Nociception Level During

Anaesthesia

Analysis and Control

Ana Castro
Departamento de Engenharia Electrotcnica e de Computadores
Faculdade de Engenharia da Universidade do Porto

A thesis submitted for the degree of

Doctor in Biomedical Engineering

November 2011

Nociception Level During Anaesthesia

Analysis and Control
by Ana Castro

Advisor:

Catarina S. Nunes (PhD)

(King's College London, United Kingdom)

Co-Advisor:

Fernando Gomes de Almeida (PhD)

(Faculdade de Engenharia da Universidade do Porto, Portugal)

Clinical Advisor:

Pedro Amorim (MD)

(Instituto de Cincias Biomdicas de Abel Salazar da Universidade do Porto,

Portugal)

Financial Support by the Fundao para a Cincia e a Tecnologia.

Under the reference SFRH/BD/35879/2007.

Reserve your right to think,

for even to think wrongly
is better than not to think at all.
Hypatia of Alexandria

I would like to dedicate this doctoral dissertation to my loving parents.

Acknowledgements

The work developed during the course of this doctoral program, and here presented,
was possible with the contribution of a wide group of people that somehow touched me
during this period, and to only some of whom it is possible to give particular mention.
Let me begin by thanking the support and scientic rigor forwarded by my advisor
Prof. Catarina S. Nunes, as well as for the refreshing look and tireless meetings with
my co-advisor Prof. Fernando Gomes de Almeida.
My special thanks to the person who introduced me into the hospital environment,
and gave me a privileged knowledge on the medical specialty of Anesthesiology, Dr.
Pedro Amorim, for his endless enthusiasm for discovery and readiness to face new challenges and projects.
I would like to acknowledge the Investigao Clnica do Servio de Anestesiologia,
of Santo Antnio Hospital, who welcomed me during this period, to the Director of
the Servio de Anestesiologia Dra. Isabel Arago Fechs, the Director of the Servio de
Bloco Operatrio Dr. Simo Barros Esteves, and to the Director of the Departamento
de Anestesiologia Cuidados Intensivos e Emergncia Dr. Antnio Marques da Silva, for
their support in the approved research studies.
I would like to thank the prompt and tireless collaboration of a large group of
professionals, who always made available to support data collection performed at the
Urology operating room: nurses, surgeons, auxiliaries, anesthesiologists, interns, there
were many who contributed for a successful outcome, and to which I leave a warm thank
you. I leave my special thanks to the anesthesiologists who accompanied me throughout
these years Dra. Eduarda Amadeu, Dra. Ftima Martins and Dra. Paula S Couto,
it was thanks to their support, patience, teaching, and personal involvement that this

vii

work has gained strength, it was with them that I have learned and gained inspiration
for Medicine. A warm thank you to the intern Diana Afonso, to which was a pleasure
to work and share experiences with.
I would like to thank all patients and volunteers who agreed to participate in the
studies, from whom I always received the maximum collaboration. I also would like to
acknowledge Neurinbloc, Helena and Filipe, for their expertise and collaboration.
I was fortunate to nd several funny and interesting friends during the course of
this work, I would like to leave a particular warm mention to my lab mates, Susana
Brs and Nadja Bressan, who attended over the years always with a friendly word and
stories of happiness. To Snia Gouveia for the wise counsels, and Ana Maria and Aura
Maia for the interesting discussions, my warm thank you.
To my friends Fernanda and Tiago, for the friendship and good years in the university, and to my friend Clia Cruz, who has accompanied me since the rst steps in
school, companion and witness of life, thank you for the certainty of mutual support
that has always united us.
A special and caring thank you to Silvio for the love, and companionship that carried
and kept me going in the nal moments.
Of course, I would not forget my doggy roommates, scar and Maggie, for the love,
friendship, and no doubt endless joy with which they wake me up, and receive me at
home, every single day.
I thank my family, in particular my parents in my heart for all the dedication and
unconditional love. My eternal gratitude and love.
Last but not least, I would like to acknowledge the nancial support provided by
the Fundao para a Cincia e a Tecnologia, Portugal (SFRH/BD/35879/2007), for the
scholarship which allowed my dedication in full time to the development of this thesis.

Porto, November 2011

Ana Castro

Abstract

Nociception Level During Anaesthesia: Analysis and

Control

This thesis aims to study and develop objective measures of the nociception/antinociception balance during anaesthesia. In spite of the increasing interest in recent years
on both pain regarding to surgery and objective assessment of the analgesia component
adequacy, these are still open issues due mainly to the fact that general anaesthesia presupposes an unconscious subject, which implies lack of collaboration from the subject.
Due to this fact, the term nociception is used instead of pain whenever regarding
to unconscious subjects.

However, diculties in assessing pain in the conscious sub-

ject also apply to the nociception/anti-nociception assessment under anaesthesia, due

to the particular characteristics of pain sensing, and the inter-patient variability regarding baseline and amplitude responses to stimulus intensity and attenuation drugs,
which makes the identication of the optimum state for the individual a dicult task.
This thesis is focused on the problem of analysis and control of nociception, in subjects
submitted to general anaesthesia.
Seeking to answer some of the questions on nociception/anti-nociception balance
assessment in anaesthesia, this thesis comprises a survey applied to clinical anaesthesiologists, a clinical study in surgical patients under general anaesthesia, a clinical study in
volunteers, and a simulation study on closed-loop control of an anaesthetic drug. This
thesis also provides a comprehensive review on the the topic of objective assessment of
nociception and analgesia balance during general anaesthesia.

ix

The rst step to a objective nociception/anti-nociception balance assessment is to

answer the question how intense is the noxious stimulus applied to the patient triggering
the observed physiological responses?. To answer this question a survey was conducted
in clinical anaesthesiologists to assess clinical perception of the intensity of noxious
stimuli related to anaesthetic and surgical procedures. A total of 57 anaesthesiologists
answered the survey, rating 35 stimuli in an ordinal scale from 0 to 10.

Later, this

information was used to construct a measure of noxious stimulus intensity, using Rasch
analysis, which was presented in the form of scale. Regarding to a surgical procedures
under general anaesthesia, this scale allows an estimative of the stimuli intensity during
the procedures, and also allows to model the input drugs' doses and stimuli intensities
impact on the physiological variables linked to noxious activation.
The rst clinical study was focused on the response to precise noxious stimuli, at
dierent analgesic drug doses under a TIVA TCI general anaesthesia of propofol (hypnotic) and remifentanil (analgesic).

Collected physiological data were pre-processed,

with emphasis on the collected waves in order to extract relevant information on noxious activation.

Such information includes heart rate, pulse plethysmography wave

amplitude, systolic blood pressure, respiration rate and EEG derived indexes.

Data

were inspected in order to identify physiological variables responding adequately to the

stimulus and to the analgesic dose attenuation. A total of 34 patients were enrolled in
the study following informed consent and institutional approval, and randomly divided
in three study groups according to the remifentanil dose (Ce=2.0, 3.0 or 4.0 ng/ml).
Based on current state of the art, two dierent methodologies were used.

The rst

was an homeostasis index, combination of steady-state individual indexes of each variable shown to contain information on noxious activation, applying wavelet analysis
adjusted to each physiological signal characteristics.

The second was a physiological

dynamic model incorporating the previously identied physiological signals linked to

noxious activation, capable of estimating patients' perceived stimulus as a combination
of the developed scale of noxious stimuli intensity and analgesic drug dose attenuation/depression.

Both methodologies provided complementary information, and were

shown to adequately respond to precise noxious stimulation and analgesic attenuation.

The second clinical study was designed to investigate the applicability of somatosensory evoked responses to assess the nociception/anti-nociception balance, for dierent
combinations of propofol and remifentanil. A total of 10 healthy volunteers were enrolled
in this study, following informed consent and institutional approval. Evoked responses

to painful electrical stimuli were measured on the somatosensorial brain area, and related to the stimulation intensity and anaesthetic drugs' attenuation.

This approach

allowed the detection of individual sensing thresholds and expected response to pattern
stimulation, which presented large interindividual variability. In this study cortical somatosensory evoked potentials were found to be related to the stimulus intensity, to
pain reported by the volunteer in a numerical rating scale (0 to 10), and to drug attenuation in a dose dependent manner, both with propofol and remifentanil. This may
congure a direct and objective method to translate the nociception/anti-nociception
balance in uncommunicative patients, allowing for the construction of normalized measures, comparable between subjects.
Finally, a simulation study on the closed-loop control of a general anaesthetic drug
was conducted.

Control in anaesthesia has been extensively investigated in the last

years, with proposed automatic administration systems for the muscle relaxant and
hypnotic drug, which correspond to general anaesthesia components that already have
validated monitors to translate the patient's state. A new approach to the control of
anaesthetic drugs' administration is here presented for the hypnotic component, taking
into account inter-patient variability described in the modelling process. The technique
employed was a sliding-mode controller, that presented robust responses to patients
diering from the nominal. The common features to the analgesia component, and the
necessary studies to further evolve in the automatic administration of general anaesthesia were discussed.
Summarizing, two approaches to the nociception/anti-nociception balance assessment were proposed in this thesis, diering from the methods presented in the literature, and shown to adequately respond to noxious stimulus and anaesthetic drugs.
Future studies should be conducted to further develop and validate the proposed indexes. Following its validation, the indexes may be employed in the closed-loop control
of the nociception/anti-nociception balance during anaesthesia, taking into account the
observed inter-patient variability.

Resumo

Anlise e Controlo do Nvel de Nocicepo Durante a

Anestesia

Esta tese tem como nalidade o estudo e desenvolvimento de medidas objetivas do

balano nocicepo/anti-nocicepo durante a anestesia. Apesar do crescente interesse,
nos ltimos anos, acerca da dor relativa cirurgia e tambm de uma avaliao objectiva
da adequao da componente analgesia, estas questes ainda permanecem em aberto
devido, principalmente, ao facto da anestesia geral pressupor um indivduo inconsciente,
o que implica a impossibilidade de colaborao por parte do paciente. Devido a este
facto, o termo nocicepo usado no lugar de dor sempre no que se refere a indivduos inconscientes. No entanto, as diculdades em avaliar dor no indivduo consciente
tambm se aplicam avaliao da nocicepo/anti-nocicepo sob anestesia, devido
s caractersticas particulares da percepo de dor, e a variabilidade inter-individual
observada nos valores basais e amplitude de respostas relativamente intensidade do
estmulo e drogas atenuantes, o que torna a identicao do estado ptimo para o indivduo, uma tarefa difcil. Esta tese foca o problema da anlise e controlo da nocicepo
em indivduos submetidos a anestesia geral.
Procurando responder algumas das questes relacionadas com a avaliao do balano nocicepo/anti-nocicepo em anestesia, esta tese compreende um questionrio
aplicado a anestesiologistas clnicos, um estudo clnico em pacientes cirrgicos sob anestesia geral, um estudo clnico em voluntrios e um estudo em simulao do controlo
em malha fechada de uma droga anestsica. Esta tese compreende ainda uma reviso

xiii

abrangente sobre a avaliao objectiva da nocicepo e analgesia durante a anestesia

geral.
O primeiro passo para uma avaliao objectiva do balano nocicepo/anti-nocicepo
responder questo quo intenso o estmulo nxico aplicado ao paciente que despoleta as reaces siolgicas observadas?. Para responder a esta questo um questionrio
foi conduzido e aplicado a anestesiologistas clnicos para avaliar a percepo da intensidade de estmulos nxicos relativos aos procedimentos da anestesia e cirurgia. No total,
57 anestesiologistas responderam ao questionrio, avaliando 35 estmulos, numa escala
ordinal de 0 a 10. Posteriormente, esta informao foi utilizada na construo de uma
medida de intensidade do estmulo nxico, aplicando anlise de Rasch, e apresentada
sob a forma de uma escala. Relativamente a um procedimento cirrgico sob anestesia
geral, esta escala fornece uma estimativa da intensidade de estmulos nxicos, e permite
tambm a modelao do impacto das entradas intensidade de estmulo e doses de drogas
atenuantes, nas variveis siolgicos relacionadas com activao nxica.
O primeiro estudo clnico foi focado na avaliao das respostas a estmulos nxicos
precisos, para diferentes doses de anestsicos durante uma anestesia geral TIVA TCI de
propofol (hipntico) e remifentanil (analgsico). Os dados siolgicos recolhidos foram
pr-processados, com nfase nas ondas recolhidas de modo a extrair informao relevante sobre activao nxica. Tal informao inclui frequncia cardaca, amplitude da
onda de pulso, presso arterial sistlica, frequncia respiratria e ndices derivados do
EEG. Os dados foram inspeccionados de modo a identicar as variveis siolgicas que
respondem adequadamente ao estmulo e atenuao pela droga analgsica. No total,
34 pacientes foram includos no estudo, aps consentimento informado e aprovao institucional, e divididos aleatoriamente em trs grupos de estudo de acordo com a dose de
remifentanil (Ce=2.0, 3.0 ou 4.0 ng/ml). Com base no estado da arte, duas metodologias
diferentes foram aplicadas. A primeira, um ndice de homeostasia, combinao de ndices individuais de estacionaridade para cada varivel que se mostrou conter informao
sobre activao nxica, aplicando anlise

wavelet

ajustada s caractersticas de cada

sinal siolgico. A segunda, um modelo siolgico dinmico incorporando as variveis

siolgicas anteriormente identicadas como relacionadas com activao nxica, capaz
de estimar a percepo do estmulo por parte do paciente, como combinao da escala
de intensidade de estmulos nxicos e a atenuao/depresso fornecida pelo analgsico.
As duas metodologias forneceram informao complementar, e demonstraram responder
adequadamente a estmulo nxicos precisos e atenuao analgsica.

O segundo estudo clnico foi desenhado para investigar a aplicabilidade das respostas somatosensoriais evocadas na avaliao do balano nocicepo/anti-nocicepo,
para diferentes combinaes de propofol e remifentanil. No total, 10 voluntrios saudveis foram includos no estudo, aps consentimento informado e aprovao institucional.

Respostas evocadas a estmulos elctricos dolorosos foram medidas na rea

somatosensorial do crtex e relacionadas intensidade de estimulao e atenuao por

drogas anestsicas.

Esta abordagem permitiu a deteco de limiares sensitivos e da

resposta esperada a uma estimulao padronizada, apresentando elevada variabilidade

inter-individual. Neste estudo, potenciais evocados somatosensitivos corticais foram relacionados com a intensidade do estmulo, com a dor reportada pelo voluntrio numa
escala numrica de dor (0 a 10), e com a atenuao fornecida pelas drogas propofol
e remifentanil.

Esta abordagem pode congurar um mtodo directo e objectivo na

traduo do balano nocicepo/anti-nocicepo em pacientes incapazes de comunicar,

permitindo a construo de medidas normalizadas, comparveis entre indivduos.
Finalmente, foi conduzido um estudo em simulao do controlo em malha fechada
de uma droga anestsica geral. Controlo na anestesia tem sido consideravelmente investigado no ltimos anos, com sistemas propostos para o controlo automtico das drogas
relaxante muscular e hipntico, que correspondem a componentes da anestesia geral
que j dispem de monitores validados para traduzir o estado do paciente. Uma nova
abordagem ao controlo de drogas anestsicas aqui apresentada para a componente da
hipnose, levando em considerao a variabilidade inter-individual descrita no processo de
modelao. A tcnica utilizada foi

slinding-mode control, que apresentou uma resposta

robusta para pacientes diferentes do nominal. As caractersticas comuns componente

analgesia da anestesia geral, e os estudos necessrios para o desenvolvimento de um
sistema de controlo automtico da anestesia geral foram discutidos.
Resumindo, duas abordagens para a avaliao do balano nocicepo/anti-nocicepo
durante a anestesia geral foram propostas nesta tese, diferindo dos mtodos propostos
na literatura, com resposta adequada introduo de estmulos nxicos e drogas anestsicas.

Estudos futuros devero ser conduzidos no sentido de desenvolver e validar

os ndices propostos. Aps validao, os ndices podero ser aplicados no controlo em

malha fechada do balano nocicepo/anti-nocicepo durante anestesia, levando em
considerao a variabilidade inter-individual observada.

Rsum

Analyse et Contrle de le Niveau de la Nociception dans

Anesthsie

Cette thse adresse l'tude et le dveloppement des mesures objectives de l'quilibre

nociception/anti-nociception pendant l'anesthsie. En dpit de l'intrt croissant sur
l'tude des deux douleurs relatives la chirurgie, et l'valuation objective de la pertinence de la composante analgsique, ces questions restent encore ouvertes en raison
principalement du fait que l'anesthsie gnrale suppose un sujet inconscient, ce qui
implique l'absence de collaboration du sujet. En consquence, le terme nociception est
utilis au lieu de douleur chaque fois qu'on vise des sujets inconscients. Toutefois, la
dicult en ce qui concerne l'valuation galement de la douleur dans le sujet conscient
s'appliquer l'valuation nociception/anti-nociception sous anesthsie, en raison des
caractristiques particulires de la dtection de la douleur, et de la variabilit interpatient concernant les rponses de base et l'amplitude de l'intensit du stimulus et de
la drogue d'attnuation, ce qui rend une tche dicile l'identication de l'tat optimal
pour l'individu. Cette thse est axe sur le problme de l'analyse et du contrle de la
nociception, chez les sujets soumis une anesthsie gnrale.
Cherchant des rponses certaines questions sur l'quilibre nociception et antinociception en anesthsie, cette thse comporte une tude applique des anesthsistes
clinique, une tude clinique chez les patients chirurgicaux sous anesthsie gnrale, une
tude clinique chez des volontaires, et une tude de simulation sur la fermeture d'une

xvii

boucle de contrle d'un drogue anesthsique. Cette thse propose galement un examen complet sur le sujet de l'valuation de l'objectif de la nociception et l'quilibre
analgsique pendant l'anesthsie gnrale.
La premire tape d'une valuation de l'quilibre nociception/anti-nociception est
de rpondre la question suivante Quelle est l'intensit de la stimulation nociceptive
applique au patient qui dclenche les rponses physiologiques observes ?. Pour rpondre cette question, une questionnnaire a t mene dans les anesthsistes cliniques
an d'valuer la perception clinique de l'intensit des stimuli nocifs lis aux procdures
d'anesthsie et de chirurgie. Un total de 57 anesthsistes ont rpondu, classiant 35
stimuli dans une chelle ordinale allant de 0 10. Ultrieurement, cette information a
t utilise pour construire une mesure de l'intensit du stimulus nociceptif, en utilisant
l'analyse de Rasch, qui a t prsent sous la forme d'chelle. En ce qui concerne une
intervention chirurgicale sous anesthsie gnrale, cette chelle permet une estimative
de l'intensit des stimuli au cours des procdures, et permet galement de modliser les
doses des drogues d'entre et les stimuli intensits impact sur les variables physiologiques
lis l'activation nocives.
La premire tude clinique a porte sur la rponse des stimuli nocifs, notamment
direntes doses de drogues anesthsiques sous anesthsie gnrale TIVA TCI du propofol (hypnotique) et rmifentanil (analgsique). Les donnes physiologiques recueillies ont
t prtraites, en mettant l'accent sur les ondes recueillies visant l'extraction des informations pertinentes sur l'activation nocives. Ces informations comprennent le rythme
cardiaque, pouls plthysmographie amplitude de l'onde, la pression artrielle systolique,
taux de respiration et EEG driv indexes. Les donnes ont t inspectes an d'identier les variables physiologiques qui permettent une rponse adquate la stimulation
et l'attnuation de la dose analgsique. Un total de 34 patients ont t inclus dans
l'tude aprs consentement clair et l'approbation institutionnelle, et rpartis au hasard en trois groupes d'tude en fonction de la dose de rmifentanil (Ce=2,0, 3,0 ou
4,0 ng/ml). Bas sur l'tat actuel de la technique, deux mthodologies direntes ont
t utilises. Le premier tait un indice homostasie, la combinaison de l'tat stable des
indices individuels de chaque variable a montre contenir des informations sur l'activation nocive, en appliquant l'analyse en ondelettes ajusts chaque caractristique du
signal physiologique. Le second tait un modle physiologique dynamique intgrant les
signaux physiologiques lie l'activation nocives prcdemment identies, capable de
reproduire les relations entre l'estimation de relance peru par le patient comme une

combinaison de l'chelle dveloppe de l'intensit des stimuli nocifs et dose de drogue

analgsique d'attnuation/dpression . Les deux mthodologies ont fourni des informations complmentaires, et ont montr tre capables de rpondre adquatement une
stimulation nociceptive prcis et d'attnuation analgsique.
La seconde tude clinique a t conue pour tudier l'applicabilit des rponses somesthsiques voques pour valuer l'quilibre nociception/anti-nociception, pour diffrentes combinaisons de propofol et de rmifentanil. Un total de 10 volontaires sains
ont t inclus dans cette tude, aprs consentement clair et l'approbation institutionnelle. Des rponses voques des stimuli lectriques douloureuses ont t mesures
sur la zone du cortex somatosensoriel, et lies l'intensit de la stimulation et l'attnuation des drogues anesthsiques. Cette approche a permis la dtection de seuils
individuels de sensibilit et rponse attendue la stimulation standardise, qui prsentait une grande variabilit interindividuelle. Dans cette tude, a t trouve une liaison
entre les potentiels voqus somesthsiques corticaux et l'intensit du stimulus de la
douleur rapporte par le bnvole dans une chelle d'valuation numrique (0 10), et
l'attnuation de drogue d'une manire dose dpendante, la fois avec du propofol et de
rmifentanil. Cela peut congurer une mthode directe et objective de traduire l'quilibre nociception/anti-nociception chez les patients peu communicatifs, permettant la
construction de mesures normalises, comparables entre les sujets.
Finalement, il a aussi t mene une tude de simulation sur le contrle en boucle ferme d'un drogue anesthsique gnral. Le contrle de l'anesthsie a t largement tudi
dans ces dernires annes, en ce qui concerne les systmes d'administration automatique
pour le relaxant musculaire et hypnotique, qui correspondent aux composantes gnrales de l'anesthsie qui ont t dj valids moniteurs qui traduisent l'tat du patient.
Une nouvelle approche pour le contrle de l'administration des drogues anesthsiques
est ici prsente pour la composante hypnotique, en tenant compte de la variabilit
inter-patient dcrite dans le processus de modlisation. La technique employe est base sur un contrleur en mode glissant, qui a prsent des rponses robustes dans le
cas des patients qui dirent de la valeur nominale. Les caractristiques communes
la composante analgsique, et les tudes ncessaires pour continuer l'volution dans
l'administration automatique de l'anesthsie gnrale ont t discuts.
En rsum, deux approches pour la valuation de l'quilibre nociception et antinociception ont t proposes dans cette thse, qui dire de la mthode prsente dans
la littrature. Ces approches ont montr une rponse adquate aux stimuli nocifs et

aux drogues anesthsiques. tudes futures doivent tre menes an de dvelopper et
de valider les indices proposs. Aprs sa validation, les indices pe uvent tre utilises
dans le contrle en boucle ferme de l'quilibre nociception/anti-nociception pendant
l'anesthsie, en tenant compte de la variabilit inter-individuelle.

Contents

List of Figures

xxvii

List of Tables

xxxix

List of Abbreviations

xlv

1 Introduction
1.1

General Anaesthesia Triad . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.1

1.2

1.3

1.5

Measuring Nociception - A Parallel Problem to the Hypnosis Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1.2

Hypnosis Monitors . . . . . . . . . . . . . . . . . . . . . . . . . .

Pain

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1

Sense Organs

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.2

The Nervous System . . . . . . . . . . . . . . . . . . . . . . . . .

12

1.2.3

Pain and Nociception

. . . . . . . . . . . . . . . . . . . . . . . .

14

1.2.4

Methods of Induction and Assessment of Pain . . . . . . . . . . .

16

Anaesthetic Drugs

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

1.3.1

Analgesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

1.3.2

Hypnotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

1.3.3

Pharmacokinetic and Dynamic Modelling and Target Controlled

Infusion

1.4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

State of the Art on the Analysis and Control of the Nociception Level

24

32

1.4.1

Objective Methods of Nociception Measurement

. . . . . . . . .

32

1.4.2

Automatic Control . . . . . . . . . . . . . . . . . . . . . . . . . .

41

Motivation, Objectives and Thesis Outline . . . . . . . . . . . . . . . . .

43

xxi

CONTENTS
2 Data Collection and Pre-Processing

47

2.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47

2.2

Clinical Protocol Design . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

2.2.1

Design Goals

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

2.2.2

Selection Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

2.2.3

Pre-Medication . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

2.2.4

Clinical Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

2.2.5

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51

2.3

Institutional Approval

2.4

Data Overview
2.4.1

2.5

2.6

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

54

Files' Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

Waves Pre-Processing

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

2.5.1

Electrocardiogram and Beat-to-Beat Interval

2.5.2

Invasive Blood Pressure

2.5.3

Photoplethysmography and Wave Amplitude

. . . . . . . . . . .

68

2.5.4

Respiration Rate . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

Summary

. . . . . . . . . . .

62

. . . . . . . . . . . . . . . . . . . . . . .

67

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 Stimulus Intensity Analysis

70

71

3.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

71

3.2

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

73

3.2.1

Survey Design and Dissemination . . . . . . . . . . . . . . . . . .

74

3.2.2

Scale Construction . . . . . . . . . . . . . . . . . . . . . . . . . .

74

3.2.3

Pharmacological Studies Comparison . . . . . . . . . . . . . . . .

77

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79

3.3.1

Clinical Perception: Monitoring for Nociception . . . . . . . . . .

81

3.3.2

Measurement Construction Process . . . . . . . . . . . . . . . . .

82

3.3.3

Pre-Intensities vs Rasch Scale . . . . . . . . . . . . . . . . . . . .

88

3.4

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

89

3.5

Summary

92

3.3

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxii

CONTENTS
4 Data Analysis - Passive Nociception Measures

95

4.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

95

4.2

Phase I Precise Stimuli Analysis

. . . . . . . . . . . . . . . . . . . . . .

96

4.2.1

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

97

4.2.2

Results

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

100

4.2.3

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

110

Phase II Maintenance Analysis - Steady-State Detection . . . . . . . . .

112

4.3.1

Wavelet Analysis and Steady-State Detection

. . . . . . . . . . .

113

4.3.2

Homeostasis Index

. . . . . . . . . . . . . . . . . . . . . . . . . .

124

4.3.3

Surface Modelling: Input/Output Steady-State

. . . . . . . . . .

127

4.3.4

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

128

Phase II Maintenance Analysis - Physiological Modelling . . . . . . . . .

130

4.4.1

Individual Modelling . . . . . . . . . . . . . . . . . . . . . . . . .

136

4.4.2

Merged Modelling

. . . . . . . . . . . . . . . . . . . . . . . . . .

140

4.4.3

Perceived Stimulus Estimator . . . . . . . . . . . . . . . . . . . .

143

4.4.4

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

144

4.3

4.4

4.5

Summary

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Evoked Potentials - Active Nociception Measures

146

149

5.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

149

5.2

Evoked Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

151

5.3

State of the Art on Evoked Potentials and Nociception . . . . . . . . . .

155

5.4

Clinical Protocol Design . . . . . . . . . . . . . . . . . . . . . . . . . . .

164

5.4.1

Design Goals

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

164

5.4.2

Selection Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . .

164

5.4.3

Clinical Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

165

5.4.4

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

166

5.4.5

Institutional Approval

. . . . . . . . . . . . . . . . . . . . . . . .

169

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

169

5.5

5.5.1

Electrical Stimulus Currents and Evoked Responses Prior to Drugs'

Administration

5.5.2

5.5.3

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

172

Numerical Rating Scale Evaluations and Remifentanil and Propofol Concentrations . . . . . . . . . . . . . . . . . . . . . . . . .

173

Normalized Somatosensory Evoked Responses Ratio

174

xxiii

. . . . . . .

CONTENTS
5.5.4

Cortical Somatosensory Evoked Responses and Numerical Rating

Scale Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.5.5

176

Cortical Somatosensory Evoked Responses and Remifentanil and

Propofol Concentrations . . . . . . . . . . . . . . . . . . . . . . .

177

5.5.6

Cervical Somatosensory Evoked Potentials . . . . . . . . . . . . .

179

5.5.7

Maximum Drug Doses Cortical Eect

183

5.5.8

Nonlinear Nature of Pain Sensation an Dynamic Modelling

. . . . . . . . . . . . . . .
. . .

183

5.6

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

185

5.7

Future Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

188

5.8

Summary

190

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6 On Nociception Control

193

6.1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

193

6.2

General Anaesthesia Triad Control

198

6.3

Hypnosis Control - A Parallel Problem to the Nociception Control

. . . . . . . . . . . . . . . . . . . . .
. . .

202

6.3.1

State Entropy Modelling . . . . . . . . . . . . . . . . . . . . . . .

203

6.3.2

Feedback Linearization . . . . . . . . . . . . . . . . . . . . . . . .

205

6.3.3

Sliding-Mode Control

. . . . . . . . . . . . . . . . . . . . . . . .

207

6.3.4

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

210

6.3.5

Results

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

211

6.3.6

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

213

6.4

Prediction and Advisory System

. . . . . . . . . . . . . . . . . . . . . .

216

6.5

Summary

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

217

7 Conclusions and Future Work

7.1

Future Research Directions

219
. . . . . . . . . . . . . . . . . . . . . . . . .

List of Publications

222

225

A Remifentanil Recommendations in Portugal

229

A.1

Remifentanil Side Eects . . . . . . . . . . . . . . . . . . . . . . . . . . .

232

A.2

Pharmacologic Properties of Remifentanil

235

. . . . . . . . . . . . . . . . .

B Informed Consent Form

239

C Remifentanil Recommended Doses

243

xxiv

CONTENTS
D Informed Consent Form - Volunteers' Study

245

E Simulations and Control

249

References

288

xxv

CONTENTS

xxvi

List of Figures

1.1

First public demonstration of diethyl ether general anaesthesia at the

Massachusetts General Hospital, on October 16, 1846. Recreation of the
event by Robert Cutler Hinckley completed in Paris in 1892. . . . . . . .

1.2

General anaesthesia components and corresponding informative physiological signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3

Guedel's stages and planes of ether anaesthesia. . . . . . . . . . . . . . .

1.4

Bispectral index monitor during data collection:

manufacturer recom-

mended target range for general anaesthesia of 40 to 60.

. . . . . . . . .

1.5

Schematic representation of the sensory process in the human body. . . .

1.6

Schematic representation of the dorsal horn of the spinal cord layers and
terminations of the three types of primary aerent neurons pathways. . .

13

1.7

Brain areas involved in somatic sensation.

14

1.8

Schematic representation of pain assessment techniques.

. . . . . . . . . . . . . . . . .
From top to

bottom: visual analogue scale, numerical rating scale (0 meaning no pain

and 10 worst possible pain), and a verbal scale.
1.9

. . . . . . . . . . . . . .

19

McGill pain questionnaire: on left the original version in English, and on

right the adapted version in Portuguese (Brazil).

. . . . . . . . . . . . .

20

1.10 Remifentanil 2D molecular schematic representation (methyl 1 - (3 methoxy - 3 - oxopropyl) - 4 - (N-phenylpropanamido) piperidine - 4
- carboxylate).

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

1.11 Propofol 2D molecular schematic representation (2,6 - bis(propan - 2 yl)phenol). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

1.12 Drugs' action path in the human body, from drug dose administration to
observed clinical eect. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxvii

24

LIST OF FIGURES
1.13 Three compartments pharmacokinetic (PK) model, followed by the pharmacodynamic (PD) model and Hill equation, translating the eect-site
concentration into the output measurable eect. Where
the drug transfer rate from compartment
is the elimination rate), and
mes; in the Hill equation
drug in the system (Ce

E0

V1 , V2

and

V3

kij ,

represents

to the compartment

j (k10

are the compartments' volu-

is the eect baseline value, when there is no

= 0), EC50

is the drug eect-site concentration

necessary to achieve half of the maximum measurable eect (E ), and

represents the steepness of the descent of the eect response to the drug.

26

1.14 Screenshot of the educational software specially developed for the simulation of target-controlled infusion anaesthesia using BIS or Entropy. It
shows the surface relational model between propofol and remifentanil's
synergistic eect on the hypnosis indexes.

. . . . . . . . . . . . . . . . .

28

1.15 Results obtained by Seitsonen and colleagues: Relative (to prestimulus

interval) group average ( standard error of mean) responses to skin incision in movers (dashed line) and non-movers (solid line) of (A) EEG
response entropy (n=10 and n=12 for movers and non-movers, respectively), (B) RRI (n=12 and n=14), (C) PPG amplitude (n=12 and n=14),
and (D) PPG notch amplitude (n=12 and n=14). Skin incision marked
with t=0.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

1.16 Bispectral Index (BIS) trend during general anaesthesia, with the induction, maintenance and recovery phases. The dashed lines represent
the BIS manufacturer recommended target range for general anaesthesia
(40-60). Data from one of the patients in this study.

. . . . . . . . . . .

36

1.17 Facial muscles involved in the innate response to noxious stimulation. . .

38

1.18 Control structure for a complete automatic administration of general

anaesthesia, considering its three components: hypnosis, analgesia and
muscle paralysis.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

1.19 Nociception/anti-nociception balance index, translating the optimum state

(around 0), the use of excessive analgesic leading to system depression
(near -10), and the use of insucient analgesic for the noxious stimulation
(near 10).

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxviii

44

LIST OF FIGURES
1.20 Representation of the relation between the stimulus and administered
drugs on a specic patient, producing an amplitude response (patient
sensitivity) in the nociception/anti-nociception balance index. . . . . . .

2.1

Representation of the dened study groups, according to the eect-site

concentration of the analgesic (remifentanil) using Minto model. . . . . .

2.2

55

T M monitor; B)
Clinical setup used in data collection: A) BIS VISTA
Orchestra

R

Base Primea syringe pumps in TCI mode (propofol and re-

mifentainl); C) General Electric Aisys

R

monitor and ventilator; D) Com-

puter used in data collection with Rugloop

2.4

49

Flow diagram of the hospital's required procedures to obtain institutional

approval for the realization of clinical studies. . . . . . . . . . . . . . . .

2.3

45

c Waves installed. . . . . . .

56

Monitored patient during a standard data collection with a BIS bilateral

sensor, on the left, and the BIS monitor with corresponding BIS and CVI
trends for the left channel, on the right.

2.5

57

Software developed to analyze and navigate o-line through the collected

data for each patient (pati , where

2.6

. . . . . . . . . . . . . . . . . .

i = 1, ..., 34).

. . . . . . . . . . . . . .

61

Original ECG signal (black line), ltered ECG signal (QRS complex
enhancing, gray line), and correspondent QRS complex detection (black
dot). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.7

65

Original RR time intervals (black line), and interpolated signal re-sampled

at 1 Hz: using hermitian cubic spline interpolation (gray line), and cubic
spline interpolation (dashed line). . . . . . . . . . . . . . . . . . . . . . .

2.8

65

Original RR sequence (black line) obtained after interpolation at 1 Hz,

correspondent ltered signal after outlier removal (gray line), and RR
sequence provided by the Aisys monitor through the heart rate sequence
(dashed line). Data from one of the patients in the study.

2.9

. . . . . . . .

66

Invasive blood pressure signal with overlapping systolic pressure peaks

detected (black dot). Data from one of the patients in the study.

. . . .

67

2.10 Photoplethysmography wave and correspondent wave amplitude denition (PPGA): A is the local minimum at the beginning of the pulse, and
B is the local maximum. PPGA is dened as the dierence between the
two.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxix

68

LIST OF FIGURES
2.11 Photoplethysmography wave and detected points for amplitude assessment (A, white dot, dened as the local minimum and B, black dot, as
the wave local maximum). Data from one of the patients in the study.

69

patients in the study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

69

2.12 CO2 wave and detected point of maximum CO2 for each respiration cycle,
corresponding to the end of an expiratory cycle. Data from one of the

3.1

Schematic representation of the patient's system and involved variables

in the triggered noxious responses. The amplitude of response to noxious
stimulation, in this simplied version, is dependent of the drugs' doses
(known) and of the stimulus intensity (estimated).

Other outer inter-

ferences may modify the exhibited relations (signal contamination and

changes in pharmacokinetics e.g.).
3.2

73

Representation of the hypnotic-analgesic drugs' interaction as an hierarchical model.

3.3

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

78

Anaesthesiologists' division considering their opinion on which is the most

painful stimulus, abdominal incision for apendicectomy (26%), or laringoscopy/intubation (74%). . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4

80

Total score for each stimulus within the three groups of stimuli studied:
anaesthetic procedure stimuli (Ani, i=1,...,11), pre-incision and incision
surgical stimuli (PreIncj, j=1,...,10), and post-incision surgical stimuli
(PostInck, k=1,...,14).

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.5

Total scores for each respondent, considering the 35 evaluated stimuli.

3.6

Anaesthesiologist's choice on the preferred range for a nociception/anti-

3.7

81

nociception balance index. . . . . . . . . . . . . . . . . . . . . . . . . . .

81

Representation of the measurement continuum:

i = 1, ..., 35,

stimulus location (i ,

in the Rasch model), and correspondent expected score in

the original rating scale.

3.8

. . . . . . . . . . . . . . . . . . . . . . . . . . .

84

Representation of the probability of response of each rating score (0 to

10), depending on the location in the pain intensity continuum (logit).

3.9

80

85

. . . . . . . . .

86

Relation between score rates (0 to 10) and pain intensity measures (logit),
with overlapping 0.5 points of expected score (squares).

3.10 Representation of raters' location (n ,

n = 1, ..., 57,

in the Rasch model)

distribution on the measurement continuum (logit). . . . . . . . . . . . .

xxx

86

LIST OF FIGURES
3.11 Application of the developed stimulus intensity scale: schematic representation of an urological procedure under general anaesthesia with identied stimuli intensities of the anaesthetic and surgical actions.

. . . . .

87

3.12 Comparison between pre-intensities estimated in previous pharmacological studies, and the Rasch scale obtained using anaesthesiologists noxious
stimuli perception.

4.1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

88

Schematic representation of proposed methods to assess amplitude responses prior and post-stimulation: a) and b) average versus maximum
response; c) and d) average values prior and post stimulation. . . . . . .

4.2

97

Schematic representation of the laringoscopy and intubation moments,

registered in each data set, and the corresponding time lag - stimulus
duration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.3

Steady-state detection in the input drugs' eect-site concentration using

the arithmetic rule. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.4

99

114

Diagram presenting the algorithm steps: the patient is the system, with
propofol and remifentanil drugs' eect-site concentrations (Ce) as inputs,
and output physiological variables entering the wavelet based algorithm
to determine the combined steady-state index as summarized.

Dashed

line rectangle presents the initially used input detection rule, replaced by
the wavelet based algorithm. . . . . . . . . . . . . . . . . . . . . . . . . .
4.5

Detailed representation of steady-state detection of the systolic blood

pressure trend, in one of the collected data sets. . . . . . . . . . . . . . .

4.6

118

121

Representation of the input and output steady-state indexes. The bottom

trend may be interpreted as an homeostasis index of the subject considering BIS, heart rate, systolic blood pressure, pulse wave amplitude and
respiration rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.7

122

Schematic representation of the signal processing steps to obtain the individual and combined steady-state indexes, from data collection, to signal
pre-processing unit, and homeostasis index presentation (individual and
combined), as implemented in the developed tool for on-line homeostasis
assessment.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxxi

125

LIST OF FIGURES
4.8

Designed software for on-line assessment of the developed steady-state indexes: individually for each input (propofol and remifentanil Ce), output
(bispectral index, electromyography, heart rate, blood pressure, photoplethysmography wave amplitude, respiration rate), and combinations of
input and output signals (In and

Out ).

Data presented in this example

is from one of the patients of the study, simulating an on-line assessment. 126
4.9

Combined periods of steady-state (SS) input/output in all data sets:

relation between drugs' eect-site concentrations (Ce) and average heart
rate (HR). The stimuli intensities are represented in dierent colors, with
white representing periods without stimulation (0 in the Rasch scale), and
increasing gray intensity with increasing stimulus intensity.

. . . . . . .

128

4.10 Representation of the attenuation and depression eects of the analgesic drug on a stimulus of known intensity, with the introduction of an
additional depression curve.

. . . . . . . . . . . . . . . . . . . . . . . . .

130

4.11 From top to bottom: stimulus intensity trend, using author's annotations and intensities estimated by the Rasch analysis;

preopioid intensity

estimated based on the Rasch scale extrapolation; estimated remifentanil eect-site concentration (Ce in ng/ml);

postopioidintensity estimates

preopioid intensity

and remifentanil attenuation;

based on extrapolated

total surgical stress. Data from one of the patients in the study. . . . . .

134

4.12 Structure representing the impact of anaesthetic drugs and noxious stimulus on the physiological measurable eects. . . . . . . . . . . . . . . .

135

4.13 Structure of the physiological model, taking into account physiological

variables linked to noxious activation dynamical relations, hypnotic drug
interference, and output estimation of the perceived stimulus.

. . . . . .

135

4.14 Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil (output), and bispectral index, frontal electromyography, heart rate, systolic blood pressure,
pulse plethysmography wave amplitude, and propofol eect-site concentration as inputs.

(a) Individually adjusted state-space model of or-

der 1 to TSS. (b) Individually adjusted state-space model of order 1

to

postopioid intensity .

. . . . . . . . . . . . . . . . . . . . . . . . . . .

xxxii

138

LIST OF FIGURES
4.15 Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil, using total surgical stimulus
(TSS) as output signal, and bispectral index, frontal electromyography,
heart rate, systolic blood pressure, pulse plethysmography wave amplitude, and propofol eect-site concentration as inputs.

(a) Individually

adjusted state-space model of order 1, in the data set with higher correlation to the output signal TSS (correlation of 0.97). (b) Individually
adjusted state-space model of order 1, to the data set with lowest correlation (correlation of 0.51). . . . . . . . . . . . . . . . . . . . . . . . . . .

139

4.16 Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil, using total surgical stimulus
(TSS) as output signal, and bispectral index, frontal electromyography,
heart rate, systolic blood pressure, pulse plethysmography wave amplitude, and propofol eect-site concentration as inputs. (a) Data set with
higher correlation to the output signal TSS, considering the merged data
model. (b) Data set with lowest correlation, considering the merged data
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

142

4.17 Designed software for on-line assessment of the developed steady-state

indexes, individually or for combinations of input and output signals
(In and

Out ), and the estimator of the perceived stimulus, based on the

physiological model for the population. Data presented in this example

is from one of the patients of the study, for a simulation of an on-line
assessment.

5.1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Schematic representation of the stimulation pathway from stimulus site

to higher processing.

5.2

144

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

151

Somatosensory evoked potential representation with latency, amplitude

and respective vertex potential marks (positive and negative): (a) cortical
site; (b) cervical site. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.3

Equipment to obtain somatosensory evoked potentials, and process data

with time-locked stimulus. . . . . . . . . . . . . . . . . . . . . . . . . . .

5.4

153

154

Summary of a review of studies evaluating the impact of anesthetic drugs

on evoked potentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxxiii

158

LIST OF FIGURES
5.5

Brain schematic representation with sensory, motor and auditory areas

outlined. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.6

International 10-20 system for electroencephalogram collection. The letters used stand for:

F-frontal, T-temporal, C-central, P-parietal, O-

occipital, and Z referring to the electrodes positioned in the midline.

5.7

161

Clinical setup used in the volunteer's study data collection:

. .

166

A) SEPs

T M CR (Nicolet-Vyasis); B)

monitor and stimulation device Endeavor

Electrical stimulus electrodes placed on volunteer median nerve; C) Aisys

monitor and ventilator; D) BIS sensor and SEPs needles montage.
5.8

169

Collected cortical somatosensory evoked potential waves, with dened

peaks by the neurophysiology technicians.

5.9

. . .

. . . . . . . . . . . . . . . . .

171

BIS trends for all volunteers during the experiment (a), and one case in
detail (b), centered according to propofol start (time=0). The study was
interrupted as soon as the volunteer lost response to verbal command,
maintaining response to mechanical stimulation, or by clinical advice. . .

172

5.10 Sensitive (ST), motor (MT), painful (PT) and 1.3 painful (1.3PT) threshold currents of stimulation for each volunteer. . . . . . . . . . . . . .

173

5.11 Relation between the painful stimulus (a) and 1.3 painful stimulus (b)
intensity and the SEPs' latency and amplitude responses prior drug administration.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

174

5.12 Numerical rating scale pain evaluations (0-10) for each volunteer both
for the painful threshold (PT) and 1.3PT stimuli. . . . . . . . . . . . .

175

5.13 Numerical rating scale evaluations (NRS) versus original (a) and normalized (b) SEP amplitude. . . . . . . . . . . . . . . . . . . . . . . . . . . .

178

5.14 Numerical rating scale evaluations (NRS) versus: (a) SEP amplitude/SEP
latency; (b) normalized SEP amplitude/normalized SEP latency.

. . . .

179

5.15 Anaesthetic drugs impact on somatosensory evoked potentials normalized

ratio SEP Normalized Amplitude/SEP Normalized Latency versus: (a)
Remifentanil only; (b) Propofol and remifentanil eect-site concentration
of 1.0 ng/ml; (c) Remifentanil and propofol eect-site concentration of
1.2 g/ml. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

180

5.16 Original (a) and normalized (b) ratio between cervical SEP amplitude
and latency, versus evaluations in the numerical rating scale. . . . . . . .

xxxiv

181

LIST OF FIGURES
5.17 Normalized cervical SEP ratio versus remifentanil eect-site concentrations for each volunteer.

. . . . . . . . . . . . . . . . . . . . . . . . . . .

182

5.18 Boxplot of the normalized SEP amplitude/normalized SEP latency ratio (RN orm ) for the maximum drug doses combination achieved in each
volunteer in each administration scheme: 1 - No drugs; 2 - Remifentanil
only; 3 - Propofol increasing steps and constant remifentanil (1.0 ng/ml);
4 - Remifentanil increasing steps and constant propofol (1.2 g/ml). . . .

182

5.19 Normalized SEP cortical ratio in the original linear scale (a) and logarithmic scale (b), versus evaluations in the numerical rating scale. . . . .

184

5.20 Collected data points from all volunteers considering propofol and remifentanil eect-site concentrations (Ce) as inputs, and the normalized
SEPs ratio (RN orm ) as output measurable eect.

. . . . . . . . . . . . .

185

5.21 Representation of a possible patch to place on the arm for example, to

administer the somatosensory stimulus, with multiple sources of contact
heat or lasers, that would be activated randomly, and irregularly in time. 189

6.1

Schematic representation of the analogy between the automatic pilot systems in the airplane, and the anaesthesiologists' action during general
anaesthesia: induction, maintenance, noxious stimulation regulation and
recovery. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6.2

194

Schematic representation of the general anaesthesia triad, with corresponding anaesthetic drugs, and their direct or indirect interference on
each other: line arrows represent the synergy relation between the hypnotic and the analgesic drugs, dashed arrows represent the impact that
the paralysis state of the patient has on the monitoring devices used to
assess the hypnosis and analgesia components, and nally dash-dot arrow
represents the impact that the hypnotic drug may have on the variables
related to noxious activation used to monitor the analgesia component. .

6.3

199

Schematic representation of an automatic control system for the administration of general anaesthetics, using a multiple-input and multipleoutput (MIMO) approach for the general anaesthesia triad.

6.4

. . . . . . .

System general structure and feedback for the sliding-mode controller.

xxxv

201
210

LIST OF FIGURES
6.5

Control activity for the nominal patient using the sign function: on top
SE and SE reference, below the error sequence, and on bottom the infusion rate (ml/h).

6.6

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

211

Control activity for the nominal patient using the saturation function:
on top SE and SE reference, below the error, and on the bottom the
infusion rate (ml/h).

6.7

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

212

Control activity for the minimum patient using the saturation function:
on top SE and SE reference, below the error, and on the bottom the
infusion rate (ml/h).

6.8

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

212

Control activity for the maximum patient using the saturation function:
on top SE and SE reference, below the error, and on the bottom the
infusion rate (ml/h).

6.9

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

213

Representation of the patient modelled system, with possible error inputs, controller, and state observer, for an on-line implementation of the
automatic control of the hypnotic.

. . . . . . . . . . . . . . . . . . . . .

214

6.10 Schematic representation of the necessary steps to model the relation

between the administered analgesic drug dose and the measurable eect,
when a tool for this eect is fully validated in the translation of the
nociception/anti-nociception balance. . . . . . . . . . . . . . . . . . . . .

E.1

Simulink block diagram for the three compartments pharmacokinetic and

dynamic model. The
ments,

massi

kij

constants are transfer rates between compart-

is the drug mass in each compartment and

concentration estimated (i, j

E.2

= 1, ..., 3).

ce the eect-site

. . . . . . . . . . . . . . . . . . .

251

Step response simulations for the Minto model, considering dierent combinations of age, weight, height and gender.

E.4

250

Step response simulations for the Schnider model, considering dierent

combinations of age, weight, height and gender. . . . . . . . . . . . . . .

E.3

215

. . . . . . . . . . . . . . . .

252

R

, with the system re-

Block diagram implemented in Simulink Matlab

presentation and feedback for the sliding-mode controller, used in the

simulations for the nominal, minimum and maximum patients.

xxxvi

. . . . .

253

LIST OF FIGURES
E.5

Representation of the diagram block for a possible on-line implementation, with the input incoming from the monitored values of State Entropy
(SE) in the patient, and with the output infusion rate (ml/h) directed to
the syringe pump actuator.

. . . . . . . . . . . . . . . . . . . . . . . . .

xxxvii

254

LIST OF FIGURES

xxxviii

List of Tables

1.1

The Modied Observer Assessment of Awareness and Sedation Score

(OAAS/S).

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Sensory modalities and corresponding receptors and sense organs. . . . .

11

1.3

Types of nerve bers that carry aerent impulses to the central nervous
system. A and B bers are myelinated, C bers unmyelinated. . . . . . .

1.4

12

Stimulus' modalities for induction of experimental cutaneous pain in humans.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.5

Stimulus' modalities for induction of muscular pain in humans.

1.6

Stimulus' modalities for induction of visceral pain in humans.

1.7

Estimated relations for the parameters of the three compartments Minto

17

. . . . .

18

. . . . . .

18

pharmacokinetic and dynamic model, and corresponding percent coecient of variation (% CV). The parameters incorporate patient's age and
lean body mass (LBM).
1.8

. . . . . . . . . . . . . . . . . . . . . . . . . . .

30

Estimated relations for the Schnider three compartments pharmacokinetic model parameters. It incorporates patient's age, weight, height and
lean body mass (LBM).

1.9

. . . . . . . . . . . . . . . . . . . . . . . . . . .

Average and standard-deviation of the estimated

(1=1,...,11) parame-

ters in the Schnider pharmacokinetic and dynamic model.

2.1

31

. . . . . . . .

31

Study interruption rules: for each case the anaesthesiologist should redene drugs' theoretical eect-site concentration targets, according to
patient's needs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

51

Patients' demographics according to the division in the three study groups:

no statistical dierence between groups, with

P < 0.05 (data as meanstandard-

deviation). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xxxix

56

LIST OF TABLES
2.3

Collected variables (47 in total) at a sampling rate of 1 Hz, exported to

*.mat les for each data set. . . . . . . . . . . . . . . . . . . . . . . . . .

58

2.4

Collected waves, number of channels, sampling rate, and data units.

59

2.5

List of noxious stimuli annotated by the author during the procedure,

and corresponding code number attributed to the stimulus variable

3.1

. .

scale.

60

List of stimuli analyzed in the study, divided according to three groups:

stimuli derived from the anaesthesia procedure; stimuli derived from the
surgical procedure, pre-incision and incision; and stimuli derived from
the surgical procedure post-incision.

3.2

. . . . . . . . . . . . . . . . . . . .

75

Total and average scores obtained for each stimulus, divided according
to the considered groups, stimuli measures (i ,

i = 1, ..., 35,

in the Rasch

model, with greater logit values indicating higher pain perception), standard errors (S.E.), expected raw score, Rasch measure parameterized in
a scale from 0 to 10, and stimulus identication, in descending order of
stimulus intensity perception (reliability of 0.98).
3.3

. . . . . . . . . . . . .

Estimated pre-intensities in previous pharmacological studies, and Rasch

measures obtained using anaesthesiologists noxious stimuli perception.

4.1

83

88

. . . . . . . . . . . . . .

100

Patients' demographic data (original and phase I nal sample): global

and for each considered study group of remifentanil target eect-site concentration (data as meanstandard-deviation).

4.2

Patients' distribution on the three possible incision sites, for each group
of study, in the nal sample (no statistical dierence between groups).

4.3

101

P < 0.05).

102

Median values for each cardiovascular variable analyzed in the study, for
each stimulus (average pre and post stimulation), divided according to
the drug dose group: remifentanil eect-site concentration (RemiCe) of
2.0, 3.0, or 4.0 ng/ml (* stands for signicant dierences with

4.4

Median values of BIS related variables analyzed in the study, for each
stimulus (average pre and post stimulation), divided according to the
drug dose group: remifentanil eect-site concentration (RemiCe) of 2.0,
3.0, or 4.0 ng/ml (* stands for signicant dierences with

4.5

P < 0.05).

. .

103

Median amplitude response of cardiovascular variables for each stimulus

and remifentanil eect-site concentration (RemiCe) target considered in
the study.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xl

105

LIST OF TABLES
4.6

Median amplitude response of BIS related variables for each stimulus and
remifentanil eect-site concentration (RemiCe) target considered in the
study.

4.7

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

106

Median normalized amplitude response of cardiovascular variables for

each stimulus and remifentanil eect-site concentration (RemiCe) target
considered in the study.

4.8

. . . . . . . . . . . . . . . . . . . . . . . . . . .

107

Median normalized amplitude response of BIS related variables for each

stimulus and remifentanil eect-site concentration (RemiCe) target considered in the study.

4.9

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

108

Median values of available CVI trends, for each stimulus (average pre and
post stimulation), divided according to the drug dose group: remifentanil
eect-site concentration (RemiCe) of 2.0, 3.0, or 4.0 ng/ml. . . . . . . . .

109

4.10 Finite impulse response of the lters H, G, K and L that correspond to

the quadratic spline wavelet. . . . . . . . . . . . . . . . . . . . . . . . . .
4.11 Normalization coecients

j = 1, 0.

for the quadratic spline wavelet. For

119

j > 5,

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

119

4.12 Estimated thresholds based on historic data for each signal analyzed, in
a training set of ve patients.

. . . . . . . . . . . . . . . . . . . . . . . .

120

4.13 Percentage of total time (total of 31 patients), that each signal analyzed
in this study phase was found to be in steady-state (SS) conditions (data
as meanstandard-deviation). . . . . . . . . . . . . . . . . . . . . . . . .

121

4.14 Median values of the proposed homeostasis index (HI): average value in
the periods baseline, prior and post-stimulation, for each remifentanil
eect-site concentration (RemiCe) group.

. . . . . . . . . . . . . . . . .

123

4.15 Normalized amplitude responses of the proposed homeostasis index (HI)

to the precise stimuli considered in study phase I, for each remifentanil
eect-site concentration (RemiCe) group.
4.16 Variable

scale

. . . . . . . . . . . . . . . . .

identies each stimulus with a code number.

124

Stimulus

intensities estimated and translated in the Rasch scale, as well as extrapolated values for the pre-intensities, considering the linear relation
shown in Figure 3.12. Table entries presented in bold correspond to the
stimuli that have pre-intensities estimations; original values presented
between parenthesis.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xli

132

LIST OF TABLES
4.17 Order 1 state-space models adjusted to the individual data, t statistics,
and correlation between output and modelled output. Model 1 corresponds to the output TSS, and Model 2 to the output

postopioidintensity .

Three input sets were considered using bispectral index (BIS), frontal electromyography (EMG), heart rate (HR), systolic blood pressure
(SBP), pulse photoplethysmography wave amplitude (PPGA), and propofol eect-site concentration (PropCe). Data presented as meanstandarddeviation.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

137

4.18 Order 1 state-space models adjusted to the merged data sets, t statistics, and correlation between output and modelled output, considering
baseline removed data, and baseline normalized data. Three input sets
were considered using bispectral index (BIS), frontal electromyography
(EMG), heart rate (HR), systolic blood pressure (SBP), pulse photoplethysmography wave amplitude (PPGA), and propofol eect-site concentration (PropCe). Data presented as meanstandard-deviation.

. . .

141

4.19 Average, minimum and maximum values of the merged model output perceived stimulus during the maintenance phase, considering all data sets
and input set of BIS, EMG, HR, SBP, PPGA and PropCe: population
trends. Data presented as meanstandard-deviation. . . . . . . . . . . .

143

4.20 Parameters of the state-space models using the merged data, considering
the three input sets.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.1

Volunteers' demographic data (data as meanstandard-deviation).

5.2

Propofol and remifentanil eect-site concentration targets and interrup-

. . .

143

170

tion rules applied in the study: the study began with remifentanil administration only, in increasing steps as described in the text, until one of
the interruption rules occurred; after interruption the drug doses were reduced to the rst step considered in the next phase, and drug doses were
increased again according to the scheme. The study had three sequent
schemes, rst remifentanil only, followed by constant remifentanil and
propofol increasing steps, and nally constant propofol and remifentanil
increasing steps. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xlii

170

LIST OF TABLES
5.3

Numerical rating scale (NRS) evaluations in each volunteer for the baseline and maximum drug doses combination achieved in each administration scheme (NE - no evaluation). . . . . . . . . . . . . . . . . . . . . . .

5.4

175

Spearman correlation coecient (rho) for each volunteer between the

normalized cortical somatosensory evoked potentials ratio (RN orm ), numerical rating scale evaluations, and remifentanil eect-site concentration
in the rst administration scheme, where remifentanil is administered in
ascending steps, without propofol (N is the number of points used to determine the correlation coecient, and * stands for signicant correlation
with P<0.05). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.5

176

Spearman correlation coecient () for the population data between the

cortical somatosensory evoked potentials normalized ratio (RN orm ) and
the numerical rating scale evaluations, and drug's eect-site concentration in each administration scheme (where N is the number of points used
to determine the correlation coecient).

. . . . . . . . . . . . . . . . . .

177

6.1

Estimated parameters for the Hill equation using State Entropy monitoring.205

A.1

Recommended doses of remifentanil in co-administration with other anaesthetic agents for adult patients. . . . . . . . . . . . . . . . . . . . . . .

230

A.2

Recommended infusion rates of remifentanil according to body weight. .

231

A.3

Infusion velocity of remifentanil and equivalent plasmatic concentrations

by approximations of several stabilized manually perfusion of remifentanil
when using target controlled infusion.

E.1

. . . . . . . . . . . . . . . . . . .

232

Considered variation intervals in the simulations, regarding the reported

parameters and dispersion measures of the pharmacokinetic and dynamic
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xliii

255

LIST OF TABLES

xliv

List of Abbreviations

AAI

Auditory Evoked Potential Index

ACC

Anterior Cingulated Cortex

AEP

Auditory Evoked Potential

ANSSI

Autonomic Nervous System State Index

ASA

American Society of Anaesthesiologists Physical Status Index

BIS

Bispectral Index

BP

Blood Pressure

Ce

Drug Eect-Site Concentration

CHEPS

Contact Heat Evoked Potentials

CO2

Carbon Dioxide

Cp

Drug Plasma Concentration

CSSA

Clinical Signs-Stimulus-Antinociception

CVI

Composite Variability Index

DLPFC

Dorsolateral Prefrontal Cortex

DNIC

Diuse Noxious Inhibitory Control

ECG

Electrocardiogram

ECoG

Electrocorticogram

EEG

Electroencephalogram

xlv

List of Abbreviations
EKF

Extended Kalman Filter

EMG

Electromyography

fMRI

Functional Magnetic Resonance Imaging

HBI

Heart Beat Interval

HR

Heart Rate

HRbaseline

Baseline Heart Rate

HRV

Heart Rate Variability

IBP

Invasive Blood Pressure

IIR

Innite Impulse Response Filter

LOC

Loss of Consciousness

MAC

Minimum Alveolar Concentration

MIMO

Multiple-Input and Multiple-Output

MIR

Minimum Infusion Rate

Noc/ANoc

Nociception/Anti-Nociception

NRS

Numerical Rating Scale

NSRI

Noxious Stimulation Response Index

PET

Positron Emission Tomography

PKPD

Pharmacokinetic and Dynamic

PPC

Posterior Parietal Cortex

PPG

Pulse Plethysmography

PPGA

Pulse Plethysmography Wave Amplitude

RE

Response Entropy

RespR

Respiration Rate

xlvi

List of Abbreviations
RN

Response Index of Nociception

ROC

Recovery of Consciousness

RR

Time interval between QRS complexes

SBP

Systolic Blood Pressure

SBPbaseline

Baseline Systolic Blood Pressure

SD

Standard-Deviation

SE

State Entropy

SEP

Somatosensory Evoked Potential

SI

Primary Somatosensory Cortex

SII

Secondary Somatosensory Cortex

SISO

Single-Input and Single-Output

SPI

Surgical Pleth Index (former SSI)

SS

Steady-State

SSI

Surgical Stress Index (see SPI)

TCI

Target Controlled Infusion

TSS

Total Surgical Stress

TTPE

Time to Peak Eect

VAS

Visual Analogue Scale

VEP

Visual Evoked Potential

WT

Wavelet Transform

xlvii

List of Abbreviations

xlviii

Chapter 1
Introduction

The term anaesthesia derives from the Greek word 'anaisthesia',

meaning lack of sensation, or insensibility.

Anaesthesiology is a medical specialty that has greatly evolved in the last 50 to 60 years,
from the simple idea of inducing pain relief and unconsciousness for a clinical procedure
to occur, to a complex set of concepts and interactions, with action in a wide range of
clinical procedures [1].
Nowadays, anaesthesia involves sophisticated technology, administration of multiple
drugs, trained professionals and demanding regulations. From being a very unsafe and
simple procedure, requiring almost no specialized skills, equipment or monitoring (just
a few drops of ether poured over the face were enough to provide anaesthetic state), to
a complex and high technology procedure.

In fact, Anaesthesiology is now a leading

specialty in the use of technology and patient safety.

The care and medical standards have taken anaesthesia to a level of demanding and
accuracy that challenges the clinician to an active interaction between patient observation, monitoring and drugs' administration. Technological advances and the development of new drugs have changed the way anaesthesiologists behave in the operating
room, improving patient care quality.
From an historical perspective, before anaesthesia was developed and introduced
in clinical practice, surgical procedures occurred while the patient remained conscious

1. INTRODUCTION
and sensible to painful stimulation. Besides being a traumatic and terrifying experience,
these interventions had a high risk of complications, during and following the procedure.
Aware of these facts, the concept of producing insensibility was quite appealing for
both patients and clinicians, improving patients' wellbeing, and also allowing surgical
technical developments with superior quality and success.

Figure 1.1: First public demonstration of diethyl ether general anaesthesia at the Massachusetts General Hospital, on October 16, 1846. Recreation of the event by Robert
Cutler Hinckley completed in Paris in 1892 [2].

The rst surgical procedure to occur under anaesthesia administration (Oct.

16

1846) was wrapped in controversy, due to the use of a secret compound which was in fact
pure unadulterated sulfuric ether. Although ether use was registered two years before,
for short dentistry procedures, however only in the referred public demonstration, a
longer surgical intervention took place at the Massachusetts General Hospital (see Figure
1.1) [2, 3].
Later, in 1847, the use of chloroform was introduced and the concept of general
anaesthesia began to gain form. Although this technique showed great advantages compared to previous approaches, there were (and still are) associated risks that soon proved
to be the worst possible. In January 1848, soon after the introduction of chloroform for
anaesthetic use, the rst of many fatalities occurred: a 15-year-old girl died while receiving a chloroform anaesthesia [4], showing that it required careful titration, beginning
the long road of drug development, patient monitoring and eects' balance, which the
anaesthesiologist still needs to go through.

1.1 General Anaesthesia Triad

1.1 General Anaesthesia Triad

Every cause produces more than one eect.
Herbert Spencer

Figure 1.2: General anaesthesia components and corresponding informative physiological signals.

In the specic literature general anaesthesia is commonly described as being composed

by three main components: hypnosis, analgesia and paralysis (see Figure 1.2). For each
of these components, the anaesthesiologist administers specic drugs, aiming at an optimum combined state, conditioned by patient's characteristics, clinical background and
individual response to the treatment. A major challenge anaesthesiologists face in their
clinical practice resides in drug combination, titration, and anticipation of unwanted
events. To respond to this challenge, the clinician uses all information available (patient data, clinical history, monitored physiological responses to treatment) to provide
and maintain homeostasis in the patient, i.e.

to provide and maintain a stable and

comfortable condition regarding the patient as a whole [5].

To evaluate patients' state regarding each component of general anaesthesia, the
clinician stays alert for any indication of inadequate anaesthesia in dierent biological
signals and derived measures. For example,

Electroencephalogram

(EEG) based moni-

tors give information on the patient's state regarding the hypnosis component. Usually

1. INTRODUCTION
these monitors translate the information contained in the EEG patterns into simple and
easy to use indexes, varying from 0 (isoelectric EEG) to 100 (fully awake patient) [6].
For the paralysis component, neuromuscular monitors allow to assess the state of the
patient regarding muscle relaxation, through the evaluation of the response to dierent
electrical stimulation patterns and currents, using movement sensors (accerelerometry)
or

Electromyography

(EMG) analysis, to evaluate the amplitude response and conse-

quently the paralysis degree of the patient.

Regarding the analgesia, there is not yet a widely accepted monitor capable of objectively translating this component. Currently, anaesthesiologists rely on physiological
signals related to the autonomous nervous system to indirectly assess adequacy of analgesia (see Figure 1.2). Analgesia and pain are dicult to assess due to its subjective
nature. In conscious subjects numerical scales are used to assess pain, however in unconscious subjects these scales become useless. These factors account for absence of a
monitor widely accepted or used to objectively assess the analgesic component of anaesthesia. In the last few years, several researchers have applied their eorts in the search
for a tool able to translate the

Nociception/Anti-Nociception

(Noc/ANoc) balance in

unresponsive or uncommunicative patients. Such a tool would be of great importance

to control drug administration and assure homeostasis throughout surgical procedures, preventing adverse eects of overdosing/underdosing potent drugs, with its adverse
physiological responses.

1.1.1 Measuring Nociception - A Parallel Problem to the Hypnosis

Measurement
To assess the Noc/ANoc balance it is necessary to identify the steps to be taken before
a fully accepted and eective monitor to measure the patient's state regarding this
anaesthesia component becomes available. Because this road has already been walked
for the monitors developed in the assessment of the other anaesthesia components, it is
useful to analyze the steps taken to produce hypnosis indexes, as a parallel problem to
the analgesia.
Most of the hypnosis indexes available in the market are calculated using the EEG
in the translation of drugs' eect on the human brain (frontal cortex EEG), and consequently the hypnosis state of a patient. As aforementioned, EEG analysis has been
explored for years in the detection of normal and abnormal patterns considering all
consciousness states, both in animal and human subjects [6, 7].

Due to the chaotic

1.1 General Anaesthesia Triad

characteristics of the signal, few people are capable of observing the raw EEG and discern about the subjects' state. It becomes even worse in an operating room, where the
clinician has to deal with information incoming from a variety of sources. Due to the
need of rapid EEG interpretation and awareness prevention during surgical procedures,
both scientists and industry soon realized the urging need for a tool to translate the
information contained in the EEG, into a simple and userfriendly index [7].
Unconsciousness monitors developed to the moment apply mathematical techniques
over the EEG to calculate indexes, usually varying from 0 to 100, translating drugs'
eect on the brain [8].

Figure 1.3: Guedel's stages and planes of ether anaesthesia [9], withdrawn from [6].

Guedel dened the anaesthesia stages and planes presented in Figure 1.3 for ether
anaesthesia [9]; the idea was to dene clinical signs characteristic of dierent anaesthesia
levels, that would place the patient in consecutive clinical states.

However, Guedel's

signs cannot be fully applied due to the dierent action mechanisms in the brain of the
currently used anaesthetic drugs. This is worsened by the use of other agents that may
mask the physiological responses. Jameson and Sloan [6] reviewed the use of the EEG
to monitor patients hypnosis state during general anaesthesia from the rst denition of
the anaesthesia planes by Guedel, to the proliferation of EEG monitors in the operating
room.
In addition to the use of clinical signs to assess the patients' state, several studies
aimed at dening average population drug doses to achieve a certain clinical eect. The
term

Minimum Alveolar Concentration

(MAC) has been dened for inhalation agents

as the minimum alveolar concentration to produce immobility in 50% of the patients in

1. INTRODUCTION
reaction to surgical incision [6]. The anaesthetic eect that causes immobility occurs in
the spinal cord, not guaranteeing adequate cortical eect, and with the use of muscle
relaxants the use of movement as a sign of adequate anaesthesia has decreased, however
the MAC that causes 50% of patients to be unconsciousness is lower than the MAC
for immobility. Other anaesthetic agents (intravenous agents) may not have the same
action, and the use of MAC initially developed in volatile anaesthesia, was replaced by
the

Minimum Infusion Rate (MIR). Althoug the MAC and MIR are usefull guides, they

have limited application for the challenges anesthesia encounters nowadays.

Table 1.1:

The Modied Observer Assessment of Awareness and Sedation Score

(OAAS/S), reproduced from [10].

Expected Response

Score

Responds readily to name spoken in normal tone

(Alert) 5

Responds only after name is called loudly and/or repeatedly

Lethargic response to name spoken in normal tone

Responds only after mild prodding or shaking

Responds only after painful trapezius squeeze

Does not respond to painful trapezius squeeze

(Unconscious) 0

A dierent score scheme used to evaluate patients' sedation state is the

Assessment of Awareness and Sedation Score

Observer

(OAAS/S, [10]) presented in Table 1.1.

This score scheme was developed and used to relate the EEG patterns with dierent
sedation levels. However, its use is limited to light sedation levels, since it depends on
the subject's collaboration to verbal and mechanical stimulation. The core idea of this
approach is to determine levels of sedation, to identify and classify EEG patterns on
each level.
Although clinical assessment of patient's state, and denition of optimum drug doses, are essential to deliver anaesthesia, the existence of objective and reproducible
methods to assess the hypnosis state of a patient is quite appealing. Thus several tools
have been developed based both on EEG and derived signals, such as the

Evoked Potentials

Auditory

(AEP). These tools use the information of the expected patterns in

each sedation/anesthesia state, to construct valid indexes, that are correlated to the
clinical state of the patient (clinical endpoints). These indexes improved the way how
anaesthesia is delivered allowing more secure and tailored drug administration, aiding

1.1 General Anaesthesia Triad

in the prevention of drug overdosage [11], awareness and recall [12, 13] during general
anaesthesia.

1.1.2 Hypnosis Monitors

Nowadays there are several monitors proposed and commercially available to assess the
hypnosis component of anaesthesia, using as referred before, EEG and derived signals
to translate the information contained into easy to use numerical indexes. In this thesis
hypnosis monitors have been used in the assessment not only of the hypnosis sate, but
also of the Noc/ANoc balance. A brief description of these monitors is presented in this
section, and their use is discussed in more detail in section 1.4.
Two of the most widely used hypnosis monitors are the
Covidien) and the Entropy

Bispectral Index

TM,

(BIS

T M (General Electric) monitors. Both monitors use signal

processing techniques over the frontal EEG, and transform the information contained in
the original chaotic signal, into easy to understand indexes. Although both indexes have
the same manufacturer recommended target range for general anaesthesia, they dier in
the signal processing techniques employed in the index calculation. Moreover, although
both monitors try to prevent the impact of outliers and outer interference, they are subjected to factors that inuence the extracted indexes, such as EMG, electromechanical
artifacts (e.g. electrocautery or forced-air warmers), abnormal EEG states, or certain
anesthetic agents' action [14].

1.1.2.1 BIS Monitor

The BIS monitor was released by Aspect Medical Systems (now part of Covidien) in
1994. The BIS index is based on the bispectral analysis [7] of the EEG signal, extracting
several indicators from the frontal EEG in a user dened time window varying from 10
to 30 seconds. Because of the possibility of contamination of frontal EEG with muscle
activity, the EMG is also measured and displayed to the user, as well as the burst
suppression (BS) ratio (i.e. the ratio of periods of isoelectric EEG).
The BIS and EMG activity and variability have been linked in previous studies to
noxious activation [1517], leading to the development of an index referred to as the

Composite Variability Index

studies only.

(CVI), and made available by the manufacturer for clinical

The CVI index is a logistic regression of EMG power and dispersion

measures of the EMG and BIS index, as described by Greenwald and Rosow [16], varying

1. INTRODUCTION

Figure 1.4: Bispectral index monitor during data collection: manufacturer recommended target range for general anaesthesia of 40 to 60.

from 0 to 10. Figure 1.4 shows the BIS bilateral monitor during a general anaesthesia,
with both BIS and CVI trends for the left EEG channel. The CVI was also used in this
thesis, thanks to the collaboration of Aspect Medical Systems.
A bilateral BIS was provided for acquisition of the new index. The Bilateral BIS
monitor allows the EEG collection from both sides of the frontal cortex, providing two
EEG channels (left and right), allowing asymmetry assessment which is an indicator of
central ischemia.

1.1.2.2 Entropy Monitor

The Entropy monitor was developed by Datex-Ohmeda and released in 2003 (now part
of General Electric). Entropy analysis has application in several research areas, being a
measure of chaos/organization of a signal: the greater the disorganization, the greater
the entropy. The Entropy monitor applies this concept to translate the hypnosis state of
the patient, based on the entropy of the frontal EEG. It calculates two indexes applying
the Shannon entropy [18, 19] over the EEG for dierent frequency bands and time
windows, depending on the signal features:

the

State Entropy

(SE), which contains

information on the electroencephalographic activity in the frequency band of 0.8-32 Hz,

and varies from 0 to 91, and the

Response Entropy

(RE), containing information on

the electroencephalographic activity and also on the electromyographic activity, in an

extended frequency band of 0.8-47 Hz, varying from 0 to 100. By denition, the SE is
always lower or equal to the RE. Besides these indexes, burst suppression ratio is also
extracted from the EEG and presented to the user.

1.2 Pain
The information contained in the RE and SE, similar to what was described for the
BIS monitor, has been linked to noxious activation [20, 21].

1.2 Pain
The physical adjunct of an imperative protective reex.
Sherrington

According to the

International Association for the Study of Pain

(IASP) [22], pain can

be dened as an unpleasant sensory and emotional experience associated with actual

or potential tissue damage, or described in terms of such damage.

It is therefore a

multidimensional complex phenomenon.

Pain sensing, from sensor organs to pain assessment is described in the following
sections.

1.2.1 Sense Organs

Every organism needs information from the external environment in order to survive,
and the human body is not an exception.

The human body communicates with the

external and internal environments through a variety of sensory receptors which activate
the central nervous system.

These receptors act as transducers, converting forms of

energy from the environment into action potentials in the neurons [23, 24].

Figure 1.5: Schematic representation of the sensory process in the human body, redrawn
from [24].

1. INTRODUCTION
A sensation essentially involves sampling selected small amounts of energy from the
environment, and using it to control the generation of trains of action potentials as
represented in Figure 1.5.
The sensory receptors are activated in response to dierent forms of energy, which
requires dierent specialized types of receptors. Table 1.2, reproduced from [23], shows
dierent sensory modalities present in the human body, from which the rst eleven are
conscious; there are however other sensory receptors involved in information processing
that do not imply consciousness.
In the human body, the way a stimulus is perceived is conditioned by the sensing grid,
but also by the intrinsic mechanisms transmitting the information. As an example of the
rst, the cutaneous sense organs translate four dierent sensations, touch/pressure, cold,
warmth and pain. The sensors present in the skin for all these sensations are mapped
according to a grid that is not uniform, and one can only sense the stimulus if an area
of the mapping sensor is elicited. As an example of the second, if a stimulus of constant
strength is applied to a receptor, an adaptation phenomenon occurs, making the action
potentials of the sensory nerve decline over time.

This adaptation or desensitization

occurs in dierent ways depending on the sensor nerve involved in the process. Another
principle implicated in sensing is the law of projection. By this law, independently of
where the sensory pathway is stimulated, the conscious sensation produced is referred
to the location of the receptor [23].
An important issue in the sensory process is the stimulus intensity discrimination.
The way stimulus intensity is perceived is not linear and the way the information is
sent to the brain depends on two factors: the frequency of the action potentials, and
the number of receptors activated. The magnitude of the perceived sensation used to
be described by the Weber-Fechner law [23], which states that the magnitude of the
sensation is proportional to the logarithm of the stimulus intensity. The Weber-Fechner
law was replaced by the power function, as presented in Equation 1.1.

R = KS A

(1.1)

where R is the sensation felt, S is the intensity of the stimulus, K and A are constants
[23].
A power function is also used to describe the relation between frequency of the action
potentials generated in a sensory nerve ber and the intensity of the initiating stimulus.

10

1.2 Pain
Table 1.2: Sensory modalities and corresponding receptors and sense organs, reproduced
from [23].

Sensory Modality

Receptor

Sense Organ

Vision

Rods and cones

Eye

Hearing

Hair cells

Ear (organ of Corti)

Smell

Olfactory neurons

Olfactory

mucous

mem-

brane
Taste

Taste receptor cells

Taste bud

Rotational

Hair cells

Ear (semicircular canals)

Linear acceleration

Hair cells

Ear (utricle and saccule)

Touch-pressure

Nerve endings

Probably nerve endings

Warmth

Nerve endings

Probably nerve endings

Cold

Nerve endings

Probably nerve endings

Pain

Nerve endings

Probably nerve endings

Joint position and

Nerve endings

Various

Muscle length

Nerve endings

Muscle spindle

Muscle tension

Nerve endings

Golgi tendon organ

Arterial blood

Nerve endings

Stretch receptors in carotid

acceleration

movement

pressure
Central venous

sinus and aortic arch

Nerve endings

Stretch receptors in walls

pressure

of great veins, atria

Ination of lung

Nerve endings

Temperature of

Neurons in hypothalamus

blood in head
Arterial PO2

Glomus cells

pH of CSF

Receptors on ventral surface

of medulla oblongata

Osmotic pressure of

Cells

plasma

bly

in
other

OVLT

and

possi-

circumventricular

organs in anterior hypothalamus

Arteriovenous blood

Cells in hypothalamus (glu-

glucose dierence

costats)

11

Carotid and aortic bodies

1. INTRODUCTION
As the strength of a stimulus is increased not only the sense organs in immediate contact
with the stimulus are activated, but also sense organs in the surrounding area become
activated, resulting in a spread eect [24].

1.2.2 The Nervous System

Table 1.3: Types of nerve bers that carry aerent impulses to the central nervous system, adapted from [24]. A and B bers are myelinated, and C bers are unmyelinated.

Fiber Type

Function

Fiber

Conduction

Spike

Absolute

Diameter

Velocity

Duration

Refractory

(m)

(m/s)

(ms)

Period

12-20

70-120

0.4-0.5

0.4-1

5-12

30-70

3-6

15-30

2-5
<3

12-30
3-15

1.2

1.2

0.4-1.2

0.5-2

0.3-1.3

0.7-2.3

Proprioception;
somatic motor
Touch, pressure,
motor
Motor to muscle
spindles
Pain, cold, touch
Preganglionic
autonomic

Dorsal root

Sympathetic

Pain, temperature,
some
mechanoreception,
reex responses
Postganglionic
sympathetic

The human central nervous system (brain and spinal cord) is a complex system that
contains about

1011

neurons, where axons and dendrites form a complex communication

network [24]. The types of nerve bers that carry impulses generated in them to the
central nervous system are presented in Table 1.3. Some of these bers are covered with
myelin and convey information at faster rates.
The spinal cord also has a complex structure.

A cross-section of the spinal cord

allows the identication of its most important structures: it is divided in the so called
horns, due to its shape. Dorsal horns are divided into laminas I-VII, with I being the
most supercial and VII the deepest (see Figure 1.6).

12

Lamina VII receives aerents

1.2 Pain

Figure 1.6: Schematic representation of the dorsal horn of the spinal cord layers and
terminations of the three types of primary aerent neurons pathways, redrawn from
[23].

from both left and right sides of the body.

The peripheral nervous system has numerous nerves with dierent types of bers
connecting the central nervous system to the organs and limbs, conveying information
between them.
Three types of primary aerent bers mediate cutaneous sensation [23]:

1. Large myelinated A and A bers that transmit impulses generated by mechanoreceptors;

2. Small myelinated A bers, some of which transmit impulses from either cold
receptors, nociceptors or mechanoreceptors;

3. Unmyelinated C bers that are concerned primarily with pain and temperature;
some C bers also transmit impulses from mechanoreceptors.

Cerebral cortex areas related to the cutaneous senses are shown in Figure 1.7; the
areas involved in sensation in the cortex have been studied clinically using

Positron

Emission Tomography (PET) and functional Magnetic Resonance Imaging (fMRI). The
use of these technologies has increased the knowledge about the ways information is
processed in the human brain.

13

1. INTRODUCTION

Figure 1.7: Brain areas involved in somatic sensation (withdrawn from [24]).

The neuronal network is not innate or immutable, and can rapidly recongure itself
by experience to reect the use of the represented area.

Cortical plasticity may be

explained by the cortical connections of sensory units to the cortex, which have extensive
convergence and divergence, with connections that can become weak with disuse and
strong with use [23].

1.2.3 Pain and Nociception

Pain is a complex defense mechanism that alerts and prompts the body to avoid the
source of noxious stimulation and potential tissue damage. It is determined by a complex
process closely linked to the conscious perception of the sensation. Pain sense organs
are naked nerve endings that transmit pain impulses from almost every tissue in the
body to the central nervous system by two ber systems (see Table 1.3): A myelinated
bers, 2-5 m in diameter, which conduct at rates of 12-30 m/s and unmyelinated C
bers (dorsal root C bers), 0.4-1.2 m in diameter, which conduct at the slow rate of
0.5-2 ms. Both ber groups ending in the dorsal horn (see Figure 1.6). The synaptic
junction between the peripheral nociceptor bers and the dorsal horn cells in the spinal
cord are sites of considerable plasticity, for this reason the dorsal horn has been referred
as a gate; pain impulses may be subject of modulation crossing this gate, meaning
modied.

Some of the axons of the dorsal horn neurons end in the spinal cord and

brainstem, others enter the ventrolateral system, including the lateral spinothalamic
tract, a few ascend in the dorsal portion of the cord.

Some of the ascending bers

project to the ventral posterior nuclei, which are the specic sensory relay nuclei of the

14

1.2 Pain
thalamus, and from there to the cerebral cortex.

PET and fMRI studies in healthy

primary and secondary

humans indicated that pain activates cortical areas SI and SII (

somatosensory cortices ), and the cingulate gyrus of the side opposite the stimulus.

In

addition, the mediofrontal cortex, the insular cortex and the cerebellum are activated
[23, 24].
As a previous step to design a tool capable of translating the balance Noc/ANoc
during general anaesthesia, it is important to understand the inherent mechanisms
and pathways from the sensor activation to the central processing of pain potentials.
This means to understand and dene the physiological signals which could be used in
translating nociception objectively.
Pain may be described considering some studied features, such as conduction velocity, site of occurrence, and pathways [2325]. Impulses resulting from noxious activation are transmitted through two nerve ber systems, resulting in the physiological
observation of two dierent pain sensations: one sharp and localized, the other dull,
intense and diuse, usually referred as fast and slow pain, or rst and second pain.
Observations indicate that the further away from the brain stimulation is applied, the
greater the time lag between fast and slow pain.

This and other evidences suggest

that the fast pain is due to the activation of the A -bers and the second due to the
activation of the C pain bers.
Enervation in body tissues is not equally distributed, conducting to dierences in
the nature of the evoked pain for supercial and deep structures.
are poor in A nerve bers, inducing little rapid sharp pain.

Deep structures

Deep pain is poorly

localized, frequently associated with nausea and autonomic symptoms like sweating
and changes in blood pressure.

The sensory modalities in the viscera are similar to

those in the skin, however there are no proprioceptors, and few temperature and touch
receptors. In the central nervous system visceral and somatic sensations travel along
the same pathway, and cortical receiving areas of visceral sensation are intermixed with
the somatic receiving areas.
A dierent specic pain modality is muscle pain, resultant from contraction and
blood supply occlusion. It is thought that the muscle pain is the result of the accumulation of a chemical agent known as the Lewis P factor, and that as soon as the blood
supply is reestablished, it is washed out and the pain disappears.
Visceral and deep somatic pain initiate a reex contraction of nearby skeletal muscle,
more frequent in the abdominal wall, originated to protect the structures beneath from

15

1. INTRODUCTION
trauma.

Irritation of a viscus producing pain may not always be felt in the viscus,

but in some referred somatic structure; this pain is said to be referred to the somatic
structure. Deep somatic pain may also be referred, but supercial pain is not. When
the pain is referred it usually is associated to a structure that has evolved from the same
embryonic segment or dermatome (dermatomal rule). This is the reason why pain from
cardiac ischemia is usually felt in the arms (besides the chest).
Pain may be centrally inhibited; examples of this are stress analgesia in battle
wounded soldiers, or the occasions when touching and shaking an injured area leads
to pain relief. The last, called central inhibition is explained by the inhibition of pain
pathways in the dorsal horn gate, by stimulating large diameter touch-pressure aerents.
Inammatory processes also perform important roles in pain sensing, leading to
hyperalgesia (minor stimuli produce an exaggerated response) and allodynia (normally
innocuous stimuli as light touch cause pain); this is the result of the liberation of
cytokines and growth factors, referred to as the inammatory soup in the inamed
area, which facilitate perception and transmission in cutaneous areas and in the dorsal
horn [23] .
Finally, another important phenomenon in pain sensing is the pain resulting from
injury in the nerve bers themselves, called neuropathic pain, which is usually accompanied by hyperalgesia and allodynia. It is a dicult condition to treat and occurs in
various forms [23, 24].

1.2.4 Methods of Induction and Assessment of Pain

Searching for new and accurate methods to assess and manage pain, numerous studies
were conducted both in animal and human subjects.
Methods to elicit pain are fully described in the literature [26] and depending on the
nerve bers and brain areas to be inspected, dierent methods are used. Tables 1.4, 1.5
and 1.6, present some experimental methods for pain induction in humans (cutaneous,
muscular and visceral, respectively). These methods are also used to explore drugs' site
of action and ecacy in pain treatment, to known standardized stimuli.
The method of pain induction in human subjects, its repeatability and standardization, are very important in the description of pain processes and to compare dierent
subjects for the same stimulus modality [27]. In spite of the fact that stimulus can be
easily controlled, namely in an experimental setting, one encounters a challenge when

16

1.2 Pain
Table 1.4: Stimulus modalities for induction of experimental cutaneous pain in humans,
adapted from [26].

Modality

Stimulus

Electrical

Transcutaneous
Intracutaneous
Intraneural

Thermal
Heat

Radiant (light, infrared)

Hot water
Contact thermode

Cold

Ice water
Contact thermode

Mechanical

Pinprick
Pressure
Impact Stimuli
Pinch

Chemical

Capsaicin
Mustard oil
Hydrogen ions
Inammatory-related substances (serotonin, bradykinin, histamine,
potassium ions, substance P)

trying to translate the way humans perceive pain, and the way the sensation may be
objectively described.

Thus, objective methods to measure pain have attracted wide

attention from researchers and clinicians.

The most widely used methods to measure pain intensity or unpleasantness are the

Visual Analogue Scales

(VAS),

Numerical Rating Scales

(NRS) and verbal scales [28].

In the rst method the subject is asked to rate pain by positioning a cursor along a ruler;
this ruler has the left side extremity marked as No Pain, and the right side extremity
identied as Very Severe Pain. The back of the ruler, not seen by the patient, usually
has a scale from 0 to 10 or 100, allowing the measurement of relative positions for pain
assessment. In the second method a numerical rating scale usually varying from 0 (no
pain) to 10 or 100 (worst possible pain) is explained to the subject who auto-evaluates
his/her pain in the ordinal scale. In the last method, the subject is asked to classify
his/her pain in one of the verbal descriptors, usually translated into an ordinal scale. A

17

1. INTRODUCTION
Table 1.5: Stimulus modalities for induction of muscular pain in humans, reproduced
from [26].

Type

Modality

Stimulus

Endogenous

Ischemia

Tourniquet + contractions

Exercise

Concentric contractions
Eccentric contractions

Exogenous

Electrical

Intramuscular
Intraneural

Mechanical

Pressure

Chemical

Hypertonic saline,

potassium,

levoascorbic acid,

capsaicin bradykinin, serotonin, calcitonin generelated peptide, neurokinin A, substance P, hydrogen ions

Table 1.6: Stimulus modalities for induction of visceral pain in humans, reproduced
from [26].

Modality

Stimulus

Electrical

Transmucosal

Thermal
Heat

Water

Cold

Water

Mechanical

Distension

Chemical

Capsaicin

schematic representation of these pain assessment techniques are shown in Figure 1.8.
Other methods of pain evaluation using the subject's participation are the multimodal questionnaires, in which pain is evaluated as a multifaceted process.

As an

example, the McGill Pain Questionnaire, also known as McGill pain index, is a scale
for pain rating developed at the McGill University, which is used to evaluate pain as a
multifaceted phenomenon using Psychology techniques (see Figure 1.9) [29]. This is one
of the developed methods to evaluate pain in patients able to cooperate, describe pain
sensation and activity limitation; several were developed for specic recurrent chronic

18

1.2 Pain

Figure 1.8: Schematic representation of pain assessment techniques. From top to bottom: visual analogue scale, numerical rating scale (0 meaning no pain and 10 worst
possible pain), and a verbal scale.

pain patients [3032].

These methods are useful to describe pain sensations and consequent limitations,
however due to the complex nature of pain the terms used to describe it must be carefully chosen.

The McGill pain questionnaire was originally developed in English,

nevertheless to use this method it is important to have the document translated, adapted and validated to the subjects' mother language. In [33], the original McGill pain
questionnaire was translated and adapted to Portuguese (Brazil), trying to maintain
the original objective of the questionnaire, and adapting some of the pain descriptors
to better t the subjects. Figure 1.9 shows the original (on left) and adapted (on right)
versions of the McGill Pain Questionnaire.
The use of the described rating scales is limited by patients' collaboration and selfevaluation of pain sensation. In the case of the multifaceted questionnaires they have
practical applicability only in the assessment of chronic pain, with no practical use
in acute pain patients (time consuming).

Moreover, the produced measures have

limitations in their mathematical interpretation and processing within and between

subjects.
Although the described methods are useful in pain management, they are insucient

19

1. INTRODUCTION

Figure 1.9: McGill pain questionnaire: on left the original version in English, and on
right the adapted version in Portuguese (Brazil), withdrawn from [33].

in many demanding practical clinical situations, such as in patients unable to communicate. This reveals the need in the clinical setting for a tool to objectively translate
the patient's state concerning pain or Noc/ANoc balance. This is the case for patients
under general anaesthesia, subjected to severe surgical noxious stimulation, and unable
to express the Noc/ANoc balance state. Objective methods for nociception assessment
are discussed in section 1.4.

1.3 Anaesthetic Drugs

Divide each diculty into as many parts as is feasible and necessary to resolve it.
Ren Descartes

General anaesthetics are potent drugs that depress the central nervous system; they
are used to allow patients to be subjected to otherwise painful or very unpleasant
procedures, namely surgery [34].

Under the eect of anaesthetics and the monitored

care of a specialized physician, severe noxious stimuli may be rendered virtually harmless. Anaesthesiologists titrate a set of drugs according to specic requirements of each

20

1.3 Anaesthetic Drugs

procedure, and individual responses, to obtain a balanced anaesthesia based on the
equilibrium between the three components shown in Figure 1.2. In this section analgesics and hypnotics are described, with particular focus on the drugs used in the studies
comprised in this thesis, namely the analgesic remifentanil and the hypnotic propofol.
The pharmacokinetic and dynamic models described in the literature and their use in
the administration of these drugs will also be described.

1.3.1 Analgesics
Analgesics are potent anaesthetic drugs used to abolish responses to noxious stimulation, meaning nociception [34].

During a surgical procedure these drugs are used

according to patients' needs, and manually titrated to obtain homeostasis in conjunction with other anaesthetic agents.

These drugs must be carefully used in order to

prevent adverse eects, and only by experienced personnel properly assisted by all the
required equipment to quickly respond to a crisis situation [35].
Opioid analgesics are currently used in clinical practice to produce analgesia. The
term

opioid

refers to all compounds related to opium, a natural product derived from

the poppy ower. Well known examples of drugs derived from opium include morphine,
codeine, pethidine, and many semisynthetic derivatives. Opioids mimic the endogenous
opioid peptides endorphins, the naturally occurring ligands for opioid receptors, that
are part of the endogenous opioid system [34].

1.3.1.1 Remifentanil

Figure 1.10:

Remifentanil 2D molecular schematic representation (methyl 1 - (3 -

methoxy - 3 - oxopropyl) - 4 - (N-phenylpropanamido) piperidine - 4 - carboxylate)

[36].

21

1. INTRODUCTION
Remifentanil, also known by its molecular description methyl 1- (3-methoxy -3oxopropyl) -4- (N-phenylpropanamido) piperidine -4- carboxylate, (see Figure 1.10) is a
fast acting opioid drug, used to provide analgesia to patients under general anaesthesia
or in the intensive care unit under mechanical ventilation [35, 37].
The endogenous opioid system plays an important role in the sensory processes, by
inhibiting responses to noxious stimuli. It has also impact in gastrointestinal, endocrine
and autonomic functions, and emotional and cognitive roles in the modulation of learning and memory. There are four major opioid receptor types:

, k ,

and N/OFQ

[34].
Remifentanil is a selective
tion. The

-opioid

-opioid

agonist with rapid onset and short action dura-

action is antagonized by narcotic agonists, like the naloxone. After

administration of the recommended doses of remifentanil, its eective half-life time is

of 3 to 10 minutes, being metabolised by plasma and non-specic tecidular sterases.
After being metabolised remifentanil results in the formation of a carboxylic acid metabolite (essentially inactive, and excreted by the kidneys).

The metabolite half-life

in adult healthy patients is of about 2 hours. In patients with normal renal function
the time for 95% of renal elimination of the primary metabolite of remifentanil is of 7
to 10 hours. Due to its very short action duration, the side eects of remifentanil are
limited to the immediate period following administration. Interruption or cessation of
drug administration leads to the return to the initial state in 10 minutes [34, 35, 37].
Remifentanil use is approved by the Portugal National Authority of Medicines and
Health Products (INFARMED), and its main characteristics are presented in Appendix
A, including side-eects and dosage recommendations. Remifentanil may be administered using either manual infusion or model based infusion, with the

and Dynamic

Pharmacokinetic

(PKPD) models described in the literature [38, 39].

1.3.2 Hypnotics
Hypnotic drugs are used either for induction of an altered state of consciousness or
for its maintenance, and are titrated according to patient's needs and procedures' requirements. As the analgesic drugs, hypnotics are also potent anaesthetic drugs with
associated risks of overdosing (e.g. higher mortality or cognitive dysfunction incidence)
[4042] and underdosing (awareness) [12, 13].

22

1.3 Anaesthetic Drugs

Regarding the use of hypnotic drugs, anaesthesia maintenance requires continuous
administration and individual titration of drugs. For that, inhalational and intravenous
hypnotic drugs may be used, diering in the mechanism of action, and obviously on
the route of administration. Another important dierence is the fact that administration of inhaled anaesthetics allows continuous measurement of its concentration in the
body, since expired concentration of the gas may be easily measured, indicating blood
concentration of the anaesthetic. Continuous measurement of blood concentration of
intravenous anaesthetics is possible, but performed only in a research context.
The most commonly used inhalational anaesthetics are sevourane, desurane, isourane, nitrous oxide and xenon; the most commonly used intravenous agents drugs are
propofol, etomidate, ketamine, sodium thiopental and dexmedetomidine.

1.3.2.1 Propofol

Figure 1.11: Propofol 2D molecular schematic representation (2,6 - bis(propan - 2 yl)phenol) [43].

Propofol, also known by its molecular description 2,6-bis(propan-2-yl)phenol (see

Figure 1.11), is an intravenous hypnotic drug with rapid onset and oset of drug eect,
and fast elimination from the body. Pharmacokinetics of propofol are characterized by
a rapid initial distribution of 2-8 minutes, with high clearance and slow diusion of
drug from the peripheral compartment. The propofol is metabolised in the liver. It is
highly protein bound, and pharmacokinetics may be aected by conditions which alter
the serum protein levels [34, 44]. It produces hypnotic eects by altering the GABAA
receptor function, stimulating the inhibitory function of the neurotransmitter GABA
and depressing nervous system activity [34, 44].
As any anaesthetic drug, propofol also has secondary eects, producing a dosedependent cardiovascular depression, which most visible eects are the decrease in blood
pressure, vasodilation, and cardiac output reduction. It also produces respiratory depression, by aecting chemoreceptor sensitivity, reducing the ventilatory response to

23

1. INTRODUCTION
hypercapnia [34, 44]. Propofol decreases brain metabolism, cerebral blood ow, intracranial pressure, and presents a neuroprotective eect [44].
Propofol is the most widely used hypnotic anaesthetic.

There has been dispute

whether propofol, besides being an hypnotic drug, could also have some analgesia eect
[45, 46]. It is known to act on a specic site on the

subunit of the GABAA receptor,

which could explain some analgesic eect [34].

The Portugal National Authority of Medicines and Health Products (INFARMED),
has approved propofol use. It may be administered by manual infusion or model based
infusion [4750].

1.3.3 Pharmacokinetic and Dynamic Modelling and Target Controlled Infusion

Figure 1.12: Drugs' action path in the human body, from drug dose administration to
observed clinical eect (adapted from [51]).

When the clinician administers a certain drug amount to a patient, a specic clinical
eect is expected; this is based on the understanding of what happens following drug
administration, meaning the drug path in the human body, as represented in Figure
1.12.

This drug action path may be resumed in the relations between administered

24

1.3 Anaesthetic Drugs

dose, plasma concentration and eect-site concentration, the pharmacokinetic phase,
and in the relation between eect-site concentration and observable clinical eect, the
pharmacodynamic phase [51].
Due to the fact that plasma drug concentrations are not measured on real time
in the clinical setting, models describing drugs' pharmacokinetics have been used to
simulate and to predict the concentrations at dierent body compartments and the
clinical eect, or

eect-site concentration

(Ce).

A few studies addressed this issue

[38, 39, 4952], in which blood samples were collected to evaluate

plasma concentration

(Cp), to adjust pharmacokinetic models and to reproduce the expected behaviour on

the patient. Mathematical equations are used to represent what the body does to the
drug (pharmacokinetic phase) and what the drug does to the body (pharmacodynamic
phase), employing the concept of body compartments to represent drug transfer in
the system.
A compartmental model structure is commonly used to describe the pharmacokinetic and dynamic phases. The Figure 1.13 shows a compartmental model, where the
constants
ment

kij ,

j (k10

represent the drug transfer rate from the compartment

is the elimination rate), and

V1 , V2

and

V3

to the compart-

are the compartments' volumes.

The central compartment may be considered as the plasma compartment (although the
volume

V1

may be higher than the real plasma (blood) volume), the compartment 2

corresponding to the fast distribution compartment or vessel rich, and the compartment
3 represents the slow distribution compartment or vessel poor.

V1 , V2

and

V3

are howe-

ver theoretical volumes, used to predict plasma concentrations, with no anatomical or

physiological meaning [51].
The compartment 1 and compartment
by the parameter

ke0 .

The

ke0

achieve the maximum value,

e,

also presented in Figure 1.13, are linked

value is obtained from the time required for the eect to

Time To Peak Eect

(TTPE), after the rapid introduction

of the drug in the system (bolus). The TTPE represents the moment in which the eectsite and plasma concentrations equalize. In the studies leading to the development of
PKPD models, EEG derived parameters are used as surrogates of the Ce to assess the
clinical/dynamic eect, and determine the point of maximum eect (TTPE).
The PKPD presented in Figure 1.13 is a dynamic model, with state-space representation as presented in Equation 1.2.

25

1. INTRODUCTION

Figure 1.13: Three compartments pharmacokinetic (PK) model, followed by the pharmacodynamic (PD) model and Hill equation, translating the eect-site concentration
into the output measurable eect.
compartment

Where

i to the compartment j (k10

kij ,

represents the drug transfer rate from

is the elimination rate), and

V1 , V2

and

V3

are

the compartments' volumes; in the Hill equation

E0

there is no drug in the system (Ce

is the drug eect-site concentration

= 0), EC50

is the eect baseline value, when

necessary to achieve half of the maximum measurable eect (E ), and

represents the

steepness of the descent of the eect response to the drug.

k12 k13 k10 k21
k31
0
m 1 (t)
m 2 (t)
k
k
0
0
12
21

m 3 (t) =
k13
0
k31
0
ke0

0
0
k
Ce(t)
e0
V1

m1 (t)
1
m2 (t) 0

m3 (t) + 0
Ce(t)
0

r(t)

(1.2)

where

r(t)

is the drug infusion rate, and

mi

is the drug mass in compartment

i (i =

1, 2, 3).
The relation between eect-site concentration and the measurable clinical eect (E )

26

1.3 Anaesthetic Drugs

(e.g. EEG derived indexes), is usually modelled using a sigmoid equation [5355]. The
Hill equation, presented in Figure 1.13, is the most commonly used. In this equation

E0

is the baseline value, when there is no drug in the system (Ce

= 0), EC50

is the

drug eect-site concentration necessary to achieve half of the maximum eect (drug
potency), and

represents the steepness of the descent of the eect response to the

drug, as given in Equation 1.3.


E(t) = E0 1

Ce(t)

Ce(t) + EC50


(1.3)

A similar approach for the concomitant administration of an hypnotic (H) and an

analgesic (A), uses a bivariate Hill equation, which also takes into account drugs' relations and synergistic eects [53, 56, 57]. The core idea is to use a combination of the
two drugs, which behaves as a new drug. Normalization of the drugs' concentration by
their potencies (EC50A and

EC50H ), as described for the univariate Hill equation, leads

to Equations 1.4 and 1.5.

UA (t) =

CeA (t)
EC50A

(1.4)

UH (t) =

CeH (t)
EC50H

(1.5)

As the normalized concentration of analgesic (UA ) and hypnotic (UH ) are computed
by Equations 1.4 and 1.5, a family of drugs may be dened as a ratio of

UA

and

UH

given by the Equation 1.6.

(t) =
By denition,

UH (t)
UH (t) + UA (t)

(1.6)

ranges from 0 (analgesic only) to 1 (hypnotic only).

Thus, the

concentration response relation for any ratio of the type of interaction may be described
by Equation 1.7.


E(t) = E0 1
where

((UH (t) + UA (t)) /U50 (t))

1 + ((UH (t) + UA (t)) /U50 (t))

is the measurable eect,

E0

is the eect at zero concentrations,

steepness of the concentration response relation, and

associated with 50% of maximum eect at ratio

27

U50

(1.7)

is the

is the number of units (U)

1. INTRODUCTION
The relation in Equation 1.7, and the expected response to noxious stimulation, were
employed in research studies [5860] addressing the combination of the hypnotic and
analgesic drugs' concentrations to obtain an index of probability of response to noxious
stimulation,

Noxious Stimulation Response Index

(NSRI).

Figure 1.14: Screenshot of the educational software specially developed for the simulation of target-controlled infusion anaesthesia using BIS or Entropy. It shows the surface
relational model between propofol and remifentanil's synergistic eect on the hypnosis
indexes [56, 57].

The bivariate Hill methodology was also employed in some studies developed by
us, nevertheless outside the context of this thesis, in the assessment of the relations
between hypnotic propofol, analgesic remifentanil and the measurable eect translated
by BIS and Entropy indexes [56, 57]. In [61, 62], data during propofol and remifentanil
anaesthesia, with BIS and Entropy monitoring were used to adjust dierent surface
model structures, and construct the synergy curves of eect, later implemented in an
education simulation tool (see Figure 1.14).
Although these models try to predict the average patient response to drug administration, there is no such thing as an average patient. The problem resides in the
known inter and intra-individual variability, which should be taken into account [54, 63]
when administering these or any drugs.

The aforementioned concepts and all the PKPD modelling, allow to implement these

28

1.3 Anaesthetic Drugs

models into medical equipment for drug delivery. Such drug delivery systems administer drugs based on the simulated concentrations for a specic patient, usually taking
into account age, weight, height and gender. These systems, called

Infusion

Target Controlled

(TCI) systems, continuously simulate the drug course in the human body,

and allow drugs' administration by eect-site or plasma concentration target steering.

It means the anaesthesiologist may choose the drug target concentration for that specic patient based on the modelled expected response, and administer the drug dose
necessary to achieve the desired target, automatically.
TCI systems have been implemented in some commercially available electronic syringe pumps, and may be used in clinical environment.

Although these systems are

commercialized in several countries, the United States of America Food and Drug Administration (FDA) has not yet approved it.
Two clinical studies have been conducted for the purpose of this thesis, both used
propofol as hypnotic, and remifentanil as analgesic. For both studies, TCI was applied
as method of drug administration, with the Schnider [50] and Minto [39] PKPD models.
The clinical protocols for these studies are detailed in Chapters 2 and 5. In the next
sections, the main features of these models are described.

1.3.3.1 Remifentanil Pharmacokinetic and Dynamic Model

Minto PKPD model was developed based on a database of 65 healthy adult volunteers
(4 excluded), between 20 and 85 years of age [38, 39]. EEG was used to measure the
cerebral eect of the opioid.

The model structure chosen was a three compartment

model, and the model parameters found to be related to the demographic data of the
subjects were age, weight, height and gender.
Table 1.7 shows the relations between patient's demographic data and the estimated
model parameters, with corresponding estimated percent coecient of variation (CV).
Weight (in kg), height (in cm) and gender were used in the assessment of lean body
mass (LBM), one of the model variables, given by Equation 1.8 for men, and Equation
1.9 for women.

Men:


LBM = 1.1W eight 128

29

W eight
Height

2
(1.8)

1. INTRODUCTION

Women:


LBM = 1.07W eight 148

W eight
Height

2
(1.9)

The relation used to obtain the LBM is a quadratic expression, that produces paradoxical values for overweighted patients [64]. Pharmacokinetic models were not adjusted
or validated on obese patients, and should be carefully used in this population. Some
methods to overcome the LBM paradoxical values issue have been proposed in the commercially available TCI systems, suchlike limiting the body weight for a certain height,
guaranteeing that the LBM calculus is not on the descending portion of the quadratic
curve [65].

1.3.3.2 Propofol Pharmacokinetic and Dynamic Model

Several pharmacokinetic models have been proposed for propofol in adults [49, 50] and
children [4749].

The model used to administer propofol by TCI in the studies that

comprise this thesis was the three compartments Schnider PKPD model [50].
The Schnider PKPD model parameters were obtained using a population of 24
healthy adult volunteers, between 26 and 81 years of age. Tables 1.8 and 1.9 present the
estimated population parameters, adjusted to subjects' demographic data age, weight,

Table 1.7: Estimated relations for the parameters of the three compartments Minto
pharmacokinetic and dynamic model, and corresponding coecient of variation (CV).
The parameters incorporate patient's age and lean body mass (LBM) [38].

Compartments' Volume (l)

V1
V2
V3
k10
k12
k13
k21
k31
ke0

CV (%)

5.1-0.0201*(Age-40)+0.072*(LBM-55)

26

9.82-0.0811*(Age-40)+0.108*(LBM-55) (fast)

29

5.42 (slow)

66

(2.6-0.0162*(Age-40)+0.0191*(LBM-55))/V1

14

(2.05-0.0301*(Age-40))/V1

36

(0.076-0.00113*(Age-40))/V1

41

(k12*V1)/V2

(k13*V1)/V3

0.595-0.007*(Age-40)

60

Transfer Rates (min1 )

30

1.3 Anaesthetic Drugs

Table 1.8: Estimated relations for the Schnider three compartments pharmacokinetic
model parameters.

It incorporates patient's age, weight, height and lean body mass

(LBM) [50].

Compartments' Volume (l)

V1
V2
V3
k10
k12
k13
k21
k31
ke0

Transfer Rates (min1 )

[

2 +7 *(Age-53)
3

4 +8 *(Weight-77)+9 *(LBM-59)+10 *(Height-177)]/V1

[ 5 + 11 *(Age-53)]/V1
6 /V1
[ 5 + 11 *(Age-53)]/V2
6 /V3
0.456

Table 1.9: Average and standard-deviation of the estimated

(1=1,...,11) parameters

in the Schnider pharmacokinetic and dynamic model [50].

Estimated Parameters Average Standard-Deviation

1
2
3
4
5
6
7
8
9
10
11

4.27

0.278

18.9

2.33

238

34.9

1.89

0.059

1.29

0.112

0.836

0.044

-0.391

0.07

0.0456

0.009

-0.0681

0.017

0.0264

0.009

-0.024

0.005

height and gender. LBM was employed as described for the Minto model in the previous
section (Equations 1.8 and 1.9).
The parameters obtained by the models are average population parameters, and

31

1. INTRODUCTION
exhibit variability. Although these models try to predict the drug path in the human
body, they are only estimations of the expected drug's concentrations for an average
patient.

1.4 State of the Art on the Analysis and Control of the

Nociception Level
Learn from yesterday, live for today, hope for tomorrow.
The important thing is not to stop questioning.
Albert Einstein

Nociception may be dened as the aerent activity produced in the peripheral and
central nervous system by stimuli that have the potential to damage tissue [23]. Nociception triggers responses that include tachycardia, hypertension, sweating and electromyographic activity.
Every time a noxious/painful stimulus is applied, the human body reacts, alerting
the organism to avoid or stop the noxious stimulus source.

This activity is initiated

by nociceptors capable of detecting mechanical, thermal or chemical changes, that rise

above a certain threshold. Once stimulated, a nociceptor transmits a signal that travels
along the spinal cord, to the brain, triggering the nociceptive responses [23, 24]. The
way sensing occurs is conditioned by many factors as attention [66], handedness [67, 68],
source of stimulation and activated nerve bers [6971], subject gender [7274], among
others.

Concomitant exposure to another noxious stimulus may attenuate sensing,

a phenomenon known as

Diuse Noxious Inhibitory Control

(DNIC), which was also

studied, although not as part of this thesis [75].

Because the objective of this thesis is to measure nociception responses to noxious stimulus during general anaesthesia, the conscious perception is not a bias factor
however, measuring nociception responses without the patient's collaboration and pain
auto-evaluation, is also a research challenge.

1.4.1 Objective Methods of Nociception Measurement

The study of objective methods for nociception measurement has attracted the attention
of the scientic community over the last decades. Researches in this eld dier mostly
on the physiological signals explored in the translation of the Noc/ANoc balance. In

32

1.4 State of the Art on the Analysis and Control of the Nociception Level
order to give a comprehensive analysis of this topic, the following sections are organized
according to the physiological signals employed in the search of a Noc/ANoc balance
index.
Most of the proposed approaches to the assessment of the Noc/ANoc balance use
pre-processed indexes, and are unitless.

1.4.1.1 Cardiovascular, EMG and EEG Derived Indexes

Figure 1.15: Results obtained by Seitsonen and colleagues, withdrawn from [20]: Relative (to prestimulus interval) group average ( standard error of mean) responses to
skin incision in movers (dashed line) and non-movers (solid line) of (A) EEG response
entropy (n=10 and n=12 for movers and non-movers, respectively), (B) RRI (n=12 and
n=14), (C) PPG amplitude (n=12 and n=14), and (D) PPG notch amplitude (n=12
and n=14). Skin incision marked with t=0.

Cardiovascular and EEG derived indexes were explored to translate nociceptive

responses both in animal [76] and human subjects [20, 7779].

33

1. INTRODUCTION
Seitsonen et al. [20] found a relation between several variables and the nociceptive
response to skin incision during general anaesthesia.

Electrocardiogram

(ECG),

Pulse

Plethysmography (PPG) and EEG-related variables showed signicant dierences before

and after surgical incision.

According to the authors, the combination that better

distinguished between patients who moved following incision and those who did not,
was composed by the Response Entropy of the EEG (RE, Entropy hypnosis monitor),
the

time interval between ECG QRS complexes (RR interval) and PPG notch amplitude

(see Figure 1.15).

In other study, Rantanen et al. [77] developed a scale to evaluate the nociceptive-antinociceptive balance, which was called

Clinical Signs-Stimulus-Antinociception

(CSSA)

score. This score was a combination of patients' reactions and clinical signs of nociception, remifentanil levels and an estimative of noxious intensity of incision. The proposed
scale was based on few experts opinion. Dierences between post and pre-incision values of

Heart Rate Variability (HRV), PPG and pulse transition time related parameters

were analyzed o-line to evidence the best predictors of CSSA. Those best predictors of
CSSA were used to develop a so called

Response index of Nociception (RN), scaled from

0 to 100. This index was later tested in a dierent data set composed by 10 patients'
data. The authors concluded that HRV, RE, RE-SE and PPG variability were the best
predictors of CSSA [77].
In a dierent study by Huiku et al [78], a similar approach was used in the assessment of surgical stress, where nger photoplethysmography and electrocardiography
waveforms were recorded and various waveform parameters extracted o-line. The

tal Surgical Stress

To-

(TSS) for a patient was estimated based on stimulus intensity and

remifentanil concentration.

A so called

Surgical Stress Index

(SSI) was then develo-

ped to correlate with the TSS estimate and the performance of SSI validated within
a validation data set [78].

Beat Interval

(HBI) and

SSI was computed as a combination of normalized

Plethysmography Pulse Wave Amplitude

Heart

(PPGA), expressed

by HBInorm and PPGAnorm respectively, as shown in Equation 1.10.

SSI = 100 (0.7 P P GAnorm + 0.3 HBInorm)

(1.10)

According to the authors, in this study the developed index (SSI) responded to
changes in the remifentanil level and to dierent nociceptive stimuli.

34

The proposed

1.4 State of the Art on the Analysis and Control of the Nociception Level
index was then used in dierent clinical studies [8085] to verify its applicability, seeming
to reect changes in surgical stress and analgesic dose.
In a study by Struys et al. [83], the SSI was monitored under general anaesthesia for
dierent combinations of propofol and remifentanil target eect-site concentrations [78].
At each steady-state of the remifentanil eect-site concentration, the maximum change
of the SSI, SE, RE and PPGA was compared with its baseline value. It was found that
the static and dynamic values of SSI correlated better to the remifentanil eect-site
concentration than the other recorded variables, and that this was independent of the
hypnotic eect-site concentration.
Recently a new method was proposed in [86], also using PPG amplitude (P P GA)
and pulse-to-pulse interval (P P I ), the so called

Autonomic Nervous System State Index

(AN SSI ). According to the authors the proposed index adequately reected stimulation and was able to dierentiate between patients who received an inltration with a
local anaesthetic in the tonsils, from patients who did not, demonstrating a greater amplitude response when compared to the

Surgical Pleth Index

(SPI, previously referred

to as SSI).

AN SSI = 100 (AN SS/AN SSmax) 90

(1.11)

AN SS = P P I P P GA

(1.12)

where

and

AN SSmax

is the maximum

AN SS ,

given by Equation 1.12, observed in the pati-

ent.
ANSSI is an open source index, and may be computed by Equations 1.11 and 1.12.
Some authors [15, 16, 8789] have related hypnosis indexes variability to the nociceptive response of anaesthetized patients, using BIS, Entropy indexes (SE, RE) and

Auditory Evoked Potential Index

(AAI).

Figure 1.16 shows a typical trend of the Bispectral Index (hypnosis index) during general anaesthesia, with the descending phase (anaesthesia induction), the steady phase
(maintenance), and the nal ascending phase (recovery). Examining the index during
the maintenance phase, while the patient is subjected to surgical noxious stimulation, it
may be observed that the index has an intrinsic variability. This variability may be affected by several external factors. The possibility that BIS variability could be aected

35

1. INTRODUCTION

Figure 1.16: Bispectral Index (BIS) trend during general anaesthesia, with the induction, maintenance and recovery phases. The dashed lines represent the BIS manufacturer recommended target range for general anaesthesia (40-60). Data from one of the
patients in this study.

by the Noc/ANoc balance was explored [15, 17]. The core idea behind this approach
is to use the apparently random variability of the signal, to translate the nociceptive
responses of an anaesthetized patient, based on the arousal responses induced in the
EEG by noxious stimulation.
Noxious stimulation activates peripheral nociceptors which send aerent signals via
the spinal cord, midbrain and thalamus to a broad distribution of regions in the cerebral
cortex.

Activating the nociceptive pathways during general anaesthesia appears to

stimulate the cerebral cortex and elicit changes in the EEG [15].
Following this line of thought, one of the developed indexes is the so called Composite
Variability Index (CVI) [16]. It was developed as a logistic regression based on the BIS
index and frontal EMG activity, recorded by the same BIS monitor.

The authors

in [15, 16] related the increase of the standard-deviation of each signal and of EMG
power to somatic responses, and demonstrated that early detection of arousal could
be achieved this way. In another study [17], the authors also observed that CVI, BIS
standard-deviation, but not EMG standard-deviation increased in response to surgical
incision, with slightly increased amplitude in the contralateral site to stimulation. A
bilateral sensor was used in these studies.
Entropy indexes were also proposed as surrogate measures of nociception [90].

study from Takamatsu et al. [89] employed Entropy monitoring to assess the depth of
anaesthesia, and used the dierence between the two indexes provided in this monitor

36

1.4 State of the Art on the Analysis and Control of the Nociception Level
- the State Entropy (SE) and the Response Entropy (RE) - to estimate nociception
during general anaesthesia.

According to the authors, noxious stimulation increased

the dierence between RE and SE, however, it did not always indicate inadequate
analgesia.
In another study [91], authors observed that RE increased in response to tracheal
intubation, while SE and BIS did not, and also that landiolol (an antiarrhythmic) and
remifentanil suppressed this increase.

Suppression of the response was also observed

by Kawaguchi et al. [92], where the authors demonstrated that the dierence between
SE and RE was suppressed by increasing the muscle relaxant, meaning that it diers
according to the paralysis state of the patient. In a dierent study conducted by Valjus
et al. [93], the authors concluded that RE did not seem to be more sensitive than SE to
guide opioid administration, and Weil et al. [94] observed that an increase in Entropy
predicted motor response to intubation and incision, but not a haemodynamic response.
Although the methodology presents several limitations, specially for paralyzed patients,
a study by Mathews et al. [21] proposed an automatic algorithm for the administration
of remifentanil based on the RE-SE dierence, demonstrating its feasibility.
It should be highlighted that Entropy based, BIS based and PPG wave based indexes, were in the majority studied through the direct initiative or direct support of the
monitors' developers. Nevertheless, from all the indexes described to this point, only
SPI became commercially available.
Some studies [95, 96] using AAI monitoring also proposed a method that used this
index to translate the nociceptive response of an anaesthetized patient. The dierences
between this index and the previously referred are the site of signal collection, postprocessing and involved processes.

SE, RE and BIS use the frontal EEG, reecting

cortical activity, while the AAI index is obtained as a response to an auditory stimulus, corresponding to the transfer of auditory information from the medial geniculate
body to the primary auditory cortex (closer to the somatosensory area). This method
was evaluated in [95], demonstrating that during a BIS guided anaesthesia, the AAI
response diered according to the analgesic regimen, and it could be used to translate
the anti-nociception of a patient under balanced anaesthesia.
A dierent approach was proposed by Cividjian et al. [97], where beat-by-beat blood
pressure and heart rate (Finapress

R

measurements), were used to assess the baroreex.

Nociception is known to depress cardiac baroreex, and the authors proposed a new
index, the

Cardiovascular Depth of Anesthesia

37

T M ), varying from 0

index (CARDEAN

1. INTRODUCTION
to 100, that reects baroreex inhibition [97]. It was also demonstrated that CARDEAN
allows to predict intraoperative movement better than BIS [97, 98], and its use reduced
the occurrence of unexpected movement during endoscopy [99, 100].

In a dierent

study [101], CARDEAN was directly associated to stimulus intensity and inversely to
analgesic concentration.

1.4.1.2 Facial Electromyography

Figure 1.17: Facial muscles involved in the innate response to noxious stimulation [102].

A study conducted by Bennet et al. [103], explored the human facial expression in
response to painful stimuli.

Facial expression is the result of the action of numerous

muscles as shown in Figure 1.17. This study the autonomic facial expression response
to noxious stimulation as a measure of Noc/ANoc balance.
The innate face expression to noxious stimuli has well dened patterns according to
the sensation, with dierent muscles contraction and intensities. Basically this study
[103] suggested that to measure electromyography activity based on two specic muscles - the

orbicularis oculi

and the

corrugator

- could provide information about the

nociceptive response, even when muscular paralysis was induced.

In this study specially developed and highly sensitive facial electrodes were placed

corrugator muscle
these two and the frontalis

over the

(C) and the

orbicularis oculi

(O), and the relation between

(F) evaluated as described in Equation 1.13.

R=

(C F ) + (O F )
C + O + 2F

38

(1.13)

1.4 State of the Art on the Analysis and Control of the Nociception Level
The collection of physiological signals such as facial EMG must be driven with
caution due to the possible internal and external interferences, suchlike cable motion
artifacts [5, 104, 105].

1.4.1.3 Skin Conductance

From the early days of ether anaesthesia, presence of sweat and tears were widely
recognized as clinical indicators of anesthesia inadequacy.

Storm [106] developed a

monitor that measures skin conductance, a surrogate of sweat production, analyzing

uctuations of skin conductance caused by the sweat segregation in the patient's hand.
This method was rst developed for use in preterm infants [107] and later evolved for
adult patients' use. The skin conduction uctuations analysis was used in the translation
of stressful behavior, in a commercially available monitor called Stress Detector (MedStorm Innovation AS) [106, 108]; it reects the sympathetic nervous system activation,
which causes a release of acethylcholine that acts on muscarine receptors, and causes
bursts of sweat and consequent increase in skin conductance [109].

The developed

monitor displays the number of peaks per second of skin conductance bursts, and the
area under the curve.
Skin conductance has been applied to asses postoperative pain intensity and in response to standardized stimuli, responding adequately to stimulation [110112], however
its use during general anaesthesia has some limitations. In a study conducted in children, the monitor was not able to distinguish evaluations in the NRS, showing low
correlation between the NRS and the number of skin uctuations [113].
The Stress Detector monitor application was also explored by us. Initial usage in
anaesthetized patients did not show promising results, and actually presented technical
problems (unpublished data).

Following the Stress Detector was used in the post-

operative setting, and it demonstrated to correlate in some patients to the reported

pain in a NRS [112, 114].

1.4.1.4 Pupil Diameter

In response to noxious stimulus the pupil becomes dilated, increasing pupil area. This
area may be measured, and pupillometers are widely used for instant objective assessment of pupil diameter. Some researchers [72, 115118] used this approach to evaluate

39

1. INTRODUCTION
the nociceptive response to noxious stimulation and the impact of analgesic drugs to
abolish these reexes.
Barvais et al. [115] explored the changes in pupil diameter and reactivity to noxious stimulation, and observed that increasing remifentanil doses decreased pupillary
responses. In a dierent study [116], the authors explored the pupillary response and
correspondent delay under general anaesthesia, inspecting if the reex was conducted by
slow C bers (increased delay for distant sites of stimulation). The authors concluded
that there was no dierence in the time delays and therefore the reex is not conducted
by slow C bers, observing that higher doses of fentanyl blocked the pupillary reex.
Although pupillary reactions are related to nociceptive responses, this approach has
limited applicability due to its potential for eye damage.

1.4.1.5 Evoked Responses

Most of the proposed methods to monitor Noc/ANoc balance during general anaesthesia, namely all those addressed earlier in this section, are based on passive measures of
physiological responses to noxious stimulation. Understanding the way pain is processed may allow a more in-depth perspective on how to design a tool for the Noc/ANoc
objective assessment. Knowing that pain is sensed and transmitted through nerve impulses to the brain, which vary according to site of stimulation and intensity, promoted
the search for a tool capable of measuring the evoked responses in the central processing
of pain sensing, or in its pathway [119, 120]. One of these methods, called

Evoked Po-

tentials (EP), is broadly used in clinical neurophysiology to determine the state of nerve
conduction for dierent stimulation sources such as visual, auditory and somatosensory.
The somatosensory evoked potentials are of interest for the problem of nociception measurement, and represent a more active approach in pain response measurement, dierent
from the passive methods proposed so far.
As part of this thesis addresses a dierent perspective on nociception measurement,
involving active interference of the system, it is treated separately in Chapter 5

Evoked

Potentials - Active Nociception Measures, were its background, specic state of the art
and clinical protocol designed for data collection in a volunteers' study, are described
in detail.

Most of the developed research in this topic is concerned with the impact

of anaesthetics on the evoked potentials, not for their potential information on the
Noc/ANoc balance, but rather as an interference to the wave used to monitor nerve

40

1.4 State of the Art on the Analysis and Control of the Nociception Level
conduction.

Also, most of the studies suggesting the use of evoked potentials to the

pain assessment are antique, and only few address the topic during general anaesthesia.

As a conclusion, indexes and monitors of the hypnosis component are clinically

widely used, and commercially available. At least ten dierent models of these types
of monitors received FDA or EE approval and are available in the market. In contrast
to this, not only almost none of the proposed methods and indexes for the objective
assessment of Noc/ANoc balance have been made commercially available, but also its
clinical use is currently almost non-existing. Noc/ANoc balance objective measurement
is far from being settled.

1.4.2 Automatic Control

Figure 1.18: Control structure for a complete automatic administration of general anaesthesia, considering its three components: hypnosis, analgesia and muscle paralysis.

Nowadays automation is used in a wide variety of elds, from car assembling to

airplane landing.

In what concerns drug administration, automatic control is almost

non-existing and currently approved only for insulin administration.

The future of

anaesthesia lies in a multivariate control of the three main components, considering

the drugs relations and inter and intra-patient variability, as shown in Figure 1.18
[39, 50, 54, 63].

41

1. INTRODUCTION
Automatic drug administration may improve drug titration and combination, maintaining homeostasis during the intervention and improving patient outcome.

Some

studies [55, 121127] have addressed this issue for the paralysis and hypnosis control,
however there is still not a way of closing the loop for the analgesia component. Knowing
this, the development of a nociception index would bring a way of closing the loop for
the remaining component of general anaesthesia.
A nociception index would most likely lead to a totally independent drug delivery
or advisory system. Combined to a robust control algorithm it would adjust to each
patient the drug doses, according to his/her needs throughout the procedure.
Most or all of the observed physical and chemical phenomena are nonlinear, and
although linear control techniques are well dened, its required applicability conditions
usually can not be met. A good example of nonlinearity is the human body, the drug
interactions and respective nal eect. Due to the adverse consequences that human
error can carry, it is very important to know and estimate the drug processing stages in
the human system. Nonlinear control may be a good approach for the three anaesthesia
components multivariate control, due to their nonlinear course in the human body, and
for being a regulatory control for each component, taking in consideration the known
deviations from the average patient [55, 128].
To maintain the patient in an optimum state, drugs are used in the correct amount
at the right moment, to minimize risks of sub or overdosing. Each case has negative
consequences that need to be minimized for maximum satisfaction and rapid recovery
[13, 40]. The importance of guaranteeing the use of the strictly necessary drug amount
would be translated in a lower incidence of side eects and lesser costs in anaesthesia.
Although for the analgesia level, monitors to assess the nociceptive responses of a
patient are still being developed, attempts have been made using the dierence between
Response and State Entropy, to automatically guide opioid administration [21]; recently
a controller for co-administration of propofol and remifentanil (hypnotic and analgesic)
based on BIS has been proposed and tested in a multi-center study [127].

Although

feasibility of these methods was demonstrated, because the Noc/ANoc assessment presents limited reliability, the control methods present practical limitations and limited
performances in the analgesic administration.
Until a monitor of the Noc/ANoc balance is validated, automatic analgesic administration may be feasible, but with limitations. In fact, considering the model of the

42

1.5 Motivation, Objectives and Thesis Outline

patients' system shown in Figure 1.13, until an adequate measure for the Noc/ANoc balance is available, only the relation between introduced drug and plasma concentration
may be assessed, as presented in Minto model [38, 39], were direct blood samples were
used to assess the real Cp in the patient. Although the

ke0

parameter has been found in

this study, using an EEG derived measure to assess the opioid maximum eect, this may
not adequately translate the maximum drug eect directly on the Noc/ANoc balance,
and even more if one considers that depending on the eect measure used, dierent
delays are associated and should be regarded in the dynamic model adjustment for the
controller design. Further, the relation between the Ce and the measured eect is still
unknown (maybe a sigmoidal relation), both the structure and inter-patient variability
in the adjusted parameters need to be studied in order to produce a full patient model
from the drug dose to the nal measurable eect.
Current research [129] suggests that soon it will be possible to control the three
main drugs for general anaesthesia (see Figure 1.18), allowing a more secure and stable
state for each component. These systems must be robust in the detection of monitoring
errors and unexpected events like hemorrhage or hemodilution [123], and the relation
between drugs (e.g. synergy between analgesic and hypnotics), must also be taken into
account and modelled.

1.5 Motivation, Objectives and Thesis Outline

Leap and the net will appear.
John Burroughs

Anaesthesia is a medical eld that has rapidly evolved in recent years accompanying the
technological developments and the tools they carried. General anaesthesia is a complex
combination of drugs used to induce a state of hypnosis, analgesia and paralysis, and
depends on adequate monitoring to provide to the patient the best care. Hypnosis and
muscle paralysis components may be monitored, since several monitors for these anaesthesia components are available, informing the clinician on patients' state throughout
the procedure, aiding to prevent inadequate dosing and possible awareness, which may
be worsened by conscious pain perception during the clinical procedure. Nevertheless
there is not still an available monitor fully accepted to translate the Noc/ANoc balance
during general anaesthesia.

43

1. INTRODUCTION
Pain is a complex and multifaceted process, necessary for the organism preservation
and survival. For years, the physiological process behind pain sensing has been a theme
of great research interest due to the increasing need of managing pain therapies, and
improving patient welfare. The understanding of pain pathways, central processes and
inhibitory drugs' actions, is fundamental in the search for a tool capable of translating
the Noc/ANoc balance.

Figure 1.19: Nociception/anti-nociception balance index, translating the optimum state

(around 0), the use of excessive analgesic leading to system depression (near -10), and
the use of insucient analgesic for the noxious stimulation (near 10).

Although several studies have already addressed the nociceptive response during
general anaesthesia, there is still no broadly accepted index to translate this anaesthesia
component, nor clinically broadly used. The proposed indexes so far present limitations
in the denition of the adequate state of Noc/ANoc balance. Studies suggested that
a multivariate approach is necessary for a robust nociception index. To this moment
SPI, and Stress Detector are the only developed indexes commercially available. ANSSI
is an open source index, available for individual implementation, and CVI is available
only for clinical studies.
Similar to the process of obtaining a valid hypnosis index, a Noc/ANoc indicator
should be able to dierentiate levels of responsiveness to painful stimulation.

This

process necessarily involves not only the painful stimulus intensity assessment, but also
the impact on the sensory responses of dierent levels of analgesic drug.

44

1.5 Motivation, Objectives and Thesis Outline

Figure 1.20:

Representation of the relation between the stimulus and administered

drugs on a specic patient, producing an amplitude response (patient sensitivity) in the

nociception/anti-nociception balance index.

A possible structure for the conception of a Noc/ANoc balance index is proposed and
shown in Figure 1.19. An index that follows this structure is able to translate the patient
sensitivity to the stimulation, meaning the sensitivity to a stimulus while receiving a
determined anaesthetic drug dose (see Figure 1.20).

One of the major concerns and

challenges in Noc/ANoc measurement is the denition of the adequate analgesia state

(the origin in Figure 1.19), since baseline values vary between patients.
The existence of a Noc/ANoc balance index would allow a more adequate titration
of anaesthetic drugs, with information incoming from dierent monitors, specialized
to translate the dierent anaesthesia components. Similar to what happens with the
hypnosis monitors, a Noc/ANoc index could allow better individual adjustment of drugs'
administration, leading to lower incidence of over or underdosing anaesthetic drugs,
and possible adverse outcomes (arousal episodes with pain experience or chronic postoperative pain e.g.).
The

objectives of this thesis are:

to explore passive and active measures linked to the Noc/ANoc balance during
general anaesthesia;

to develop an index capable of translating the patient's Noc/ANoc balance state

during general anaesthesia;

45

1. INTRODUCTION

to implement the Noc/ANoc developed index in a system to aid the anaesthesiologist control the analgesic administration and, in conjunction with the information
provided by the monitors present in the operating room, maintain patient's homeostasis.

This thesis is organized as follows: Chapter 2 to 4 present the study of passive nociception measures in anaesthetized patients during general anaesthesia, Chapter 5 is
focused on active measures of nociception on anaesthetized volunteers using somatosensory evoked potentials, Chapter 6 discusses the control in anaesthesia, and Chapter 7
presents the thesis global conclusions.

46

Chapter 2
Data Collection and Pre-Processing

Errors using inadequate data are much less

than those using no data at all.
Charles Babbage

This chapter is focused on a clinical protocol design to collect data under general anaesthesia. It also includes the submission for institutional and ethics committee approval,
informed consent management, and data collection. The data were collected in the urology operating room of the Hospital de Santo Antnio (Centro Hospitalar do Porto),
located in Porto-Portugal, following approval and written informed consent. The structure of the data les collected is here presented, as well as the pre-processing techniques
applied to the wave data les collected, in order to extract the information of interest.

2.1 Introduction
Responses to noxious stimulation are widely described in the literature, expressed in
dierent physiological signals such as heart rate, blood pressure, skin conductance,
electromyography, pupil diameter, pulse wave and evoked responses. In order to fulll
the thesis objectives, a clinical protocol was designed to collect as much usable data as
possible, and exploit the relations between the physiological signals, in the search for
a tool to translate the clinical Noc/ANoc balance of the patient. The present chapter

47

2. DATA COLLECTION AND PRE-PROCESSING

is focused on protocol design, data acquisition, les' structure, and data pre-processing
for later developments.
The designed clinical protocol was submitted for Santo Antnio Hospital's approval
(Centro Hospitalar do Porto), both by the Administration Council and Ethics Committee; after approval, data collection was conducted by the author, who was present
throughout the anaesthetic and surgical procedures, gathering all information considered relevant to the study.
The study was divided into two distinct phases: the rst phase in which autonomic
responses were evaluated in precise moments of the anaesthetic or surgical procedures
(phase I), and a second phase during the surgical procedure for dierent drugs' levels,
and stimuli intensities (phase II).
Data collected from the monitoring devices, and annotations registered by the
author, must be pre-processed for subsequent analysis of relations between surgical stimulus, drug doses and physiological responses. Data pre-processing techniques includes
data re-sampling, ltering, and waves feature detection, so that important information
contained in the data is revealed through the employment of adequate techniques.
Annotations gathered by the author throughout the clinical procedures were transformed into a stimulus scale. This scale, initially with no clinical meaning, was later
post-processed to translate noxious stimuli intensity.
In this study a BIS Bilateral monitor was used, with the inclusion of the experimental
CVI index from both left and right channels, in a research collaboration with Aspect
Medical Systems, who provided the adequate equipment and BIS bilateral sensors.
Patients who participated in the study were aware of this collaboration and agreed
to share the data with the company (see written informed consent in Appendix B),
with the guarantee of privacy and condentiality preservation. There were no monetary
compensations in this research collaboration.
In the following sections, data collection procedures and pre-processing is described
in detail.

2.2 Clinical Protocol Design

2.2.1 Design Goals
The clinical protocol was designed to:

48

2.2 Clinical Protocol Design

evaluate autonomic responses to precise noxious stimuli during intravenous general

anaesthesia, for dierent analgesic levels;

evaluate the eect of changes in surgical stimulation and drug doses on the physiological signals during the intervention;

evaluate metrics proposed in the literature, and propose an adequate measure for
the Noc/ANoc balance during general anaesthesia;

use the gathered information in the development/implementation of a clinical

index translating the patient's Noc/ANoc balance state.

Figure 2.1:

Representation of the dened study groups, according to the eect-site

concentration of the analgesic (remifentanil) using Minto pharmacokinetic and dynamic

model [39].

2.2.2 Selection Criteria

Patients subjected to scheduled urological surgery under general anaesthesia, adults,
physical status I-III in the classication scale of the

gists

American Society of Anaesthesiolo-

(ASA), lled the requirements for enrollment. Consecutive patients scheduled for

surgery that fullled the inclusion criteria, and accepted to participate in the study by
signing the informed consent presented in Appendix B, were included. Informed consent
was obtained by one of the anaesthesiologists who participated in the study during the
pre-anaesthetic visit.
The set of enrolled patients was randomly divided in three groups (see Figure 2.1),
dened according to the theoretical eect-site concentration of the analgesic remifentanil

49

2. DATA COLLECTION AND PRE-PROCESSING

(Minto PKPD model [39]) used for induction and phase I of the study. Each patient
was identied with a generated code, preserving privacy and condentiality. Exclusion
criteria were body weight 30% above or lower the average recommended weight, serious
cardiovascular or respiratory disease, neurological disturbs, or counter-indication to the
anaesthetic technique.
For each patient a folder was created to register patient data (gender, age, weight,
height, clinical background), and store les during the procedures; this folder was identied with the generated code.

2.2.3 Pre-Medication
Patients were pre-medicated with 10 mg of diazepam, at 22h on the previous day, and
at 7 h on the day of surgery. In the operating room, following patient monitoring, 2 mg
of midazolam were administered intravenously.

2.2.4 Clinical Setup

R

All patients were monitored according to hospital's protocols, using the General Electric
monitor and ventilator equipping the operating room.

Patient's physiological signals

were continuously registered through the synchronization and data acquisition software
RugloopII

c Waves (Demed). Invasive blood pressure monitoring was set up either at

the beginning or following anaesthesia induction.

The material used in data collection is listed below:

R

- monitoring and automatic ventilation;

General Electric Aisys

ASPECT BIS Vista

BIS bilateral sensors;

Propofol 1% Lipuro

Remifentanil Ultiva

Fresenius-Kabi Orchestra

Perfusion pump, B.Braun

Computer with RugloopII

T M monitor;

R
R
R



/Fresenius
in B.Braun
Perfusor 50 ml syringe;

R
R


in B.Braun
Perfusor 50 ml syringe;
R

Base Primea pump (2 infusion pumps);
R

Infusomat;

c Waves (data collection and synchronization software)

installed and fully functional.

50

2.2 Clinical Protocol Design

R

Anaesthetic drugs' administration was conducted using the TCI system Orchestra

Base Primea. Remifentanil (C=20 g/ml) administration was conducted using Minto
[38, 39] PKPD model and propofol (C=10 mg/ml) was administered using Schnider
PKPD model [50]. All stimuli and events considered important for the study were annotated and registered in RugloopII

c (e.g. loss of response to verbal and mechanical

stimulus, intubation, incision, electric cauterization, drugs' administration, suture, extubation). After data collection, the les recorded were extracted to spreadsheets and

c (Demed).

text les using Labgrab2.03

2.2.5 Methods
Anaesthetic induction was conducted similarly in all patients using TCI of propofol
and remifentanil.

The study was centered in haemodynamic and hypnosis EEG ba-

sed indexes changes, in the periods preceding and following precise noxious stimuli:
laringoscopy/intubation, tetanic stimulus and incision.
In the determination of pattern responses exclusively due to the noxious stimulus,
the patient was kept with no external interference, for one minute prior and two minutes
after stimulation, with the same doses of propofol and remifentanil (phase I).
Besides pattern responses to precise noxious stimuli, data referring to dierent combinations of drugs were obtained to analyze the relations between drugs' doses, noxious
stimulation and nociceptive activation. For that purpose, during the surgical procedure,
dierent drugs' combinations were dened and kept constant for 5 minutes intervals, to
obtain both dynamic and steady-state information (phase II).
Following, the designed clinical protocol for both phases of the study is described
step-by-step.
Table 2.1: Study interruption rules: for each case the anaesthesiologist should redene
drugs' theoretical eect-site concentration targets, according to patient's needs.

Study Interruption Rules

BIS>60 or BIS<40
HR>1.2 HRbaseline or HR<0.8 HRbaseline
SBP>1.2 SBPbaseline or SBP<0.8 SBPbaseline
Movement

51

2. DATA COLLECTION AND PRE-PROCESSING

PHASE I:
Study physiological responses to precise noxious stimuli, of equivalent intensity
between patients, for dierent analgesic doses.

1. Start monitoring noninvasive arterial pressure in continuous mode. Register

art Rate

(HR) and

Systolic Blood Pressure

He-

(SBP) baseline values (HRbaseline ,

SBPbaseline );

2. Start saline at 400 ml/h and remifentanil infusions, dening the eect-site concentration of remifentanil according to the study group attributed in the randomization (2, 3 or 4 ng/ml, as shown in Figure 2.1);

3. After the eect-site concentration of remifentanil reaches the dened theoretical

target, start administration of propofol 1% at a slow rate of 200 ml/h until loss
of response to verbal and mechanical stimulus (

Loss Of Consciousness, LOC). At

this point the propofol eect-site concentration observed is registered (induction

dose);

4. After LOC propofol infusion is administered by TCI, according to patient's needs;

5. Administer muscle relaxant for intubation, and preferentially only re-administer

after the second phase of the study starts;

6. When the eect-site of propofol reaches the target, patient should be maintained
with no external interference;

7. One minute after the eect-site concentration of propofol dened after induction
is stable, start laringoscopy; after intubation keep the patient with no stimulus
for two minutes;

8. During the period of preparation for the surgical procedure, change drugs' eectsite concentration, according to patient's needs;

9. Before surgery start, set eect-site concentration of remifentanil to the dose used
during intubation, and one minute after it reaches the theoretical target, apply a
tetanic stimulus (50 Hz for 5 s, and post-tetanic count); keep the patient with no
external interference for 2 minutes;

52

2.2 Clinical Protocol Design

10. In case of occurrence of any of the responses described in Table 2.1, increment the
eect-site concentration of remifentanil in 0.5 ng/ml and repeat tetanic stimulation. This step should be repeated until loss of the listed responses;

11. Start incision.

PHASE II:
Study responses to noxious stimuli during surgery, for dierent ascending and descending analgesic steps.

1. During surgical procedure drugs' eect-site concentration should be redened

according to patient's needs.

Ten minutes after the beginning of the surgical

procedure, register baseline heart rate and systolic blood pressure (HRbaseline ,
SBPbaseline );

2. In the case of remifentanil increment after surgery start, the study should be
initiated with ascending remifentanil steps, otherwise start second phase of the
study with descending remifentanil steps;

(a) Increase/decrease the eect-site concentration target in 0.5 ng/ml, adjusting

the hypnotic eect-site concentration according to BIS monitored values (target of 40 to 60, as recommended by the manufacturer);
(b) Keep remifentanil target stable for 5 minutes following equilibration of the
plasma and eect-site theoretical concentrations, and register surgery stage;
(c) Repeat steps a-b until one of the scenarios described in Table 2.1 occurs, or
the minimum dose allowed in the study of 1.5 ng/ml is reached; in that case
redene hypnotic and analgesic targets according to patient's needs;
(d) Start descending/ascending remifentanil 0.5 ng/ml steps, keeping the hypnotic level according to BIS monitored values;
(e) Keep remifentanil target stable for 5 minutes, and register surgery stage.
(f ) Repeat steps d-e until one of the scenarios described in Table 2.1 occurs, or
the minimum dose allowed in the study of 1.5 ng/ml is reached; in that case
redene hypnotic and analgesic targets according to patient's needs.

53

2. DATA COLLECTION AND PRE-PROCESSING

Remifentanil doses used in the study were dened according to recommended dose
ranges for induction and maintenance of general anaesthesia in co-administration with
propofol, as demonstrated in Appendix C.
Propofol administration was guided and adjusted throughout the study to maintain
BIS values within the manufacturer recommended target range of 40 to 60.
Following the intra-operative study described above, a simultaneous study was conducted in the post-operative period, for the analysis of the post-operative acute pain.
Pain was evaluated using patient's collaboration and his/her evaluation in a NRS, and
also using skin conductance monitoring (peaks/s and area under the curve) with the

T M ). The study received institutional

Stress Detector monitor (Medstorm Innovations

approval and the results presented to the scientic community [114], and published [112].
Although the study was conducted in full collaboration with the author, the results are
not in the scope of this thesis (study focused on the analysis of the post-operative acute
pain), and therefore will not be here presented.

2.3 Institutional Approval

In order to proceed with data collection, after the clinical protocol design, the necessary
documentation to obtain institutional approval was gathered. The institutional procedures for clinical studies approval include the denition of the main researcher (the
author), the denition of the responsible researcher in the hospital, obtaining approvals
from the departments and services involved in the study development, and obtaining
approvals from the Research Department and Ethics Committee. Figure 2.2 shows a
ow diagram of the institutional procedures followed to obtain a clinical study approval.

a 027/09(019-DEFI/025-CES)) data col-

Following institutional approval (N/REF.

lection was initiated.

2.4 Data Overview

Data from 34 patients, under general anaesthesia were collected, after obtaining written informed consent. Data presented as meanstandard-deviation. Table 2.2 shows
patients' demographics in the three study groups, with no statistical dierence found
between groups considering average age, weight, height and gender distribution (Lilliefors, Kolmogorov-Smirnov, Fisher's tests,

P < 0.05).

54

The les collected were extracted

55

studies.

Figure 2.2: Flow diagram of the hospital's required procedures to obtain institutional approval for the realization of clinical

2.4 Data Overview

2. DATA COLLECTION AND PRE-PROCESSING

Table 2.2: Patients' demographics according to the division in the three study groups:
no statistical dierence between groups, with

P < 0.05

(data as meanstandard-

deviation).

Global Group 1 Group 2 Group 3

Gender (M/F)

Figure 2.3:
Orchestra

18/13

5/5

8/3

5/5

Age (years)

54.913.4

58,215,4

50,813,8

56,110,6

Weight (kg)

69.912.3

66,010,9

74,914,4

68,510,3

Height (cm)

165.07.6

161,92,6

168,49,7

164,47,5

Clinical setup used in data collection:

A) BIS VISTA

T M monitor; B)

R

Base Primea syringe pumps in TCI mode (propofol and remifentainl); C)
R

monitor and ventilator; D) Computer used in data collection

General Electric Aisys

with Rugloop

c Waves installed.

using Labgrab

c to Excel les, and later exported to Matlab

R

.

Of the 34 collected

cases, 3 were discarded due to technical failure or protocol deviations.

56

2.4 Data Overview

Figure 2.4: Monitored patient during a standard data collection with a BIS bilateral
sensor, on the left, and the BIS monitor with corresponding BIS and CVI trends for the
left channel, on the right.

In the nal data set, 16 patients had an epidural catheter placed prior to anaesthesia
induction for post-operative analgesia only, and the moment of post-operative analgesia
administration registered. Type of surgery varied, with incidence of 13 prostatectomies,
8 nephrectomies, 5 nephrolithotomies, 2 cystectomies and 3 smaller procedures.
Collected data from the monitors, using RugloopII, were sampled at a rate of 1 Hz.
Nevertheless, monitors export data to the synchronization software in dierent sampling rates: Datex monitor every 5 s, infusion pumps every 2s and BIS every 1s. For
lower sampling rates the synchronization software digitizes the signal at constant steps.
Waves data les were also sampled at dierent sampling rates, depending on signal
characteristics. Due to the presented dierences, les' structure was carefully dened
for subsequent analysis.
Figure 2.3 shows the clinical setup assembled in every data collection, with the

R
R

T M monitor, Orchestra
, the BIS

physiological variables monitor and ventilator Aisys

syringe pumps in TCI mode, and the computer used for data collection and synchronization, with the Rugloop

c Waves system installed and fully operational. Figure

2.4 shows a patient during data collection, with the BIS bilateral sensor placed on the
forehead (left), and corresponding BIS monitor and collected trends (right).

2.4.1 Files' Structure

All data were exported to Matlab

R

for o-line analysis.

Table 2.3 shows the list of

variables analyzed in the study. Data collected at 1 Hz containing information from the

57

2. DATA COLLECTION AND PRE-PROCESSING

Table 2.3: Collected variables (47 in total) at a sampling rate of 1 Hz, exported to *.mat
les for each data set.

Datex

BIS Bilateral and CVI

TCI Data

1. Time

16. Suppression Rate L (%)

37. Infused Volume (ml)

2. ECG Heart Rate

17. Spectral Edge

38. Remifentanil Cp

1 )
(beats/min

Frequency L (Hz)

(ng/ml)

3. BP Systolic (mmHg)

18. BIS L

39. Remifentanil Ce
(ng/ml)

4. BP Diastolic (mmHg)

19. Total Power L (dB)

40. Remifentanil Ct
(ng/ml)

5. BP Mean (mmHg)

20. EMG L (dB)

41. Infusion Rate

(ml/h)

6. BP Heart Rate

21. SQI L (%)

42. Infused Volume (ml)

22. ASYM (rw)

43. Propofol Cp

1 )
(beats/min
7. Non-Invasive BP

(g/ml)

Systolic (mmHg)
23. SD BIS L

44. Propofol Ce (g/ml)

9. Non-Invasive BP Mean

24. SD EMG L

45. Propofol Ct (g/ml)

o
10. Temperature ( C)

25. CVI L

46. Infusion Rate

8. Non-Invasive BP
Diastolic (mmHg)

(ml/h)
11. SPO2 (%)

26. Impedance L ()

47. SPO2 Amplitude

(%)

12. SPO2 Pulse Rate

27. Supression Rate R (%)

1 )
(beats/min
13. CO2 Et (%)

28. Spectral Edge

Frequency R (Hz)

14. CO2 Fi (%)

29. BIS R

15. CO2 Respiration

30. Total Power R (dB)

1 )
Rate (cycles/min
31. EMG R (dB)
32. SQI R (%)
33. SD BIS R
34. SD EMG R
35. CVI R
36. Impedance R ()

58

2.4 Data Overview

c , were

infusion pumps, Datex and BIS monitors, exported to Excel les using Labgrab
later imported to Matlab

R

using ve structure variables as follows:

datex = struct('time', {}, 'ecghr', {},'bps', {},'bpd', {},'bpm', {},'bphr', {},'nibps',

{},'nibpd', {},'nibpm', {},'temperature', {},'spo2', {},'spo2pr', {},'co2et', {},'co2',
{},'co2rr', {})
bisl = struct('sr', {},'sef ', {},'bis', {},'totpow', {},'emg', {},'sqi', {},'sdbis', {},'sdemg',
{},'cvi', {},'impedance', {})
bisr = struct('sr', {},'sef ', {},'bis', {},'totpow', {},'emg', {},'sqi', {},'sdbis', {},'sdemg',
{},'cvi', {},'impedance', {})
prop = struct('infvol', {},'cp', {},'ce', {},'ct', {},'infrate', {})
remi = struct('infvol', {},'cp', {},'ce', {},'ct', {},'infrate', {})

Table 2.4: Collected waves, number of channels, sampling rate, and data units.

Signal

Number of Channels Unit

EEG

4 (eeg0, eeg1, eeg2, eeg3)

ECG

1 (ecg)

PPG

1 (pleth)

IBP

1 (invp)

CO2

1 (co2)

V
V

128

100

mmHg

100

300

The collected waves' data les - EEG, ECG, PPG,

Carbon Dioxide

Sampling Rate (Hz)

25

Invasive Blood Pressure

(CO2 ) - were extracted to ASCII les using Labgrab

(IBP),

c , and later im

R

. The collected waves included four EEG channels (ch0, 1, 2 and 3),

ported to Matlab

ECG, CO2 , PPG and IBP, all with dierent sampling rates as shown in Table 2.4. Each
wave le is composed by two columns: the rst with the sampling time, synchronized
with the le sampled at 1 Hz; the second with the collected physiological signal wave
values.
For each patient

i in the study data were saved to a *.mat le pati , where i = 1, ..., 34,

containing the ve dened structures and the eight wave les.
To facilitate data inspection, a

Graphical User Interface

(GUI) tool was developed

R

using Matlab
to visualize and assess collected data. This tool allows the synchronized
navigation through the les, with a zoom feature for more detailed analysis of each
signal and event (see Figure 2.5).

59

2. DATA COLLECTION AND PRE-PROCESSING

Table 2.5: List of noxious stimuli annotated by the author during the procedure, and
corresponding code number attributed to the stimulus variable

scale.

Noxious Stimulus

Code

No stimulus
Ureteroscopy/Cystoscopy
Retractors placement and Introduction of laparoscopic surgery instruments
Retractors placement in abdominal surgery
Tetanic stimulus
Lesion extraction in laparoscopic surgery
Extubation
Skin incision for nephrectomy
Abdominal incision
Percutaneous nephrostomy incision
Skin incision for prostatectomy
Inate abdominal cavity for laparoscopy
Intubation
Laringoscopy
Tissue manipulation open surgery
Tissue manipulation cystectomy
Tissue manipulation laparoscopy
Tissue manipulation nephrectomy
Tissue manipulation prostatectomy
Skin incision for drain or Small laparoscopic incision
Periods in the surgery without surgical stimulus (any break in surgery)
Wall suture/Skin staples
Skin suture
Abdominal cavity tissues traction
Gastric drainage tube/Temperature probe nasal/oral
Vein puncture/Arterial puncture/Jugular or sub-clavian puncture
Folley catheter placement
Laparoscopy nalization
Intraoperative lithotripsy with ultrasound
Laringoscopy for oro-pharyngeal throat-packing
Visceral tissue manipulation
Nasal broscopy for intubation
Skin puncture and epidural catheter placement under general anaesthesia without
local anaesthesia
Ganglia removal in the inguinal region
Internal sutures (muscle or viscera)

60

0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

2.4 Data Overview

Figure 2.5: Software developed to analyze and navigate o-line through the collected
data for each patient (pati , where

i = 1, ..., 34).

Events registered throughout anaesthesia, annotated by the author in a text le,

were also analyzed to determine important events, and changes in stimulation.

As

referred before, the study was divided in two phases: the rst phase focused on the
analysis of punctual events (laringoscopy, tetanic stimulus and incision), and a second
phase during the maintenance in which changes in the drugs' doses and stimulus intensity were to be related to changes in the physiological signals. To analyze changes
in stimulus two variables were included in each patient le, the rst being an array
containing the time points of precise events in the procedure, such as loss of response
to verbal and mechanical stimulus (LOC),

Recovery Of Consciousness

(ROC) and in-

tubation (times variable), and a second variable (scale), which translated by a code
number the dierent stimuli annotated in the events' text le throughout the whole
case. Table 2.5 shows the code numbers attributed to each stimulus annotated by the
author during data collection, and later introduced in the
array was composed by the following events in time:

61

scale

variable. The

times

2. DATA COLLECTION AND PRE-PROCESSING

times = [Start IBP Monitorig - LOC - Laringoscopy - Intubation - Tetanic Stimulus
- Incision - Start 2nd Phase - Post-Operative Analgesia - ROC - Extubation]

2.5 Waves Pre-Processing

Collected waves' data les contain important information on the Noc/ANoc balance
assessment, and may allow an evaluation of the compensation mechanisms between heart rate and blood pressure, for example, taking into account other variables such as
pulse wave amplitude, and respiration rate. To perform this analysis it is necessary to
analyze the data in synchronization, and obtain useful information contained in each
signal. From the ECG one wants to extract each beat-to-beat interval (heart rate related); from the IBP, for each heart beat, wave analysis allows quantication of systolic,
diastolic and mean invasive arterial pressure; from the PPG, wave amplitude may be
extracted, containing information on vasodilation/constriction state of the patient [130];
and nally, from the CO2 wave, one may extract the respiration rate.
For each wave collected in the study, information of interest must be extracted from
the raw data, synchronized between signals and re-sampled at an uniform sampling
rate. Following waves' pre-processing is described; each signal was treated separately,
synchronized, and re-sampled at the same sampling rate (1 Hz).

2.5.1 Electrocardiogram and Beat-to-Beat Interval

ECG signals and the information contained in them is topic of great interest in the
scientic community [131, 132].

The collected physiological signals are vulnerable to

noise disturbances, and to proceed with the analysis waves must be pre-ltered, and
external interferences such as EMG, or electrical power line, removed.

One of the

problems in ECG signals processing is the existence of baseline drift, which should
be removed minimizing interference in heart beat morphology. Baseline drifts present
in ECG signals are usually below 0.5 Hz, nevertheless movement artifacts may lead to
increased frequency bounds. The two techniques usually applied in these cases are linear
lters and polynomial tting. In the collected data a linear lter was applied [131].
The high-pass lter designed must allow removal of all low-frequency components,
keeping the most of the signal clinical information. For the ECG, during a bradycardia
episode heart rate may drop to 40 beats/min, making the lowest frequency of the signal
to drop to 0.67 Hz. Nevertheless, heart rate is irregular, and its best to choose a cut-o

62

2.5 Waves Pre-Processing

frequency slightly lower. Another important issue is the phase response characteristics
of the lter. The objective is to obtain a linear phase lter, so that signal characteristics
such as QRS complex duration are not distorted.

Innite Impulse Response (IIR) lters

allow optimum magnitude response lters, however with non-linear phase response.
Nevertheless if the lter is applied in the two directions, a zero-phase transfer function
lter is obtained. Since in this case data were processed o-line, this was the method
applied. Let us consider

h(n)

x(n)

the original signal,

s(n)

the output ltered signal, and

the IIR lter [131], the output signal is given by the sequence of Equations 2.1,

2.2 and 2.3.

z1 (n) = h(n) x(n)

(2.1)

z2 (n) = h(n) z1 (n)

(2.2)

s(n) = z2 (n)

(2.3)

Taking the discrete Fourier transform of the signal, Equation 2.4 is obtained.

X(ejw ) =

x(n)ejwn

(2.4)

n
with the property in Equation 2.5.

x(n) X (eejw )

(2.5)

Combining the two origins Equation 2.8.

S(ejw ) = Z2 (ejw ) = H (ejw )Z1 (ejw )

(2.6)

= H (ejw )H(ejw )X(ejw )

(2.7)

= |H(ejw )|2 X(ejw )

(2.8)

The lter used was an IIR Butterworth of order 3, applied in both directions: cuto
frequency of 0.5 Hz (30 beats/min), and order 6.
After removing baseline wander, the next step is to detect the QRS complex and
extract the RR intervals. A linear lter given by Equation 2.9 was applied to the signal
to highlight the signal features of interest.

H(z) = (1 z L1 )(1 + z 1 )L2

63

(2.9)

2. DATA COLLECTION AND PRE-PROCESSING

From the ltered data, QRS peaks may be detected, and the sequences of RR
intervals obtained. Several methods use detection rules, such as the use of a threshold
above which a QRS complex is detected; this simple method is not the best, since
it will detect missing beats, or even T-waves of greater amplitude. Thus, it becomes
imperative to use methods more appropriate and adaptive in order to obtain only the
QRS complexes of interest. A proposed method to overcome this issue is an interval
depending threshold [131]. This method sets the threshold of the last detected complex
and considers the rules in Equations 2.10 and 2.11 for the detection threshold denition.

where

(2.10)

ze,i = ze,i1 + (z(i ) ze,i1 ), i 1

(2.11)

is a detected complex,

complexes and

nI =
ze,i , n = i , i + 1, ...

z(i )

ze,i

is the exponential average of the previously detected

the amplitude of the signal of the most recently detected complex.

represents the fraction used in the threshold determination, usually

and

0.5 0.7

the rate at which the threshold may change.

This method may be expanded to a time-dependent threshold, improving erroneous

detection of high amplitude T waves, with an eye-closing period. This period is usually
between 160-200 ms, which is the time of absolute refraction during which the heart
does not respond to electrical stimulation. Nevertheless, this period may be too long,
missing ectopic beats. To overcome this issue a compromise must be reached between
these two. The best option is to keep a waiting time (silence period)
no beat is detected, and a

D1

D0,

during which

period during which ectopic beats may be detected with

linearly descending threshold values [131], as given by Equation 2.12.

n = i + 1, ..., i + D0
nmax
g(n i D0 1) n = i + D0 + 1, ..., i + D1
n(n) =

ze,i
n = i + D1 + 1, ...

(2.12)

This method was applied to the RR intervals detection in the collected data. Figure
2.6 shows the results of the RR detection algorithm in one of the collected data les.
Upon detecting the QRS complexes, RR intervals sequence must be extracted. The
RR interval was dened at the moment of QRS detection (tRi ) as the time dierence
between two consecutive QRS complexes (tRi

tRi1 ).

Following RR intervals sequence

extraction, the HR sequence is easily obtained by Equation 2.13.

64

2.5 Waves Pre-Processing

Figure 2.6: Original ECG signal (black line), ltered ECG signal (QRS complex enhancing, gray line), and correspondent QRS complex detection (black dot).

Figure 2.7: Original RR time intervals (black line), and interpolated signal re-sampled
at 1 Hz: using hermitian cubic spline interpolation (gray line), and cubic spline interpolation (dashed line).

HR(tRi ) =

tRi

60
tRi1

(beats/min)

This sequence is however unevenly spaced in time due to its nature.

(2.13)

To obtain

equally spaced in time sequences, data was interpolated using a cubic Hermite spline,
maintaining the original signal trend (see Figure 2.7), and re-sampled at 1 Hz.

65

2. DATA COLLECTION AND PRE-PROCESSING

Cubic Hermite splines t a order 3 polynomial to the unevenly spaced in time
sequence of RR intervals, similar to the cubic spline but without the restriction of
producing

C2

functions. Although cubic splines produce more smooth functions, since

the second order derivative is continuous in time, the cubic Hermite spline has no
overshoot and less oscillations, which is more adequate for the current application [133,
134].

Figure 2.8:

Original RR sequence (black line) obtained after interpolation at 1 Hz,

correspondent ltered signal after outlier removal (gray line), and RR sequence provided
by the Aisys monitor through the heart rate sequence (dashed line). Data from one of
the patients in the study.

Following QRS peaks detection, RR intervals extraction and interpolation, the obtained RR sequence was still contaminated with outliers due to poor quality in the
ECG data acquisition. To proceed with the analysis, the extracted sequence was postprocessed for outlier detection and removal. To obtain clean RR sequences the originally extracted RR sequence was smoothed by a smoothing lter using the robust Loess
method [135], which uses a local regression with linear least squares, and a 2nd degree polynomial model, assigning lower weights to outlier values in the regression. This
method allows the construction of a smoother version of the original signal, diminishing
the impact of outliers. To detect an outlier a comparison between the original and smoothed signal was performed: if the dierence between the two sequences was above two
standard-deviations of the median error observed in the population data, an outlier was
identied and substituted by the corresponding point in the smoothed signal version.

66

2.5 Waves Pre-Processing

Figure 2.8 shows the results of the outliers removal and replacement, with the originally
extracted RR sequence, the ltered RR sequence and the RR sequence given directly
through the Datex monitor (obtained from the heart rate sequence, communication
every 5 s with 1 Hz sampling rate).
The methods described for the ECG wave and RR interval extraction were applied,
with the adequate modications, to the remaining waves since they are, in the majority,
heart beat related. All sequences were interpolated using Hermite cubic splines and resampled at 1 Hz, to obtain signals equally spaced and synchronized in time.

2.5.2 Invasive Blood Pressure

Figure 2.9: Invasive blood pressure signal with overlapping systolic pressure peaks detected (black dot). Data from one of the patients in the study.

For the IBP wave, information regarding systolic, diastolic and mean arterial pressure may be extracted. Due to signal physiological characteristics, systolic blood pressure is the parameter more closely related to noxious activation, and compensation
mechanisms. The methods previously described were applied to obtain blood pressure
peaks, and systolic blood pressure measured in the original signal. In this case, baseline
wander is important information contained in the physiological signal. Figure 2.9 shows
the invasive blood pressure wave, and the detected peaks later used to interpolate the
systolic blood pressure sequence.

67

2. DATA COLLECTION AND PRE-PROCESSING

2.5.3 Photoplethysmography and Wave Amplitude

Figure 2.10: Photoplethysmography wave and correspondent wave amplitude denition

(PPGA): A is the local minimum at the beginning of the pulse, and B is the local
maximum. PPGA is dened as the dierence between the two.

For the PPG wave, the objective is to obtain the wave amplitude, which has been
related to changes in vasoconstriction and dilation, as a cardiovascular response to
noxious stimulation [78, 86, 130]. Wave amplitude is the height of the pulse, usually
obtained as the amplitude dierence between the local minimum at the beginning of
the pulse (A), and the pulse wave maximum amplitude (B) as depicted in Figure 2.10
[130].
Beat-to-beat PPG wave amplitude (PPGA) was extracted considering the consecutively registered wave values at points A (tAi ) and B (tBi ), given by Equation 2.14, and
illustrated in Figure 2.11.

P P GA(tBi ) = P P G(tBi ) P P G(tAi )

(2.14)

The obtained PPGA sequence was then interpolated and re-sampled at 1 Hz.

2.5.4 Respiration Rate

When evaluating cardiovascular compensation mechanisms, respiration plays an important role, inducing changes both in heart rate and blood pressure. Although patients
were mechanically ventilated throughout the procedure following intubation, arousal
episodes may be accompanied and detected by the observation of spontaneous ventilation.

Respiration Rate

(RespR) was extracted from the CO2 wave, with a simplied

68

2.5 Waves Pre-Processing

Figure 2.11:

Photoplethysmography wave and detected points for amplitude assess-

ment (A, white dot, dened as the local minimum and B, black dot, as the wave local
maximum). Data from one of the patients in the study.

Figure 2.12: CO2 wave and detected point of maximum CO2 for each respiration cycle,
corresponding to the end of an expiratory cycle. Data from one of the patients in the
study.

method adapted from the previously described, in which the moment of maximum CO2
value was detected and considered to be the end of an expiration cycle (tExpi ).
respiration rate sequence was given by Equation 2.15.

69

The

2. DATA COLLECTION AND PRE-PROCESSING

RespR(tExpi ) =

tExpi

60
tExpi1

(cycles/min)

(2.15)

Figure 2.12 shows the CO2 collected wave with the expiration detected points used
to compute RespR in cycles/min, and re-sampled at 1 Hz using cubic Hermite splines.

2.6 Summary
A clinical protocol to evaluate nociceptive responses to changes in stimulation intensity,
and drug doses was designed and submitted for approval to the Hospital de Santo
Antnio Ethics Committee and Administration Council.
Following institutional approval, and patients' written informed consent, 34 patients
scheduled for urological surgical procedures under general anaesthesia were enrolled in
the study. Of the collected data, 3 cases were excluded due to protocol deviations or
technical diculties.
Following data collection, the data were pre-processed and exported to an unique

R

le. This le contained synchronized data from all monitors used in the study.

Matlab

R

in order to ex-

Waves collected during the procedures were processed in Matlab

tract the information contained in them, synchronized and re-sampled for the following
analysis.

70

Chapter 3
Stimulus Intensity Analysis

Measure what is measurable

and make measurable what is not so.
Galileo Galilei

This chapter addresses the construction of a measure of noxious stimuli intensity.

survey designed and disseminated in the Portuguese anaesthesiologists community (Sociedade Portuguesa de Anestesia and Servio de Anestesiologia do Hospital de Santo
Antnio), to evaluate clinical perception of the noxious intensity of a list of stimuli that
the patient may be subjected to during general anaesthesia, is reported.

Moreover,

a scale based on clinical perception constructed using Rasch analysis is proposed for
later application in the search for relations between drugs' doses, stimuli intensity and
amplitude response of the physiological variables related to noxious activation [136].

3.1 Introduction
During general anaesthesia patients are subjected to various sources of noxious stimuli originating from anaesthetic and surgical procedures. Anaesthesiologists rely on
their experience to recognize stimulation timing and intensity, anticipating stimulus
occurrence, and administering analgesics to control physiological responses to noxious
simulation.

71

3. STIMULUS INTENSITY ANALYSIS

There are already available monitors to aid anaesthesiologists in the hypnotic [8]
and muscle relaxant [137] drug management, for the analgesia component there is not
yet a fully accepted tool for the evaluation of the Noc/ANoc balance during general
anaesthesia, although there are already proposed indexes, some of them commercially
available; nevertheless, none seems to be the ideal method, and none includes in the
designation the notion of nociception [78, 79, 8587, 130, 138], thus the development of
new indicators continues to be a research eld of interest.
In the design of indexes that reect the Noc/ANoc balance there are three aspects to
consider: the physiological variables measured objectively, the anti-nociception provided
by anaesthetic drugs, and the intensity of the noxious stimulation (see Figure 1.20). The
latter is dicult to quantify and is the subject of this study.
Patients' response to anaesthetic drugs and noxious stimulation is known to vary,
and when developing a tool capable of recognizing pattern physiological responses to
noxious stimulation, it is important to understand the existing balance between the
administered analgesic, patient's characteristics, intensity of the noxious stimulus, and
interference of other drugs and events during the procedure. All these factors combined
interact to produce responses in the patient which may be interpreted and translated,
within context, as an objective measure of the Noc/ANoc balance state. To recognize
the variables and existing relations one must be able to quantify all inputs, in the case
anaesthetic drugs' doses and stimulus intensity. This is in fact what anaesthesiologists
have been doing for years, and any tool developed to assess the analgesia component
must have this into account, reecting changes in stimulation, and changes in the drugs'
levels.
The methods proposed so far to assess noxious stimuli intensity, in the search for
a Noc/ANoc indexes, use estimates of stimulation intensity assessed by few experts'
opinions [77], or estimation of stimulus intensity based on pharmacological studies, in
which drugs' needs to abolish responses to noxious stimuli are used to estimate noxious
stimulus intensity [139]. Both approaches present limitations, the rst by the use of few
expert opinions, which may be prone to bias, and the second limited by the availability
of data on drug needs to assess each stimulus intensity (available data only for a small
list of stimuli).
The objective of this study was to take advantage of the clinical experience of a
wide number of anaesthesiologists, using their clinical experience and knowledge of
drugs' needs anticipation for each stimulus, to construct a measure of noxious stimulus

72

3.2 Methods
intensity for a wide list of stimuli, that can be later used in the search of pattern
responses in the objectively measured physiological variables during general anaesthesia.
The most commonly used scales in the assessment of acute pain are visual analogue
scales, numerical rating scales, and verbal scales. Although these scales are easy to understand and use, they reect individuals' assessment and cannot be compared between
subjects, troubling statistical analysis and operations between and within individuals.
This problem may be surpassed using the Rasch model [140, 141], a mathematical tool
that uses the population tendency rather than individual's assessment to obtain scaling
measurements [31, 142, 143].

Figure 3.1: Schematic representation of the patient's system and involved variables in
the triggered noxious responses. The amplitude of response to noxious stimulation, in
this simplied version, is dependent of the drugs' doses (known) and of the stimulus
intensity (estimated).

Other outer interferences may modify the exhibited relations

(signal contamination and changes in pharmacokinetics e.g.).

This carries important information in the search for relations between dierent stimuli, and the triggered responses in the physiological signals associated to noxious
activation.

The idea is to have an estimate of the stimulus intensity and assess the

impact of both stimulus intensity and drugs' attenuation, as shown in Figure 3.1.

3.2 Methods
This section intends to describe the methods employed in the development of the noxious
stimulus intensity scale. The survey design, the scale construction using Rasch analysis,
and comparison of the results with previous pharmacological studies, are reported in
detail.

73

3. STIMULUS INTENSITY ANALYSIS

3.2.1 Survey Design and Dissemination

A survey was designed and executed to evaluate anaesthesiologists' perception of stimuli
intensities during general anaesthesia. It was organized in three sections:

stimulation sourcing from the anaesthetic procedure;

stimulation sourcing from the surgical procedure pre-incision and incision;

stimulation sourcing from the surgical procedure post-incision.

A total of 35 stimuli were analyzed in this study and shown in Table 3.1.
For each stimulus in the survey, anaesthesiologists were asked to give an evaluation
in an ordinal rating score varying between 0 (non-noxious stimulus) to 10 (severe noxious
stimulus).
Besides stimulus assessment, a few questions considered relevant to the analysis
were added to the survey: clinician's perception of the existence of a tool to assess the
analgesic state of the patient, similar to the existing tools for the hypnosis state (likely
or unlikely); what scale would be chosen for such an index (0-1, 0-5, 0-10, 0-100, other);
what is the most intense stimulus, laringoscopy/intubation, or abdominal incision for
classic apendicectomy; number of years of experience since the beginning of the training
period; and nally gender of the participant (male or female).
The survey was implemented using Google Docs (https://spreadsheets.google.

com/viewform?formkey=dFg2Ul95a25zYlhRNDNPbEpGOV9TZkE6MA) and disseminated in

the anaesthesiologists community (form available on-line for consultation). The Portuguese Society of Anaesthesiologists disseminated the survey by e-mail to all its members
and publicized it in the society web page (http://www.spanestesiologia.pt/), also
the survey was sent by e-mail to all anaesthesiologists of Centro Hospitalar do Porto,
by the Anaesthesiology Service.
Collected data were organized in Excel data sheets for subsequent analysis.

3.2.2 Scale Construction

The Rasch model was used to translate the information contained in the ratings provided by anaesthesiologists into a measure of noxious stimulus intensity (Observations are
always ordinal; measurements, however, must be interval) [144]. The concept behind
this approach is to determine the magnitude relations between items, considering not

74

3.2 Methods

Table 3.1:

List of stimuli analyzed in the study, divided according to three groups:

stimuli derived from the anaesthesia procedure; stimuli derived from the surgical procedure, pre-incision and incision; and stimuli derived from the surgical procedure postincision.

Anaesthesia Procedure
Stimulation (11, An)

Surgical Procedure
Stimulation Pre-Incision
and Incision (10, PreInc)

Surgical Procedure
Stimulation Post-Incision
(14, PostInc)

1. G18 venous catheter in the

back of the hand (prior to drug
administration)
2. Laringoscopy

1. Folley catheter placement

1. Intraoperative lithotripsy
with ultrasound

2. Ureteroscopy/Cystoscopy

3. Intubation (moment of
passage of the tube through
vocal cords)

3. Skin incision for drain or

Small laparoscopic
incision/Percutaneous
Nephrostomy incision
4. Retractors placement and
Introduction of laparoscopic
surgery instruments
5. Inate abdominal cavity for
laparoscopy

2. Retractors placement in
abdominal surgery
3. Abdominal cavity tissues
traction

4. Nasal broscopy for

intubation
5. Gastric drainage
tube/Temperature probe
nasal/oral
6. Laringoscopy for
oro-pharyngeal throat-packing
7. G20 catheter in the radial
artery under general anaesthesia
without local anaesthesia
8. Puncture, dilation and
insertion of jugular or
sub-clavian catheter under
general anaesthesia without
local anaesthesia
9. Skin puncture and epidural
catheter placement under
general anaesthesia without
local anaesthesia
10. Tetanic stimulus (50 Hz for 5
seconds)
11. Extubation in awake patient
at the end of general anaesthesia

4. Internal sutures (muscle or

viscera)
5. Tissue manipulation
cystectomy

6. Transverse abdominal incision

of 15cm supra-umbilical
7. Skin incision for
prostatectomy

6. Tissue manipulation
nephrectomy
7. Tissue manipulation
prostatectomy

8. Skin incision for nephrectomy

8. Ganglia removal in the

inguinal region

9. Midline 10 cm lumbar incision

10cm for disc herniation surgery

9. Tissue manipulation
laparoscopy

10. Fronto-parietal craniotomy

incision 10 cm

10. Lesion extraction in

laparoscopic surgery
11. Tissue manipulation lumbar
disc hernia surgery
12. Periods in the surgery
without surgical stimulus (any
break in the surgery) in
abdominal surgery with patient
under general anaesthesia,
intubated and ventilated
13. Wall suture
14. Skin staples

75

3. STIMULUS INTENSITY ANALYSIS

only the respondents evaluations, but also the respondents' locations, obtaining independent measures of a certain latent dimension not observable directly.

The Rasch

model was rst developed for dichotomous data [140, 145, 146] (painful/not painful for
example) later expanded for the analysis of polytomous data, meaning for more than
two ordered response categories [147151].
This study aimed at obtaining measures of perceived stimulus intensity, based on
rates attributed by clinicians considering their clinical perception of the stimulus intensity; this problem corresponds to a two-facet Rasch model, the two facets being the
painful stimulus and the raters who evaluated each stimulus, in a rating scale from 0
to 10. The rating scores were analyzed using a two-faceted rating scale model [141], as
presented in Equation 3.1.

Px

j=0 [n (i
Pk
j=0 [n
k=0 exp

exp

nix = Pm
where

0 0

nix

so that

Pk

j=0 [n

perception of rater

(stimulus location,

i = 1, ..., 35),

n,

(rater location,

= 1, ..., 10).

x = 0, ..., 10

(i + k )]

(3.1)

(i + k )] = 1.

is the probability of rater

categories (k

+ k )]

and

giving a rating of

n = 1, ..., 57), i
k

x,

on stimulus

i, n

is the pain

is the painfulness of the stimulus

is the threshold between each pair of adjacent

This formula computes the probability that a rater will give

a particular response to a given stimulus.

The model parameters (n ,

i , k )

are estimated from the rating data obtained

for each stimulus (35 stimulus57 raters), providing insight not only on the perceived
stimulus intensity, but also on the raters' locations regarding pain scores attributed.
The subsequent analysis will include both facets analysis, since they are intertwined in
the measurement process (rater location and item location).
Model parameters are all in the same continuum of pain intensity/sensibility, in
a logit scale.

The logit scale may be linearly transformed to a more user friendly

scale varying from 0 to 10.

To perform the conversion, extreme scores logit values

were estimated. Extreme scores (perfect scores) imply extreme measures, but indenite
since they congure innite values of the latent trait. Nevertheless the location of the
boundaries may be estimated, considering a 0.5 score from the extreme perfect scores,
in order to obtain a more user friendly scale (linear transformation) [152, 153].
Item response theory methodology and Rasch analysis have been used in several
elds, such as educational applications in which the latent trait is usually referred to

76

3.2 Methods
as ability, but also dierent applications in the medical eld have explored this analysis
technique, in the search of measurement and tuned questionnaires to better evaluate
pain [31, 32, 142, 143, 154158], fatigue [159], physical functioning ability and outcome
[160165], and even analysis of the written anaesthesiology examination [166], among
other variables of interest.
Rasch analysis was performed using Minifac (Facets student version 3.66.3, Linacre
2010).

3.2.3 Pharmacological Studies Comparison

After obtaining a scale based on anaesthesiologists' clinical perception, the relation
between these results and the results obtained in previous pharmacological studies [78,
167170], in which drug needs to abolish reactions to dierent noxious stimuli were used
to quantify the stimulation intensity (response probability curves), was investigated.
This approach leads to a more direct evaluation of stimulus intensity, since the analgesic
drug needs to abolish the stimulus is used to assess stimulus intensity, and to obtain
this information for each stimulus would be a more adequate way of estimating noxious
stimulus intensity in anesthetized patients. Nevertheless, evaluating each stimulus drug
needs in pharmacological studies is time consuming, and with limited clinical relevance.
If the relation between clinical perception is shown to be comparable to the estimated
intensities through pharmacological studies, then the developed scale may be used to
assess stimulus intensity for a wide number of stimulus.
Bouillon et al. [167] used the drug needs to abolish noxious response to stimulation,
to obtain the probability curve of responsiveness to laringoscopy using a bivariate Hill
equation (Equation 1.7) as described in the previous chapter, and given by Equation
3.2.


Plaryngoscopy =

URlaryngoscopy +UP laryngoscopy

U50,laryngoscopy(laryngoscopy )


1+

laryngoscopy

URlaryngoscopy +UP laryngoscopy

U50,laryngoscopy(laryngoscopy )

laryngoscopy

where:

Plaringoscopy

is the probability of being unresponsive to laryngoscopy;

77

(3.2)

3. STIMULUS INTENSITY ANALYSIS

URlaryngoscopy

is the remifentanil concentration normalized by the

C50,remif entanil,laryngoscopy

(remifentanil concentration associated with 50% proba-

bility of no response to laryngoscopy);

UP laryngoscopy

is the propofol concentration normalized by the

C50,propof ol,laryngoscopy

(propofol concentration associated with 50% probability of

no response to laryngoscopy);

laryngoscopy = UP laryngoscopy /(UP laryngoscopy + URlaryngoscopy );

2
U50,laryngoscopy(laryngoscopy ) = 1laryngoscopy laryngoscopy +laryngoscopy laryngoscopy
;

laryngoscopy

is the interaction of propofol and remifentanil on the probability of

no response to laryngoscopy;

and

laryngoscopy

is the steepness of the relation between the drug combination

and the probability of no response to laryngoscopy [167].

Figure 3.2: Representation of the hypnotic-analgesic drugs' interaction as an hierarchical

model [167]).

In the same study, the opioid-hypnotic interaction was also modelled as an hierarchical model (see Figure 3.2), where pain is blocked by the use of opioids (attenuation),
and then the attenuated pain is projected to the cortex where hypnotics act. This is

78

3.3 Results
the concept behind the use of information contained in EEG based indexes to translate nociceptive activation, since the noxious stimulation is projected interfering with
cortical EEG patterns [129].
This approach interprets the potency of the opioids in attenuating the stimulus as
a function of the stimulus intensity given by Equation 3.3.

postopioid intensity=

preopioid intensity 1

opioid
opioid +(opioid50 preopioid intensity)

(3.3)

where:

preopioid intensity
postopioid intensity

in the intensity of the aerent noxious stimulus;

is the intensity of the noxious stimulus after opioid attenua-

tion;

opioid

is the opioid concentration;

opioid50

is the opioid concentration associated with 50% attenuation, at an inten-

sity score of 1, of the

and

preopioid intensity

is the steepness of the opioid-response (attenuation) relation [167].

The intensities estimated using the hierarchical model described are unitless and
are related to the probability of response. After attenuation the

postopioid intensity

is projected to the cortex and suppressed by the hypnotic drug, described by a similar
model structure (refer to [167] for more detailed analysis).
The drug needs were evaluated and stimulus intensities were estimated based on the
drug needs to abolish 50% of the response (pre-intensity estimation).

This approach

was further explored in a subsequent study by Huiku and colleagues [78], where the
authors estimated other noxious stimuli intensities based on drug needs reported in the
literature [167, 170, 171]. The pre-intensities estimated for the hierarchical model were
later used in the search for noxious responses in the physiological signals collected.

3.3 Results
There were 57 responders to the survey, 19 of which were men. Average years of experience in clinical anaesthesiology since the beginning of the internship of respondents

79

3. STIMULUS INTENSITY ANALYSIS

was 9.78.0 years (data presented as meanstandard-deviation).

Frequency of each

entry in the rating scale was (rating scale entry [0-10] and correspondent frequency of
occurrence): 0-25, 1-87, 2-154, 3-189, 4-281, 5-315, 6-290, 7-281, 8-218, 9-125 and 10-30.

Figure 3.3:

Anaesthesiologists' division considering their opinion on which is the

most painful stimulus,

abdominal incision for apendicectomy (26%),

or laringos-

copy/intubation (74%).

Figure 3.4: Total score for each stimulus within the three groups of stimuli studied:
anaesthetic procedure stimuli (Ani, i=1,...,11), pre-incision and incision surgical stimuli
(PreIncj, j=1,...,10), and post-incision surgical stimuli (PostInck, k=1,...,14).

Regarding the question, which is the most painful stimulus, incision for apendicectomy or laringoscopy/intubation, 74% of anaesthesiologists who answered the survey
considered laringoscopy/intubation as the most painful stimulus (see Figure 3.3).
Total scores obtained for each stimulus analyzed (anaesthetic, pre-incision and incision, post-incision; 0 to maximum of 570) are shown in Figure 3.4; total scores rated by
each respondent (0 to 350) are shown in Figure 3.5. There is a wide variation in total

80

3.3 Results

Figure 3.5: Total scores for each respondent, considering the 35 evaluated stimuli.

scores obtained for each respondent, consequence of dierent rater severities and scale
use/interpretation.

3.3.1 Clinical Perception: Monitoring for Nociception

One of the topics approached in this study was the inspection of the clinical perception
on the existence of a tool, similar to those used in the operating room in the evaluation of
patients' hypnosis state during anaesthesia, to evaluate the balance between nociception
and analgesia. 68% of the anaesthesiologists who answered the survey believe such a
tool will be in the future developed for use in the operating room, while the remaining
30% believe that the development of such tool in the future is unlikely.

Figure 3.6:

Anaesthesiologist's choice on the preferred range for a nociception/anti-

nociception balance index.

Assuming the existence of such a tool, anaesthesiologists were also asked to give

81

3. STIMULUS INTENSITY ANALYSIS

their opinions on the preferred range for a Noc/ANoc balance index (0-1, 0-5, 0-10, 0100, Other). Anaesthesiologists who answered the survey chose an index to vary from 0
to 10 (53%) or from 0 to 100 (47%), with no occurrence in the other available categories
as shown in Figure 3.6.

3.3.2 Measurement Construction Process

Following data collection, the two-facet matrix (35 stimuli57 raters) was exported
to a text specication le, used in the Minifac (Facets student version 3.66.3, Linacre
2010) software for Rasch analysis. The analysis that followed was divided in the two
considered facets: stimuli and raters. The linear measures obtained were all located in
the same stimulus intensity (sensibility) continuum.
After adjusting the two-facet Rasch model to the data, outlying observations were
detected, with recurrent incidence of unexpected responses from raters 20, 23, 28, 57
and 14.

As an example, observing Figure 3.5, respondent 23 presented a low total

score when compared to the remaining responders, demonstrating lower severity. This
may have been due to a misunderstanding of the rating criteria, leading to unexpected
responses within the measurement process, or to loss of collaboration in the survey
response, since the responses by this rater were of 0 in many of the stimuli analyzed.
Raters who had recurrent incidence of outlying entries, meaning unexpected responses within the adjusted Rasch model to the population trend (response outside modelled
boundaries resulting in high residuals), were removed from the data set, and the analysis
conducted with the remaining evaluators. Each removed rater was individually analyzed to identify the possible exclusion cause. The removed raters either had unrealistic
ratings (loss of collaboration in the survey response), or dierent opinions regarding
some of the stimuli within the population trend, leading to occurrence of high residuals.
Stimuli measures (i ,

i = 1, ..., 35,

in the Rasch model) are shown in Table 3.2, for

each group of stimuli considered. The model obtained presents a reliability index of 0.98
(98% of the observed variance resulted from true measure variance). After obtaining
the Rasch measurements for each stimulus in logit, measures were linearly rescaled to
present values in a linear scale from 0 to 10, for ease of understanding and application
(Measure [0-10] in Table 3.2).

To obtain this scale, extreme scores were estimated

using values of 0.5 from the perfect score; values of -3 for 0, and 3.3 for 10 were used.
Stimulus intensity measures (logits) provided by the adjusted Rasch model are also

82

3.3 Results
Table 3.2: Total and average scores obtained for each stimulus, divided according to the
considered groups, stimuli measures (i ,

i = 1, ..., 35,

in the Rasch model, with greater

logit values indicating higher pain perception), standard errors (S.E.), expected raw
score, Rasch measure parameterized in a scale from 0 to 10, and stimulus identication,
in descending order of stimulus intensity perception (reliability of 0.98).

Anaesthesia Stimuli

Total Observed Measure

Score Average
(logit)
399
364
332
302
269
225
219
204
203
198
194

7,67
7,00
6,38
5,81
5,17
4,33
4,21
3,92
3,90
3,81
3,73

1,28
0,88
0,56
0,28
0,00
-0,37
-0,42
-0,55
-0,56
-0,60
-0,64

S.E.
(logit)
0,11
0,10
0,10
0,09
0,09
0,09
0,09
0,09
0,09
0,09
0,09

Expected Measure
Stimulus
Score
[0-10]
7,74
7,08
6,46
5,85
5,20
4,32
4,20
3,89
3,87
3,78
3,68

6,81
6,16
5,65
5,21
4,76
4,17
4,09
3,88
3,87
3,80
3,74

An2
An3
An6
An4
An10
An11
An8
An1
An7
An5
An9

Brief Descriptor
Laringoscopy
Intubation
Throat-packing
Fibroscopy
Tetanic
Extubation
Jugular or sub-clavian puncture
G18 catheter
G20 artery
Gastric tube/Temperature probe
Epidural catheter

Pre-Incision and Incision Stimuli

Total Observed Measure

Score Average
(logit)
417
413
379
368
353
306
286
273
256
168

8,02
7,94
7,29
7,08
6,79
5,88
5,50
5,25
4,92
3,23

1,53
1,47
1,04
0,92
0,76
0,32
0,15
0,03
-0,11
-0,87

S.E.
(logit)
0,12
0,12
0,11
0,10
0,10
0,09
0,09
0,09
0,09
0,10

Expected Measure
Stimulus
Score
[0-10]
8,10
8,01
7,36
7,16
6,86
5,94
5,55
5,27
4,94
3,17

7,20
7,11
6,43
6,23
5,98
5,27
4,99
4,81
4,58
3,37

PreInc6
PreInc8
PreInc7
PreInc9
PreInc10
PreInc5
PreInc4
PreInc3
PreInc2
PreInc1

Brief Descriptor
Incision supra-umbilical
Incision nephrectomy
Incision prostatectomy
Incision lumbar
Incision craniotomy
Inate abdominal cavity
Retractors in laparoscopy
Incision drain/lapar./nephrostomy
Ureteroscopy/Cystoscopy
Folley catheter

Post-Incision Stimuli

Total Observed Measure

Score Average
(logit)
363
325
312
298
292
266
260
260
254
249
247
243
236
106

6,98
6,25
6,00
5,73
5,62
5,12
5,00
5,00
4,88
4,79
4,75
4,67
4,54
2,04

0,87
0,49
0,37
0,25
0,20
-0,03
-0,08
-0,08
-0,13
-0,17
-0,19
-0,22
-0,28
-1,51

S.E.
(logit)
0,10
0,10
0,10
0,09
0,09
0,09
0,09
0,09
0,09
0,09
0,09
0,09
0,09
0,11

Expected Measure
Stimulus
Score
[0-10]
7,07
6,31
6,05
5,78
5,67
5,13
5,01
5,01
4,89
4,79
4,75
4,67
4,53
1,95

6,15
5,55
5,36
5,16
5,07
4,72
4,64
4,64
4,56
4,49
4,46
4,41
4,32
2,35
83

PostInc3
PostInc2
PostInc6
PostInc7
PostInc5
PostInc11
PostInc13
PostInc8
PostInc14
PostInc10
PostInc1
PostInc9
PostInc4
PostInc12

Brief Descriptor
Abdominal tissue traction
Retractors abdominal
Nephrectomy manipulation
Prostatectomy manipulation
Cystectomy manipulation
Lumbar manipulation
Wall suture
Ganglia removal
Skin staples
Lesion extraction laparoscopy
Lithotripsy/Ultrasound
Laparoscopy manipulation
Internal sutures
No surgical stimulus

3. STIMULUS INTENSITY ANALYSIS

Figure 3.7: Representation of the measurement continuum: stimulus location (i ,

1, ..., 35,

i=

in the Rasch model), and correspondent expected score in the original rating

scale.

84

3.3 Results
shown in Figure 3.7, demonstrating the dispersion of the stimuli along the stimulus
intensity continuum.

Figure 3.8: Representation of the probability of response of each rating score (0 to 10),
depending on the location in the pain intensity continuum (logit).

The probability of response in a determined rating score, regarding the pain intensity
location, is shown in Figure 3.8, and the relation between scores, and pain perception
measures given by the model, shown in Figure 3.9. Approximately between -1 and 1
logit the relation is close to linear; the relation is given by an ogival shape function,
which transforms non-linear scores into linear pain perception measures.
Distribution of raters' location (n ,

n = 1, ..., 57,

in the Rasch model) is shown in

Figure 3.10, measures are not presented for brevity. Raters' model presents a reliability
of 0.94 (6% of the observed variance results from measurement error).
The most intense stimuli considered were: abdominal incision 15cm supra-umbilical,
nephrectomy incision, laringoscopy, prostatectomy incision, lumbar incision, intubation,
abdominal cavity tissues traction, and incision for craniotomy. Stimuli with lower perceived intensity were:

venous catheter, radial artery catheter, gastric drainage tube

insertion or temperature probe, epidural puncture, Folley catheter, and periods in surgery with no surgical stimulus.
As aforementioned, for the comparison of the relative pain intensity considering laringoscopy/intubation and abdominal incision, 74% of anaesthesiologists agreed that

85

3. STIMULUS INTENSITY ANALYSIS

Figure 3.9: Relation between score rates (0 to 10) and pain intensity measures (logit),
with overlapping 0.5 points of expected score (squares).

Figure 3.10: Representation of raters' location (n ,

n = 1, ..., 57,

in the Rasch model)

distribution on the measurement continuum (logit).

laringoscopy/intubation is a more intense stimulus than abdominal incision for classic

apendicectomy. Regarding the measures results, this was also observed: laringoscopy

86

3.3 Results
presented a measure of 1.28 logits, followed by intubation with 0.88 logits, with laringoscopy presenting a higher measure than incision for prostatectomy, only surpassed
by incision for nephrectomy (1.47 logits) and transversal abdominal incision of 15cm
supra-umbilical (1.53 logits), of all stimuli considered.

Figure 3.11: Application of the developed stimulus intensity scale: schematic representation of an urological procedure under general anaesthesia with identied stimuli
intensities of the anaesthetic and surgical actions.

For demonstration purposes, the developed scale was applied to describe changes in
noxious stimuli for one of the urological procedures collected under general anaesthesia,
and shown in Figure 3.11 with the illustration of dierent moments of the anaesthetic
and surgical procedures, characterized by dierent intensity levels. The stimulus map
corresponds to a prostatectomy procedure, with the dierent stimuli moments both from
anaesthetic and surgical procedures; observation of the estimated intensity variations
allows to infer the changes in the analgesic drug to compensate and anticipate the
stimulus occurrence.
Observing Figure 3.11 a limitation in the study is encountered, since moments
without stimulus post-intubation were not dierentiated from moments with no stimulation prior-intubation. The stimulation of an endotracheal tube should not be disregarded, and should be dierentiated when evaluating the stimulus intensity. Although

87

3. STIMULUS INTENSITY ANALYSIS

Periods in the surgery without surgical stimulus (any break in the surgery) in abdominal surgery with patient under general anaesthesia, intubated and ventilated was
evaluated, a new entry should be introduced in the survey, for the stimulus Patient
under general anaesthesia, intubated and ventilated, nevertheless this should pose no
problem for the remaining analysis.

3.3.3 Pre-Intensities vs Rasch Scale

Table 3.3: Estimated pre-intensities in previous pharmacological studies [78, 167], and
Rasch measures obtained using anaesthesiologists noxious stimuli perception.

Stimulus

Pre-Intensity Rasch Study Stimulus

Laringoscopy

0,83

6,81

An2

Laringoscopy + Intubation

1,57

6,16

An3

0,8

4,81

PreInc3

1,25

4,99

PreInc4

4,41

PostInc9

Intense abdominal exploration

1,5

6,15

PostInc3

Electric tetanic 30s

0,8

4,76

An10

Incision (laparoscopy)
Trocar
Laparoscopy surgery

NoStim

Figure 3.12: Comparison between pre-intensities estimated in previous pharmacological

studies [78, 167], and the Rasch scale obtained using anaesthesiologists noxious stimuli
perception.

Stimuli pre-intensities estimated based on previous pharmacological studies [78, 167]

88

3.4 Discussion
were compared to the Rasch scale obtained in this study, using the clinical perception.
The referred studies obtained an estimation of the stimulus intensity based on analgesic
drug needs to abolish the response to stimulation. Results of the comparison are shown
in Table 3.3 and Figure 3.12, with the stimuli considered in the pharmacological studies,
and correspondent stimulus in this study, the pre-intensities estimated, and corresponding value in the Rasch scale. Notice that laringoscopy+intubation was compared to
intubation in this study, leading to lower values in the Rasch scale, most probably
due to the way the question was posed to the respondents who were asked to evaluate
intubation (moment of passage of the tube through vocal cords) independently from
laringoscopy. Nevertheless, the relation between the two assessments is close to linear,
supporting anaesthesiologists' stimulus assessment and their evaluations, which lead to
the Rasch scale proposed in this thesis.

3.4 Discussion
Rasch models are probabilistic models which lead to the construction of linear continuous measures, based on ordinal observations of the population. A measure of noxious
stimuli intensity was successfully constructed, based on anaesthesiologists' perception,
which may be used in the development of a tool to assess Noc/ANoc balance during
general anaesthesia. To our knowledge this is the rst time such information was obtained. It is a simple study to conduct and its importance may be disregarded, however
we think that the information gathered through such a simple method is relevant and
may be useful.
The number of respondents to the survey (57) may seem small, however this number
corresponds to approximately 5% of the total number of Portuguese anaesthesiologists.
One may argue that the number of respondents is low, nonetheless the model of the
population trend presented a reliability of 98%, and results should not vary signicantly
with the introduction of more raters in the process. Also, although it was applied to
the Portuguese anaesthesia community only, there is no reason for bias.
Portuguese anaesthesiologists who answered the survey in their majority believe
that a tool to assess the Noc/ANoc balance, similar to the tools used in the assessment
of hypnosis and neuromuscular blockade, will be available in the future. Nevertheless
there is still a high incidence of skepticism for this matter, and 30% of the enquired
professionals are not sure that this tool may be in the future developed.

89

There are

3. STIMULUS INTENSITY ANALYSIS

however several research and commercial groups interested in this matter, with several
proposals to assess the remaining component of the anaesthesia triad. Monitoring cardiovascular changes [77, 79, 85, 86], skin conductance [109], frontal electromyography
activity [21], electroencephalogram derived indices variability [16], pupil diameter [72]
and evoked reexes [119], are some of the explored methods so far in the search of such
a tool, being some of these methods already commercially available.
As for the preferred range of such an index, assuming its existence, it was clear that
the preferred scale to be used must be similar to the currently used in the hypnosis
monitors, varying from 0-10 or 0-100. A tool developed for this purpose, must consider
the impact and acceptance of their users, and the only way a product will be successful
is if it is easy to interpret and use. The way humans perceive information is fascinating,
and several studies have already approached this matter when it comes to developing
monitors for anaesthesia use [172, 173], and the way the product is designed must take
advantage of this knowledge diminishing errors' incidence.
Laringoscopy and intubation are considered to be intense noxious stimuli, tendency
observed in the measurement of noxious stimulus intensity, where laringoscopy was the
third most intense stimulus, of all considered stimuli, only surpassed by nephrectomy
and transversal 15cm abdominal supra-umbilical incisions. The evaluation of laringoscopy as one of the most noxious stimulus in anaesthetic and surgical procedures is
corroborated by previous studies, were cardiovascular variables were monitored, and
compared for dierent noxious stimulations, demonstrating increased amplitude responses [168] and drug needs [169, 170] in laringoscopy/intubation when compared to
incision.
The way the rating scale was used diered according to the scale interpretation made
by each respondent (see Figure 3.5), conditioned by rater severity. The Rasch model takes the dierent raters severities into account to model the population trend, and obtain
independent measures. Nevertheless, occurrence of outliers is a concern with any type of
measurement, and surveys are prone to outliers, resulting from scaling rate misinterpretation, or misuse when responding to the survey, or even to dierent perspectives and
assessment, when compared to the population modelled trend. Recurrent incidence of 5
respondents, with outlying entries considering the modelled population trend, were observed. The identied outliers in the model residual analysis may have dierent reasons
of occurrence as referred before, leading to the appearance of unexpected values within
the modelled population trend and, like in any other modelling process, the outlying

90

3.4 Discussion
observations were removed. The occurrence of outlying observations was not surprising,
since the survey was long and loss of collaboration was a real concern, also some stimuli
may be dierently interpreted by raters, leading to unexpected responses considering
the population modelled trend.
Regarding the stimuli perceived as less intense, the one assessed to be the less
intense of all considered was periods in surgery with no surgical stimulus, even when
compared to a venous catheter puncture. This may be due to the way the problem
was posed to the respondents, who were asked to evaluate stimulus noxious intensity
considering a venous puncture prior to drug administration, and for the remaining
stimuli noxious intensity under general anaesthesia.

The periods in surgery with no

surgical stimulus during a surgical procedure still imply the presence of noxious stimuli
namely from the presence of a tracheal tube, positioning, or of retraction, nevertheless it
also implies the use of anaesthetics to maintain the patient in a state of unconsciousness,
as for the venous catheter puncture stimulation was considered to occur in an awake
patient, which is in many cases suering from anxiety in the period prior to anaesthesia
induction.

This was the probable reason why a venous catheter puncture in the

awake patient was considered to be a more intense stimulus than periods in surgery
with no surgical stimulation. When trying to evaluate personal opinions it is of great
importance to adequately state questions to obtain usable information.
One concern which may rise with this study is the use of subjective individual assessment of noxious stimuli intensities, which may be prone to bias and misinterpretations.
Although with the Rasch analysis the dierence in raters' severity should be compensated to produce independent measures, to understand the implications of this matter,
relation between pre-intensities obtained in pharmacological studies, which are based
in objective data of drug needs, and the proposed Rasch scale were investigated.

linear relation between the two approaches was observed; the result strengthens the
method proposed in this study, showing that clinical perception may be used as a tool
to estimate noxious stimulus intensity, as it would be expected since it is based on clinical experience and perception, that anaesthesiologists guide anaesthesia (drug doses)
in their daily activities. Since anaesthesiologists' perception of stimulation intensity is
based on the stimulus anticipation and drugs' adjustment, it was expectable that this
relation was strong, nevertheless the conrmation of this relation, leaves room for the
use of this simple method in a wider number of noxious stimuli.

91

3. STIMULUS INTENSITY ANALYSIS

General anaesthesia and surgery imply frequent changes in stimulus and drugs'
administration. When one thinks about the relationship between stimuli and drug doses,
one tends to think that a change in stimulus is followed by a complementary change in
drug concentration. Anaesthesiologists try to anticipate drugs' requirements and adjust
drugs' concentrations prior to the stimulus change in order to avoid the consequences
of inadequate anaesthesia. This is feasible and currently done for well dened moments
like laringoscopy and intubation [170]. For adjustments during surgery it would be ideal
to have a system performing a constant evaluation of the balance between stimulus
and anaesthetics, namely between nociceptive stimuli and anti-nociception.

An anti-

nociceptive drug which may be adjusted very quickly is already available - remifentanil what is lacking is an objective measure of nociception. In previous studies, a search for a
nociception index has lead researchers to compare dierent precise stimulus, for dierent
drug doses [77, 78, 86, 87, 138], and attempts have been made to obtain scaling methods
connecting stimulus and analgesic level, nevertheless taking into account few experts
opinions [77]. In this study a survey was designed to evaluate the clinical perception
of several anaesthesiologists on the stimuli intensity, leading to the construction of a
measure of stimuli intensity that can be later used to further investigate changes in
physiological parameters in response to stimulus variations.
Besides the usability of this technique in the search for a tool to evaluate and quantify
the Noc/ANoc balance, results in this study allow clinicians an overview of stimuli
intensities they encounter in their daily activity, and the inference of the relations
between them. This scale provides a term of comparison between dierent stimuli, and
a chronological prediction of intensity in the anaesthetic and surgical procedures.
The present survey must be simply regarded as a tool to obtain data and develop
a scale to be applied to clinical studies oriented at relating noxious stimulus, antinociception drugs, and correspondent measurable physiological responses in the patient.
The novelty brought by this study was the use of Rasch analysis to construct the noxious
intensity measure, with the input of a group of anaesthesiologists, for a more objective
assessment of these relations.

3.5 Summary
A measure of stimulus intensity was constructed, allowing scaling and comparability
between the considered noxious stimuli. This information was used to better understand

92

3.5 Summary
the relationship between stimulus intensity, drug doses and physiological responses in
the collected data.
The original

scale

variable, containing the code numbers annotated by the author

during the stay in the operating room, was updated to contain the corresponding stimulus intensity values of the developed noxious stimulus intensity scale. A new variable
containing this information was created (sca) and employed in the following analysis,
to translate the stimulus intensity the patient was under throughout the entire case.

93

3. STIMULUS INTENSITY ANALYSIS

94

Chapter 4
Data Analysis - Passive Nociception
Measures

Each problem that I solved became a rule

which served afterwards to solve other problems.
Ren Descartes

This chapter is focused on the analysis of passive Noc/ANoc balance measures. After the
pre-processing of the collected data, as described in Chapters 2 and 3, it was analyzed
in the search for the best descriptors of noxious activation, and of the relations between
input and output signals.

The term passive is here applied in the sense that the

analyzed signals, and variations in response to noxious stimulation, are merely observed
without outer interference.

4.1 Introduction
The objective during general anaesthesia is to maintain the patient in homeostasis, and
provide conditions for the clinical procedures to be performed. Considering the system
represented in Figure 1.20, if the input drugs are constant, the output should only vary
with a stimulus variation, and therefore measures of variability of the system outputs
may translate the Noc/ANoc balance of the patient.

95

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Most of the methods proposed in the literature for the Noc/ANoc balance assessment are based on passive measures related to noxious activation, their normalized
value and variability: heart rate, blood pressure, photoplethysmography wave amplitude, skin conductance, pupil diameter, EEG derived indexes and electromyography.
Some methods are combinations of variables which contain complementary information, and others are based solely on one physiological variable. All these physiological
signals are conditioned by many factors of the environment and natural auto-regulation
mechanisms (e.g. positioning).
Surely that if only one variable could convey the whole information on patient's
state concerning the analgesia component of general anaesthesia, this would be greatly appreciated, specially if this variable is already being collected in the usual clinical
practice. Nevertheless, a single variable that is not directly connected to the sensing
process, and is within the auto-regulation mechanisms, may lead to erroneous assumptions about the patient's state if regarded individually. As an example let us consider
heart rate, a physiological signal linked to noxious activation. It is true that with increasing stimulation an increase in heart rate is expected as a response to the stimulus,
however a paradoxical scenario may be observed, in which a decrease in heart rate occurs to compensate an elevation in blood pressure. All these factors must be taken into
account when designing a tool to translate the patient's state considering the analgesia component, and even so, a new index must always be viewed and analyzed within
clinical context.
The following sections explore collected data during general anaesthesia to evaluate
which of the physiological signals are related to noxious activation, and the best methods
to extract the information contained in these variables on the Noc/ANoc balance of the
patient.

4.2 Phase I Precise Stimuli Analysis

In this phase of the study the objective was to inspect collected signals variations
in response to precise reproducible noxious stimuli, with equivalent intensity between
patients, and also the impact of the analgesic dose (2, 3 and 4 ng/ml of remifentanil
eect-site concentration) on the response amplitude observed (see Figure 2.1).

This

analysis allows the denition of the more informative variables responding to noxious
stimulation in standardized conditions.

96

4.2 Phase I Precise Stimuli Analysis

The stimuli chosen for this phase of the study were laringoscopy/intubation, tetanic
stimulus and incision, since they are applied in all procedures, and considered to have
similar intensities between patients.

Data analysis is divided according to amplitude

response to precise stimuli, and analgesic dose impact on the amplitude response.

4.2.1 Methods

Figure 4.1: Schematic representation of proposed methods to assess amplitude responses

prior and post-stimulation:

a) and b) average versus maximum response; c) and d)

average values prior and post stimulation.

Several methods are proposed in the literature to assess amplitude response to precise stimulation, from comparison of baseline values prior stimulation with maximum
response after stimulation, average values before and after, among others. The methods
should be applied carefully since comparisons between dierent extracted measures may
lead to erroneous assumptions, for example comparing average values in a time window
to the maximum value reached post-stimulation, is doomed to bias, and tends to show
dierences (see Figure 4.1).

Knowing this, and exploring the proposed methods the

average value in a time window prior stimulation, versus the average post-stimulation
were used, followed by amplitude response assessment given by Equation 4.1.

97

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Amplitude Response = Average(tP ost ) Average(tP re )

where

tP re =60 s

and

(4.1)

tP ost =180 s.

Besides obtaining average values for each time span, dierences observed in individual baseline values of variables analyzed in this study should be taken into consideration. To increase comparability between individual responses all

Amplitude Response's

extracted were normalized by the baseline values observed in each patient prior drugs'
administration, in this way, more precise results may be guaranteed when comparing
groups of patients, for example to compare the impact of the drug dose on amplitude
response for the dierent groups of study.

Amplitude ResponseN

was obtained using

Equation 4.2.

Amplitude ResponseN =
where

tP reDrugs =180 s

Amplitude Response
Average(tP reDrugs )

(4.2)

is a time window prior drugs' administration.

The variables analyzed in this study were HR, SBP, PPGA, BIS, EMG, BIS SD,
EMG SD, CVI, SEF, TOTPOW, and ASYM. Besides the cardiovascular and EEG
derived indexes collected during general anaesthesia, an additional index proposed in
the literature for the Noc/ANoc balance assessment (ANSSI) was implemented, since it
is an open source index. ANSSI was obtained as described in Chapter 1, in Equations
1.11 and 1.12.
During data collection, a BIS bilateral sensor was used, and laterality eects reported
in the literature were taken into account [17]. Since BIS data were available for the two
EEG channels, contralateral data according to the site of surgical incision was used:
if the incision was on the left side, EEG right channel was used and vice-versa; if the
incision was central, the average of the indexes from both channels was used instead.

4.2.1.1 Amplitude Response and Analgesic Dose Impact

Amplitude response to each stimulus was assessed using the previously described methods.
The objective was to assess the amplitude response of each collected variable to the precise stimulus, and verify if there were dierences before and following stimulation (paired
samples). Amplitude responses were also compared regarding the dierent stimuli. For
this analysis, the amplitude response (Equation 4.1) was analyzed and compared within
the same subject (paired samples), for the three considered noxious stimuli.

98

4.2 Phase I Precise Stimuli Analysis

Besides intra-patient assessment, the clinical protocol described in Chapter 2, allowed the inspection of the analgesic drug dose impact on the amplitude response for
each stimulus. In this case a comparison between patients is necessary, for the same
stimulus and dierent remifentanil eect-site concentrations; original and normalized
response amplitudes (Equations 4.1 and 4.2) were used in the comparisons (independent samples).

Figure 4.2: Schematic representation of the laringoscopy and intubation moments, registered in each data set, and the corresponding time lag - stimulus duration.

Although the stimuli chosen to assess the analgesic impact between groups of patients are considered to be similar between individuals, the fact is that stimulation
duration from laringoscopy to intubation may dier according to the diculty of intubation. Although the stimulus is the same, in some cases the duration of the stimulus
diers (consequence of a dicult intubation), leading to more pronounced responses
(e.g.

anatomical dierence).

Therefore, dierences between groups considering time

from rst laringoscopy to intubation were analyzed, to verify if there was any bias
factor (see Figure 4.2).
In addition to the dierence in stimulus duration for laringoscopy/intubation, the
incision stimulus may also introduce bias. Incision stimuli intensities dier according to
the surgical procedure, as described and concluded in the previous chapter (Rasch scale).
Since the study groups are composed of dierent surgical procedures, and consequently
dierent incision types, dierences between groups considering incision types, meaning
intensities, were investigated.

99

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

4.2.1.2 Statistical Analysis
Non-parametric paired sample tests were used for the comparison of the observed values
before and following stimulation, since measurements inside the same patient (not independent) were being compared (paired sample median test, Wilcoxon signed ranks).
For the assessment of the analgesic impact, a non-parametric median test was employed,
since in this case comparison between dierent groups of patients is performed (independent, Kruskal-Wallis and Mann-Whitney median tests).
signicant. Matlab

P < 0.05

was considered

R

software was used for statistical analysis.

4.2.2 Results
Table 4.1: Patients' demographic data (original and phase I nal sample): global and
for each considered study group of remifentanil target eect-site concentration (data as
meanstandard-deviation).

Original Data Set

Gender (M/F)

Global Group 1 Group 2 Group 3

18/13

5/5

8/3

5/5

Age (years)

54.913.4

58,215,4

50,813,8

56,110,6

Weight (kg)

69.912.3

66,010,9

74,914,4

68,510,3

Height (cm)

165.07.6

161,92,6

168,49,7

164,47,5

Phase I Data Set

Gender (M/F)

Global Group 1 Group 2 Group 3

15/11

4/5

7/2

4/4

Age (years)

55.012.4

57,015,8

54,011,1

53,910,6

Weight (kg)

69.412.2

63,49,1

77,212,9

66,810,9

Height (cm)

164.57.5

161,62,5

168,79,6

163,17,6

Data were collected from 34 patients under general anesthesia (3 excluded) as described in Chapter 2. From the 31 remaining data sets, ve were eliminated from this phase
of the study due to protocol deviations: one case from Group 1, two cases from Group
2, and two cases from Group 3. The nal data set was composed by 26 patients, nine in
Group 1, nine in Group 2 and eight in Group 3. Table 4.1 shows the demographic data
of the nal sample. No statistical dierences between groups considering demographic

100

4.2 Phase I Precise Stimuli Analysis

Table 4.2: Patients' distribution on the three possible incision sites, for each group of
study, in the nal sample (no statistical dierence between groups).

Incision Site
Left Central Right
Group 1

Group 2

Group 3

data were found (Lilliefors, Kolmogorov-Smirnov and Fisher's tests). Data presented
as meanstandard-deviation.
Table 4.2 shows the frequency of occurrence in each study group of each incision
site: left, central of right. No statistical dierence was found between groups considering
incision larerality distribution.
Time from rst laringoscopy to intubation was registered and analyzed in the search
for possible bias factors. In general time from rst laringoscopy to intubation was below
60 seconds, only two cases exceeded the one minute interval, and in only one case
intubation was guided using GlideScope

R1

(median of intubation times, excluding the

guided intubation was of 16 s, ranging from 4 s to 144 s), with no statistical dierence
between groups considering time for laringoscopy (Kruskal-Wallis test).
Incision intensity was compared between study groups to verify if there was any
dierence that could bias the observations. No statistical dierence was found between
groups considering surgical incision intensity (Kruskal-Wallis test). Average observed
intensity in the study groups were 5.951.11, 5.810.93 and 6.300.92 in the Rasch
scale.

4.2.2.1 Amplitude Response and Analgesic Drug Impact

For each stimulus analyzed the average values prior and post stimulation were compared,
with observed statistical dierences between some of the variables. Tables 4.3 and 4.4
present the median values observed for each physiological variable analyzed, for the
averages prior and posterior to stimulation, in each of the study groups.

R is a medical device used to assist dicult intubations. It provides airway video

Glidescope
monitoring, while performing laringoscopy/intubation.
1

101

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.3: Median values for each cardiovascular variable analyzed in the study, for each
stimulus (average pre and post stimulation), divided according to the drug dose group:
remifentanil eect-site concentration (RemiCe) of 2.0, 3.0, or 4.0 ng/ml (* stands for
signicant dierences with

Baseline

P < 0.05).

Laringoscopy
Pre
Post

Tetanic
Pre Post

Incision
Pre
Post

HR
RemiCe=2.0

67,78*

63,98*

71,66*

60,58

59,68

58,69

59,00

RemiCe=3.0

68,98*

58,85*

65,14*

64,86

61,00

59,60

58,59

RemiCe=4.0

82,61*

59,38*

65,71*

63,81

63,48

62,72

62,35

RemiCe=2.0

126,34*

91,66*

108,16*

103,37

100,97

104,90*

108,07*

RemiCe=3.0

125,05*

97,43*

102,04*

100,95

99,23

98,51*

115,37*

RemiCe=4.0

120,54*

101,26*

96,76*

101,28

95,02

96,83*

104,75*

RemiCe=2.0

1,12*

6,14*

4,10*

1,11

1,07

1,62*

1,02*

RemiCe=3.0

1,53*

5,75*

5,18*

6,38

7,32

6,52*

2,33*

RemiCe=4.0

1,17*

2,46*

2,68

1,82

2,49

2,56

2,29

RemiCe=2.0

82,78*

38,34*

54,39*

85,70

85,52

78,49*

81,92*

RemiCe=3.0

90,15*

48,85*

55,49*

49,93

46,93

55,59*

67,09*

RemiCe=4.0

82,73*

59,48*

57,36

60,64

52,66

60,95*

67,12*

SBP

PPGA

ANSSI

Statistical dierences between HR values prior and post-laringoscopy/intubation

were found in all groups of study, with an increase in HR post-stimulation. A tendency
of decreasing HR was observed in response to tetanic stimulation, and no statistical
dierence was observed for the incision stimulus. Also a statistical signicant depression
in HR values at induction was observed in all study groups.
Statistical signicant dierences were observed globally in SBP response to the stimuli laringoscopy/intubation and incision. Signicant increase in SBP was observed in
study groups 1 and 2 to laringoscopy/intubation, not visible in study group 3 (possible total stimulus block). Signicant depression of SBP was also observed globally at
induction. No statistical dierences between SBP median values prior and post tetanic
stimulation were observed, nevertheless observing the median values extracted (see Ta-

102

4.2 Phase I Precise Stimuli Analysis

Table 4.4: Median values of BIS related variables analyzed in the study, for each stimulus (average pre and post stimulation), divided according to the drug dose group:
remifentanil eect-site concentration (RemiCe) of 2.0, 3.0, or 4.0 ng/ml (* stands for
signicant dierences with

P < 0.05).

Baseline Laringoscopy
Pre Post

Tetanic
Pre Post

Incision
Pre Post

BIS
RemiCe=2.0

81,43*

43,63*

46,60*

46,79

42,50

41,79

43,44

RemiCe=3.0

95,81*

50,51*

47,00

42,42

43,17

42,63

43,78

RemiCe=4.0

87,91*

47,14*

45,08

43,22

45,95

45,39

45,21

RemiCe=2.0

3,60

3,21*

7,13*

3,26

3,25

2,44*

3,11*

RemiCe=3.0

1,93

2,52*

4,88*

2,79

3,28

2,94*

3,33*

RemiCe=4.0

3,17

2,83*

5,33*

3,84*

4,51*

3,19*

5,25*

RemiCe=2.0

44,67*

26,52*

28,19*

26,51

26,90

25,68

26,46

RemiCe=3.0

48,62*

27,23*

27,67*

26,70

26,93

25,90

25,94

RemiCe=4.0

49,85*

28,93*

29,29*

26,51

26,65

26,67

26,63

RemiCe=2.0

2,98*

0,88*

3,41*

1,21

0,43

0,44*

1,04*

RemiCe=3.0

3,76*

1,63*

2,85*

0,40*

1,56*

0,33

0,34

RemiCe=4.0

3,18*

1,12*

3,10*

0,41

1,22

1,25

1,02

RemiCe=2.0

22,50*

16,19*

16,31

16,17

15,93

15,67

15,68

RemiCe=3.0

20,11*

16,47*

16,42

15,42

15,02

15,38

15,40

RemiCe=4.0

24,55*

16,10*

15,92

16,31

16,26

16,36

16,60

RemiCe=2.0

56,76

62,67

62,42

60,99

61,74

61,01

62,06

RemiCe=3.0

61,59

61,55

61,71

61,17

60,88

61,11

60,79

RemiCe=4.0

59,31

64,41

63,20

61,19

61,16

61,00

60,52

RemiCe=2.0

51,02

52,47

51,36

52,65

53,05

54,78

53,81

RemiCe=3.0

51,68

48,79

49,75

50,29

49,48

50,02

48,59

RemiCe=4.0

54,13

48,89

48,60

48,97

49,39

49,76

49,31

BIS SD

EMG

EMG SD

SEF

TOTPOW

ASYM

103

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

ble 4.3), a tendency of depressing SBP values, with increasing remifentanil doses was
observed.
PPGA signicantly increased value following induction in all study groups.

Also

PPGA was globally signicantly depressed in response to the laringoscopy/intubation

and incision stimuli, although not statistically signicant in study group 3 (probably
due to stimulus total block). Although no statistically signicant responses to tetanic
stimulus were observed, it may may be observed that in remifentanil group 1 a small
tendency towards PPGA decrease is observed, nevertheless this tendency was inverted
with the increase of remifentanil doses.
Concerning the variables provided by the BIS monitor, and their median values
prior and post-stimulation (see Table 4.4), a statistically signicant increase in BIS
values post-laringoscopy/intubation in study group 1 was observed, not observed in the
remaining study groups.
be clinically relevant.

Nevertheless this dierence in amplitude does not seem to

As expected a signicant depression in BIS values prior drug

administration and stimulus start was observed.

BIS SD was also analyzed, with no statistical dierence between prior and post drug
administration periods.

A statistically signicant dierence was observed globally in

response to laringoscopy/intubation and incision, with an increase in signal variability,

and also in study group 3, in response to the tetanic stimulation.
EMG was also signicantly depressed in all study groups when comparing baseline
values and the period pre-laringoscopy. This was expected since a muscle relaxant was
administered to allow laringoscopy/intubation.

A statistically signicant increase in

EMG activity was observed in all study groups for the laringoscopy/intubation stimulus.
EMG SD was assessed demonstrating that there was a signicant depression of EMG
SD in response to induction. Also a signicant increase in EMG SD was observed in all
study groups considering the laringoscopy/intubation stimuli. An EMG SD signicant
increase was also observed for the tetanic stimulus, in study group 2, and in response to
incision in study group 1, although not statistically signicant in the remaining study
groups.
SEF was globally depressed at induction, with a tendency of depression in response
to increasing remifentanil doses for the tetanic stimulus.
in response to induction.

TOTPOW was increased

No statistical dierences between median prior and post-

stimulation values of ASYM were observed.

104

4.2 Phase I Precise Stimuli Analysis

Table 4.5: Median amplitude response of cardiovascular variables for each stimulus and
remifentanil eect-site concentration (RemiCe) target considered in the study.

Original Data
Baseline/Pre-Laringo Laringoscopy Tetanic Incision
HR
RemiCe=2.0

-6,0

7,91

-0,95

0,31

RemiCe=3.0

-2,8

2,01

-1,45

1,08

RemiCe=4.0

-14,3

2,58

-1,42

-0,93

RemiCe=2.0

-30,6

11,63

0,09

6,49

RemiCe=3.0

-20,6

2,91

-0,20

4,22

RemiCe=4.0

-17,7

0,71

-3,01

6,43

RemiCe=2.0

3,9

-1,07

0,03

0,02

RemiCe=3.0

3,4

-0,39

-0,04

-0,53

RemiCe=4.0

1,1

0,36

0,06

-0,17

RemiCe=2.0

-45,1

19,77

-0,56

0,60

RemiCe=3.0

-30,9

6,64

0,64

5,15

RemiCe=4.0

-18,3

-5,64

-2,25

2,87

SBP

PPGA

ANSSI

Following the analysis of individual stimulus amplitude response, amplitude responses for the three stimuli considered in the study were analyzed within each study group.
Tables 4.5 and 4.6 present the median amplitude responses (baseline to pre-laringoscopy,
pre-laringoscopy to post-intubation, pre-tetanic to post-tetanic, and pre-incision to postincision) for each variable within each study group.
Statistical signicant dierences in HR amplitude response were observed considering the stimuli, being laringoscopy/intubation the stimulus with most pronounced
response, followed by incision. SBP amplitude responses were also dierent when comparing tetanic and incision stimuli.

Statistical signicant dierences were observed

between BIS SD amplitude response to laringoscopy/intubation, incision and tetanic

stimulus, with decreasing amplitude responses. EMG and EMG SD also presented signicant dierences between stimuli amplitude responses, with the response being more
pronounced for laringoscopy/intubation.

105

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.6: Median amplitude response of BIS related variables for each stimulus and
remifentanil eect-site concentration (RemiCe) target considered in the study.

Original Data
Baseline/Pre-Laringo Laringoscopy Tetanic Incision
BIS
RemiCe=2.0

-39,8

3,78

-0,09

0,55

RemiCe=3.0

-44,5

-4,75

-0,96

1,03

RemiCe=4.0

-42,7

-0,88

1,44

-1,40

RemiCe=2.0

-0,8

3,93

0,53

0,93

RemiCe=3.0

1,4

1,62

0,18

0,61

RemiCe=4.0

-0,5

1,59

1,12

1,67

RemiCe=2.0

-18,2

2,02

-0,19

0,33

RemiCe=3.0

-22,6

0,20

0,08

0,02

RemiCe=4.0

-18,3

1,04

0,03

0,22

RemiCe=2.0

-2,4

3,03

-0,30

0,42

RemiCe=3.0

-1,3

1,54

0,42

0,02

RemiCe=4.0

-1,7

2,27

0,62

-0,01

RemiCe=2.0

-7,3

0,18

-0,23

-0,05

RemiCe=3.0

-3,6

-0,35

-0,17

-0,01

RemiCe=4.0

-8,2

-0,10

-0,03

0,05

RemiCe=2.0

5,9

-0,25

-0,04

0,05

RemiCe=3.0

-0,7

0,16

-0,22

-0,32

RemiCe=4.0

4,9

-0,56

-0,05

-0,24

RemiCe=2.0

-1,0

-1,00

-0,51

-0,41

RemiCe=3.0

-2,0

0,76

-0,59

-0,67

RemiCe=4.0

-5,5

-0,02

-0,49

-0,87

BIS SD

EMG

EMG SD

SEF

TOTPOW

ASYM

Amplitude response for each stimulus was compared between the three study groups
considered, meaning according to the analgesic remifentanil dose. To perform this com-

106

4.2 Phase I Precise Stimuli Analysis

Table 4.7: Median normalized amplitude response of cardiovascular variables for each
stimulus and remifentanil eect-site concentration (RemiCe) target considered in the
study.

Normalized Data (%)

Baseline/Pre-Laringo Laringoscopy Tetanic Incision
HR
RemiCe=2.0

-11,8

10,8

-1,5

0,5

RemiCe=3.0

-3,6

3,0

-1,6

1,5

RemiCe=4.0

-15,9

3,6

-1,7

-1,4

RemiCe=2.0

-26,9

10,2

0,1

5,9

RemiCe=3.0

-18,7

2,7

-0,2

5,3

RemiCe=4.0

-14,8

0,6

-2,3

5,1

RemiCe=2.0

235,2

-66,3

1,7

2,2

RemiCe=3.0

224,4

-20,9

-4,3

-67,1

RemiCe=4.0

77,9

29,3

5,8

-14,0

RemiCe=2.0

-54,1

23,9

-0,8

0,7

RemiCe=3.0

-34,3

6,7

0,7

6,8

RemiCe=4.0

-21,7

-6,5

-3,1

3,3

SBP

PPGA

ANSSI

parison original and normalized amplitudes were used. Each stimulus was considered
separately and compared between groups using a non-parametric sign rank test for independent samples. Tables 4.5 and 4.6 present the median original values for each variable
analyzed, and Tables 4.7 and 4.8 present the median normalized (%) amplitude responses, by the baseline value prior drug administration (Equation 4.2), for each variable
analyzed.
Statistical dierences were found between amplitude responses observed in each
study group at induction (HR, PPGA and ANSSI) and in response to the stimulus
laringoscopy/intubation for the HR, SBP, PPGA, ANSSI, BIS, EMG and EMG SD,
with a tendency of decreasing amplitude responses for increasing remifentanil doses.

107

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.8:

Median normalized amplitude response of BIS related variables for each

stimulus and remifentanil eect-site concentration (RemiCe) target considered in the

study.

Normalized Data (%)

Baseline/Pre-Laringo Laringoscopy Tetanic Incision
BIS
RemiCe=2.0

-48,8

4,7

-0,1

0,6

RemiCe=3.0

-46,1

-5,1

-1,1

1,1

RemiCe=4.0

-46,5

-0,9

1,5

-1,6

RemiCe=2.0

-22,1

80,6

2,7

12,9

RemiCe=3.0

68,1

82,1

6,3

22,8

RemiCe=4.0

-13,8

53,4

32,9

74,5

RemiCe=2.0

-40,7

4,8

-0,3

0,5

RemiCe=3.0

-45,3

0,4

0,2

0,0

RemiCe=4.0

-40,6

1,9

0,0

0,4

RemiCe=2.0

-68,8

89,8

-9,8

14,2

RemiCe=3.0

-57,2

52,3

10,7

0,5

RemiCe=4.0

-59,6

63,3

17,7

-0,2

RemiCe=2.0

-31,5

0,8

-1,3

-0,2

RemiCe=3.0

-18,1

-1,5

-0,9

-0,1

RemiCe=4.0

-34,8

-0,4

-0,1

0,2

RemiCe=2.0

10,6

-0,4

-0,1

0,1

RemiCe=3.0

-1,1

0,2

-0,4

-0,5

RemiCe=4.0

8,5

-1,0

-0,1

-0,4

RemiCe=2.0

-1,8

-2,1

-1,1

-0,9

RemiCe=3.0

-3,9

1,6

-1,1

-1,3

RemiCe=4.0

-9,6

0,0

-0,9

-1,6

BIS SD

EMG

EMG SD

SEF

TOTPOW

ASYM

108

4.2 Phase I Precise Stimuli Analysis

4.2.2.2 Proposed Indexes in the Literature
CVI and ANSSI indexes were analyzed in this study in response to the three considered
stimuli, for the three dierent remifentanil doses.
ANSSI is a combination of PPGA and HR signals, as described in Chapter 1. It was
observed both signicant dierence of HR and PPGA in response to stimulation and to
dierent analgesic doses, so an adequate response of ANSSI was also expected. ANSSI
responded adequately to stimulation, with increasing values, statistically signicant for
the incision stimulus. Also ANSSI presented statistically signicant dierences between
amplitude responses at induction and laringoscopy/intubation stimuli, regarding the
remifentanil dose (decreasing amplitude responses with increasing remifentanil doses,
see Tables 4.3, 4.5 and 4.7).
Table 4.9: Median values of available CVI trends, for each stimulus (average pre and
post stimulation), divided according to the drug dose group:

remifentanil eect-site

concentration (RemiCe) of 2.0, 3.0, or 4.0 ng/ml.

Baseline Laringoscopy Tetanic

Incision
Pre
Post Pre Post Pre Post
CVI
RemiCe=2.0

2,47

4,80

1,75

1,54

1,23

1,81

RemiCe=3.0

5,94

3,41

1,09

1,16

1,01

1,24

RemiCe=4.0

6,57

6,61

1,52

1,53

1,39

1,79

CVI is a derivate measure of BIS and EMG variability, as described in Chapter 1.

Dierences in both EMG SD and BIS SD in response to stimulation were observed,
also with a tendency of decreasing amplitude response to increasing analgesic doses.
Being CVI a computed measure of BIS SD and EMG SD, the same pattern response
was expected for the CVI. Tables 4.4 and 4.6 do not present the CVI values since the
number of patients used in the analysis of CVI diered from the remaining analyzed
variables. This is due to a limitation imposed by the manufacturer which only allows
CVI calculation and display for BIS values below a certain threshold value. In some
of the collected cases CVI was not available during the analyzed periods due to the
occurrence of arousal periods, leading BIS values outside the admissible range for CVI
calculation. For the same reason, normalized values were not calculated since baseline

109

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

BIS was above the imposed manufacturer threshold, and CVI not available. Table 4.9
shows the CVI median values calculated for the available data (three patients excluded,
and variable window size to calculate averages). Statistical dierence between pre and
post intubation was found in study group 2, nevertheless, with decreasing values. This
was not expected since BIS SD, EMG and EMG SD were increased in this group following stimulation (see Table 4.4). Due to the reduced number of patients in each group
more statistical signicant dierences were not observed as expected for the previously
referred reasons. The imposed limitation by the manufacturer bias the observation of
dierences in the CVI index since important arousal episodes are forcefully excluded
from the analysis.

Nevertheless in the context of practical application this does not

pose as a problem, since the BIS index is already alerting to the arousal episode.

4.2.3 Discussion
Analysis of amplitude response to precise noxious stimulation of several physiological
variables was conducted. The stimuli employed in this analysis were stimuli considered
to have similar noxious intensity between patients, and therefore comparable between
individuals.
Bias factors that could introduce variations to amplitude responses, and inuence
the observed results were analyzed both for laringoscopy time and incision intensity.
No statistical dierence was found between groups, nevertheless an increase of incision
intensity was observed in study group 3. Although it was not signicant, regarding the
reduced number of patients in each study group this may account for the observation
of increased responses to incision of some of the analyzed physiological variables, even
with a higher remifentanil dose, when compared to the remaining study groups.
Of the physiological signals analyzed in this study, some responded both to stimulus,
and remifentanil attenuation. These variables were HR, SBP, PPGA, BIS SD, EMG
and EMG SD.
Two proposed indexes in the literature were also assessed to verify if they responded
adequately to stimulus and drug attenuation. Being these indexes post-processed measures which include the previously referred physiological signals responding to stimulation and drug attenuation, it was expected that they too presented adequate responses
to both factors.

This was observed for ANSSI, nevertheless the same tendency was

not possible to observe for the CVI due to limitations in the index presentation for the

110

4.2 Phase I Precise Stimuli Analysis

user. Nevertheless considering the physiological signals used in the indexes calculus, an
adequate response to noxious stimulation and drug attenuation for both is expected.
CVI limitation imposed by the manufacturer blocks the observation of important
arousal events that may be connected to noxious stimulation. For the purpose of this
study missing CVI values limited the observation of signicant dierences in the index,
however in practical aspects this limitation does not have an important impact on the
anaesthesiologists activity since BIS is used in the detection of the arousal episode, and
the clinician advised to take action.
Although dierences in ANSSI response to noxious stimulation were found, there
are some variables that were not adequately controlled in this study to guarantee more
reliable results (e.g.
perature).

PPGA is conditioned by patient's volemia and peripheric tem-

These variables should have been monitored and controlled to guarantee

comparability between patients [86]. Although this was not taken into account, indication of the information contained in this variable regarding the Noc/ANoc balance was
found, and in future studies these issues should be addressed and corrected.
In general the variables that have been linked in previous studies to noxious activation [16, 17, 20, 77, 86] presented signicant results in response to a noxious stimulus,
and for the amplitude response attenuation of the analgesic drug dose. Therefore, these
variables were concluded to contain information on noxious activation, and to provide
indicators of the patient's Noc/ANoc balance. This is in accordance to previous results
presented in the literature, and described in Chapter 1.
Statistical signicant dierences in amplitude responses considering the three noxious stimuli applied in this phase of the study were observed. Laringoscopy/intubation
was the stimulus with more pronounced response (increased amplitude response to stimulation), followed by incision, and tetanic stimulus.

This result is corroborated by

results presented in the literature [168170], and by the observed tendency described
in Chapter 3, where laringoscopy and intubation were considered to be high intensity
stimuli (6.81 and 6.16 in the Rasch scale) even when compared to some incision types
(see Figure 3.7), and tetanic stimulation (4.76).
The main objective of this phase of the study was to identify variables linked to
noxious activation, that adequately translate the impact of the analgesic drug and
stimulus intensity, this was achieved in accordance to the tendencies described in the
literature.

111

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

4.3 Phase II Maintenance Analysis - Steady-State Detection

After identifying the variables more informative on noxious activation in the study
phase I, the second phase was focused on the evaluation of drugs and stimuli intensities
combination, and their impact on the measurable output signals related to noxious
activation.
The proposed clinical protocol was designed to produce a variety of drugs' combinations in such a way information on drugs' relations and physiological signals could be
extracted, both in dynamic and

Steady-State

(SS) conditions. This approach is useful

since it allows the identication of models, to translate the patient's response to changes
in drugs' doses and stimuli intensities.
Throughout daily activities body natural balance suers constant interferences, automatically and unconsciously adjusting physiological variables to maintain homeostasis.

During general anaesthesia, the patient is subjected to the action of drugs and

noxious stimulation, which interfere with the natural balance and also trigger natural
compensatory mechanisms (see Figure 1.20). These relations were investigated to explain the involved mechanisms in response to drugs and stimuli inputs, and obtain a
measure of the patient Noc/ANoc balance.
In a rst approach a SS measure was extracted from the input signals (drugs' doses
and stimulus intensity), and output measurable physiological signals related to noxious
activation (BIS, EMG, HR, SBP, PPGA, as presented in the previous section), in order
to evaluate patient's homeostasis [174177].

In a second approach the collected data

were used in the search of dynamic models that may explain the individual responses
to noxious stimulation and drugs' doses, observed in the output signals.
Most of the indexes proposed in the literature to assess the patients' Noc/ANoc
balance are based on measures of physiological signals variability associated to noxious
activation and their normalized values: BIS, EMG, HR, PPGA, SBP [17, 20, 77, 78].
Some of the methods proposed so far present limitations in their application and solely provide information on the variation of the physiological signal, and not on the
direction of the occurring change. To respond to this problem a dierent signal processing technique was applied, adequate to non-stationary signals, capable of assessing
and translating the change and shift direction of the physiological signals associated to

112

4.3 Phase II Maintenance Analysis - Steady-State Detection

noxious activation, both individually and in a normalized combined index of stationarity.
During general anaesthesia the clinician aims at maintaining the patient in a comfortable and stable condition (homeostasis), adjusting drug doses according to patients'
needs, avoiding over/underdosing and its consequences. As described earlier, anaesthesia is composed by three main components, controlled by the clinician through the
proper administration of three distinct drugs, which may interact to produce a certain
state in the patient (see Figure 1.18).

Following general anaesthesia induction, the

hypnotic dose may be adjusted using information derived from EEG based indexes and
clinical observation (see Table 1.1) to maintain an adequate unconscious state, avoiding
awareness or overdosing, and possible adverse consequences [12, 13, 4042]. Considering
anaesthesia as a combination of the three components, if a certain unconscious state
is maintained, variations in the patient's state may be considered to derive only from
changes in the stimulation intensity. Bouillon et al. [167] present anaesthesia opioidhypnotic interaction as an hierarchical model (see Figure 3.2), where pain is blocked by
the use of opioids (attenuation), and then the attenuated pain is projected to the cortex
where hypnotics act. If one considers this hierarchical model it is possible to infer that
changes in the patient's physiological signals are a consequence of noxious stimulation
variation, and therefore closely linked to the Noc/ANoc balance. This notion is applied
by anaesthesiologists in their daily activities, detecting changes in the physiological
signals' trends, and adjusting drugs' doses to compensate the uctuations, acting as
controllers to maintain patients in optimal SS conditions. A measure of combined SS
for dierent physiological signals was obtained.

4.3.1 Wavelet Analysis and Steady-State Detection

When modeling a process it is important to obtain SS information of the process overall
performance.

Knowing this, it is of great relevance to detect SS conditions in both

inputs and outputs of the system. In the specic case the patient is the system, drugs'
doses and stimulus intensity the inputs, and the measurable physiological signals related
to noxious activation the outputs.
TCI of propofol and remifentanil was used for drugs' administration, and therefore
SS of drugs is equivalent to SS of eect-site concentration estimated by the PKPD models. A simple arithmetic rule was initially applied to obtain input drugs' SS [174176],

113

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

based both on the target (Ct) set by the user, and theoretical eect-site concentration
(Ce), as presented in Equation 4.3.

|P ropCt P ropCe| + |RemiCt RemiCe| = DIFP &R

If

DIFP &R 0.05

considered.

(4.3)

for more than one minute than SS for the input drugs was

It should be highlighted that this arithmetic rule is computed from two

dierent drugs and therefore dierent units, with no physical meaning, nevertheless the
empirical limit set to detect SS is in accordance with the TCI administration scheme.

Figure 4.3: Steady-state detection in the input drugs' eect-site concentration using the
arithmetic rule.

Figure 4.3 shows the calculus of

DIFP &R

for input SS detection in one data set,

using the rule presented in Equation 4.3.

Following detection of system SS for the inputs, the output measurable eects must
be analyzed. The output signals need more robust analysis techniques than those used
for drugs' SS detection. A wavelet based approach was employed. Wavelets are currently
widely used in several elds [178181] due to their properties, such as noise removal and
feature detection. The method used was initially proposed by Jiang et al. [178], and is
summarized here with the correspondent and adequate modications for the problem
posed in this thesis.
The concept of multi-scale representation is to have the signal as a limit of successive
approximations (Equation 4.4).

f (t) = f0 =

X
iI0

114

c0,i 0,i =

4.3 Phase II Maintenance Analysis - Steady-State Detection

=

c1,i 1,i +

iI1

d1,k 1,k = ... =

kk0

cj,i j,i +

iIj

dj,k j,k +

kkj

cJ,i J,i

J X
X

dl,k l,k =

dj,k j,k

(4.4)

j=1 kki

{z
}
|
High F requency
Components Scale 1 to J

| {z }
Low F requency
Component Scale J
j,i and j,k

l=1 kkl

iIj

where

j X
X

are the discretized dyadic scaling and wavelet functions, respectively.

Due to the known

Wavelet Transform

(WT) properties, the noise can be reduced

using a soft threshold technique over the WT modulus (Equation 4.5).

dj,k =
where

0
|dj,k | j
, 1 j J, 0 k Kj
sign(dj,k )(|dj,k | j ) |dj,k | > j

(4.5)

is the threshold value at scale j.

At scale j=1 the WT modulus is dominated completely by noise and the threshold
value

can be assigned as the mean of the modulus maxima. The following threshold

values are calculated as given by Equation 4.6.

j = 1 2(j1)/2 , 2 j J

(4.6)

Rapid changes in the signal may be detected using WT, since these changes in the
signal are identied as a maximum in the corresponding WT.
Two points, within a dened interval

tp ,

with a WT maximum and opposite signal

identify an abnormality (Equations 4.7 and 4.8).

|Wf (p1 )| and |Wf (p2 )| T1

(4.7)

sign(Wf (p1 )).sign(Wf (p2 )) < 0, p2 p1 tp

(4.8)

After detecting an abnormality, the duration of this event is determined as the

nearest point

ta

to the left of

p1

and the nearest point

rules in Equations 4.9 and 4.10.

115

tb

to the right of

p2

that satisfy

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Threshold values

|Wf (ta )| and |Wf (tb )| T2

(4.9)

|Wf (ta 1)| and |Wf (tb + 1)| < T2

(4.10)

T1

and

T2

are computed from historic data as given in Equation

4.11.

T1 = 31 w,
where

T2 = w

(4.11)

w is the standard deviation of the wavelet modulus of historic measurements with

noise eliminated, and

A SS index
status and

=1

an adjustable parameter around 1.

was used to measure the degree of SS of each signal:

= 0 for unstable

for SS.

When trying to determine SS of a signal it is important to distinguish zero-crossing

points of
WT

Wj f (t).

W Wj f (t),

To do that the WT on

Wj f (t) is performed, obtaining a second order

proportional to the second derivative of

The calculus of

(t)

f (t).

was based on these notions, detecting rapid changes in

and distinguishing zero-crossing points with

W WS f (t)

WS f (t)

values. S is the characteristic

scale, meaning the proper scale to analyze the WT where the WT represents the process
variations properly (Equation 4.12).




S = j = int log2 + 0.5

ts
where

ts

is the sampling interval, and

(4.12)

the response time constant.

SS index is dened according to the following rules:

if

|WS f (t)| > Tu

then

(t) = 0,

where

Tu

is the identication WT modulus

threshold for unsteady status;

if

|WS f (t t)| < Ts

then

(t) = 1,

threshold for steady status, and

where

Ts

is the identication WT modulus

a long enough time interval to identify SS;

to detect zero-crossing points, the second order WT is used.

and

|W WS f (t)| < Tw

then

(t) = 1

where

Tw

|WS f (t)| < Ts

is the second-order WT modulus

threshold to identify zero-crossing point in the WT.

116

If

4.3 Phase II Maintenance Analysis - Steady-State Detection

In other cases relations in Equations 4.13, 4.14, 4.15 and 4.16 are used.

(t) = [(t)]

(4.13)

(t) = |WS f (t)| + |W WS f (t)|

(4.14)

|W WS f | Tw
0
(|W WS f | Tw )/2Tw |W WS f | ]Tw , 3Tw [
=

1
|W WS f | 3Tw

(t) Tu
0
[(t)] Ts < (t) < Tu
(t) =

1
(t) Ts
where

is a smooth transfer function with range [0, 1].

(4.15)

(4.16)

given in Equation 4.17 was

used.





1
x Ts
(x) =
cos
+1
2
Tu Ts
The thresholds

Tu , Ts

and

Tw

(4.17)

were calculated from historic measurements, after

manually selecting SS periods, and performing WT and second-order WT. Then the
standard deviation of the WT modulus
modulus

W Wf

and the median of the second-order WT

are obtained, and the thresholds dened by Equation 4.18.

Ts = Wf ,
where

Wf

Tu = 32 Wf ,

Tw = W Wf

(4.18)

is an adjustable parameter around 1.

Figure 4.4 shows the algorithm steps in the SS detection for both input and output
signals. Although for the collected data in this study, input SS detection as described
in Equation 4.3 is adequate (dashed rectangle in Figure 4.4), to increase the method
robustness and generality for it to be fully applied to dierent data sets, wavelet SS
detection was applied directly to the drugs' Ce and an input SS index was obtained
similar to the output SS combined index previously described [177]. The methodology
was also applied directly to the infusion rates, nevertheless for brevity, results will not
be presented here.
The quadratic spline wavelet function and scaling function described by Mallat and
Zhong [181] were used. Tables 4.10 and 4.11 present the nite impulse response lters
correspondent to the quadratic spline wavelet, and the normalization coecients used to

117

118

based algorithm.

steady-state index as summarized. Dashed line rectangle presents the initially used input detection rule, replaced by the wavelet

concentrations (Ce) as inputs, and output physiological variables entering the wavelet based algorithm to determine the combined

Figure 4.4: Diagram presenting the algorithm steps: the patient is the system, with propofol and remifentanil drugs' eect-site

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

4.3 Phase II Maintenance Analysis - Steady-State Detection

Table 4.10: Finite impulse response of the lters H, G, K and L that correspond to the
quadratic spline wavelet, reproduced from [181].

-3

0,0078125

0,0078125

-2

0,054685

0,046875

0,171875

0,117188

-1

0,125

0,375

-2,0

-0,171875

0,65625

0,375

2,0

-0,054685

0,117188

0,125

-0,0078125

0,046875

Table 4.11: Normalization coecients

j = 1, 0.

0,0078125

for the quadratic spline wavelet. For

j > 5,

Reproduced from [181].

1,50

1,12

1,03

1,01

1,00

compute the wavelet transform. For each signal considered as output related to noxious
activation this method may be applied, with the adequate threshold detection.

The

signals analyzed in the previous section (phase I), and found to be related to noxious
activation were BIS, EMG, HR, SBP and PPGA, nevertheless an additional signal was
introduced in this analysis, the respiration rate (RespR). Although ventilation in all
collected cases was mechanically maintained, changes in the ventilation conditions may
occur by the anaesthesiologist choice. This may alter the equilibrium state of the patient
and induce changes in other physiological signals. Also, if the patient starts to recover
from neuromuscular blockade, and moves, this may be translated into spontaneous
ventilation, which may be detected and linked to arousal episodes. The following results
include data for the respiration rate, nevertheless when comparing the response to
precise noxious stimuli, the respiration rate was removed from the analysis, since it
would introduce bias due to the initiation of mechanical ventilation concomitant with
the laringoscopy/intubation stimulus.
A training set of ve patients was used to adjust the wavelet thresholds of the

119

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.12: Estimated thresholds based on historic data for each signal analyzed, in a
training set of ve patients.

Wf

WWf

Remifentanil Ce

0,08

0,06

0,02

0,00

Propofol Ce

0,04

0,05

0,01

0,00

Stimulus

6,35

0,32

0,05

0,00

BIS

16,27

3,04

2,90

0,78

EMG

23,25

10,49

0,33

0,11

HR

10,71

1,87

0,36

0,12

SBP

34,67

3,85

0,93

0,27

PPGA

1,68

0,18

0,11

0,03

RespR

4,64

0,45

0,22

0,10

eight analyzed physiological signals:

propofol Ce, remifentanil Ce, BIS, EMG, HR,

SBP, PPGA and RespR. For each data set, the output signals were visually inspected
and sequences considered as in SS, selected to extract the thresholds later used in the
SS index processing, shown in Table 4.12. Following the method was applied to each
input and output signal individually, and individual SS indexes (Signali ,

i = 1, ..., 8)

combined to produce the combined input (In ,

m = 5)

n = 3)

and output (Out ,

SS

indexes (Equations 4.19 and 4.20).

In (t) =

1X
Signali (t)
n

(4.19)

i=1

Out (t) =

1 X
Signali (t)
m

(4.20)

i=1

To inspect if the method was accurately detecting SS periods, the sequences considered as in SS used to dene the detection thresholds were analyzed. Of the total time
used in the threshold detection 91% was found to be in SS (Signali

0.9, i = 1, ..., 8),

demonstrating the method accuracy in detecting steady-state sequences.

The SS index
value: if

(t)

(t) 0.9

was used to extract SS periods in the output, by thresholding its

for more than one minute than the signals were considered to be in

SS, and the patient in homeostasis.

In Figure 4.5 one of the output signals SS individual index is presented in detail
to better illustrate the method, with the original signal, the extracted trend using the

120

4.3 Phase II Maintenance Analysis - Steady-State Detection

Figure 4.5: Detailed representation of steady-state detection of the systolic blood pressure trend, in one of the collected data sets.

wavelet technique, and the processed individual homeostasis index (Signali ).

Figure

4.6 shows the developed homeostasis index (Out ) for the same collected data set.

Table 4.13:

Percentage of total time (total of 31 patients), that each signal analy-

zed in this study phase was found to be in steady-state (SS) conditions (data as
meanstandard-deviation).

Time Percentage in SS (N=31)

Remifentanil Ce

74,17,8

Propofol Ce

79,56,6

Stimulus

90,43,1

In

59,58,5

BIS

75,09,7

EMG

67,07,6

HR

32,617,1

SBP

40,319,3

PPGA

59,428,6

RespR

88,49,6)

Out

24,314,2

The methodology was applied to all data sets (N=31, Table 2.2), data presented

121

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Figure 4.6: Representation of the input and output steady-state indexes. The bottom
trend may be interpreted as an homeostasis index of the subject considering BIS, heart
rate, systolic blood pressure, pulse wave amplitude and respiration rate.

122

4.3 Phase II Maintenance Analysis - Steady-State Detection

as meanstandard-deviation. Besides information regarding the subjects' homeostasis
state (Out ), the implemented methodology allows the inspection of whether or not the
designed clinical protocol (presented in Chapter 2) was ecient in the production of
both periods in SS and dynamic conditions.

It was observed that in the total data

set, signicant periods in SS and non-SS conditions were achieved. Table 4.13 shows
the time percentage that each physiological signal analyzed was in SS ((t)

0.9)

and

dynamic conditions, and also of input SS (In ) and output signals SS (Out ). The total
time analyzed, considering the 31 data sets, was of 6684 minutes, and average duration
per case of 21576 minutes.
Input/output simultaneous SS periods may also be extracted using this technique.
Periods lasting more than one minute of input and output simultaneous SS were extracted, and may be used in the construction of dynamic surfaces relating drugs, stimulus
and output measurable eect.
It was observed that input drug's SS was not always followed by output SS. This
is a demonstration that dierent factors interfere with the patient's homeostasis, being
noxious stimulation one of these factors. Also, it should be highlighted that SS does
not necessarily mean the patient is in optimum conditions, but only that the patient is
stable around a determined state of operation.
Table 4.14: Median values of the proposed homeostasis index (HI): average value in the
periods baseline, prior and post-stimulation, for each remifentanil eect-site concentration (RemiCe) group.

Baseline Laringoscopy
Pre
Post

Tetanic
Pre Post

Incision
Pre Post

HI
RemiCe=2.0

54,64

80,73

51,01

76,18

86,33

84,60

78,33

RemiCe=3.0

52,95

75,83

67,71

78,50

74,45

77,56

84,11

RemiCe=4.0

52,89

72,94

65,34

80,72

79,99

75,44

76,73

The proposed homeostasis index was calculated for all data sets, and the response to
precise stimulus in study phase I inspected to evaluate if the proposed index responded
adequately to the introduction of noxious stimulus, and dierent analgesic drug doses.
Tables 4.14 and 4.15 present the median values of the homeostasis index before and
following stimulation of the precise stimuli considered in study phase I, and amplitude
responses to the stimuli. The number of data sets used in this comparison is the same

123

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.15: Normalized amplitude responses of the proposed homeostasis index (HI) to
the precise stimuli considered in study phase I, for each remifentanil eect-site concentration (RemiCe) group.

Original
Baseline/Pre-Laringo Laringoscopy Tetanic Incision
HI
RemiCe=2.0

23,7

-24,64

10,22

-3,68

RemiCe=3.0

26,6

-12,22

-4,02

-3,84

RemiCe=4.0

22,0

-10,05

-4,91

5,53

Normalized (%)
Baseline/Pre-Laringo Laringoscopy Tetanic Incision
HI
RemiCe=2.0

46,2

-43,1

16,5

-6,2

RemiCe=3.0

44,4

-20,4

-6,7

-6,4

RemiCe=4.0

42,8

-18,5

-7,5

11,2

as described in phase I in section 4.2, 26 patients (see Table 4.1). Similarly to what
was presented in the phase I results (see subsection 4.2.2), non-parametric tests were
employed in the comparison of amplitude responses and analgesic impact. Statistical
dierences were found between pre and post intubation periods, with a decrease in the
homeostasis index, in a dose-dependent manner.

Also the decrease was signicantly

larger in the laringoscopy/intubation stimulus, when compared to tetanic and incision

amplitude responses. This is in concordance with the results obtained for each variable
individually in the phase I analysis.

4.3.2 Homeostasis Index

The proposed index may be used as a measure of patient's homeostasis, meaning of the
equilibrium and stability of the physiological signals associated with noxious activation.
This method allows the quantication of the changes as well as the detection of the
changes direction for each signal individually. The index was implemented to obtain online measures and automatically inform the anaesthesiologist on the occurring changes
in the patient, for a fastest detection and prediction of possible modications in the
patient's state.

Monitored signals were analyzed individually and in conjunction to

124

4.3 Phase II Maintenance Analysis - Steady-State Detection

Figure 4.7: Schematic representation of the signal processing steps to obtain the individual and combined steady-state indexes, from data collection, to signal pre-processing
unit, and homeostasis index presentation (individual and combined), as implemented
in the developed tool for on-line homeostasis assessment.

translate the patient's homeostasis state, as represented in Figure 4.7.

The developed software was designed using the GUIDE tool in Matlab, including
the steps from data pre-processing as described in Chapter 2 (waves delineation and
detection), to the individual SS analysis of the physiological signals found to be linked
to noxious activation in the study phase I, as described earlier in this chapter. The tool
displays every second an actualization of the patient's state, taking in consideration 15 s
of past values for each physiological signal. The individual SS indexes are presented for
each variable analyzed (drugs' Ce, BIS, EMG, HR, SBP, PPGA and RespR), and as a
combination of input (propofol and remifentanil Ce) and output variables (BIS, EMG,
HR, SBP, PPGA, RespR), for the combined

In

and

Out

SS indexes, respectively. It

allows the visualization of each variable value every second, and also a graphical repre-

125

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

sentation of 10 minutes windows. The individual SS indexes are numerically presented
every second. Also the combined SS indexes are numerically presented with 1 s actualizations and graphically in 10 minutes windows. This allows a fast detection of unstable
conditions, with fast graphical interpretation of the changes. It is also designed to detect infusion rates steady-state, instead of Ce, to allow its use in clinical setups that
do not use TCI. It should be highlighted that using infusion rates instead of eect-site
concentration estimates may lead to increased delay between input steady-state, and
output steady-state, due to the disregard of equilibration times for Cp and Ce.

For

example, periods when infusion rate is null and in steady-state, due to a decrease in
target eect-site concentration, does not correspond to steady-state in Ce, since the
drug is being washed-out of the system.

Figure 4.8: Designed software for on-line assessment of the developed steady-state indexes: individually for each input (propofol and remifentanil Ce), output (bispectral
index, electromyography, heart rate, blood pressure, photoplethysmography wave amplitude, respiration rate), and combinations of input and output signals (In and

Out ).

Data presented in this example is from one of the patients of the study, simulating an
on-line assessment.

Figure 4.8 shows the developed tool, using data from one of the collected data sets,
to simulate the on-line implementation. Besides numerical information on the SS index
value, the software also processes the information on the direction of the variable shift,

126

4.3 Phase II Maintenance Analysis - Steady-State Detection

using a color code: green for a sudden decrease, red for a sudden increase and orange
for a variability increase.
Although there is important information in the detection of the patient's global SS,
and more importantly in the detection of non-SS conditions, the maintenance of SS in
the conjunction of the clinical signals analyzed, it is not guarantee that the patient is in
optimum conditions, only that he/she is in equilibrium around some state of operation.
Some pattern scenarios may be studied with the aid of experienced anaesthesiologists,
to introduce in the system alarms that would reproduce the expert reaction to pattern
changes. For example, if the heart rate is increasing, followed by an increase in blood
pressure, it means that there is an active response to stimuli, and an alarm would be
triggered. Such a system may aid in a more rapid detection of critical situations, and
advise the anaesthesiologist on the recommended action to take.

4.3.3 Surface Modelling: Input/Output Steady-State

The production of SS valid models is important in the description of a system permanent state operation.

In this case the purpose is to evaluate the impact of dierent

drugs' combination on the physiological signals associated to noxious activation. The

importance of evaluating the system permanent operation behavior resides in the relations that may be inferred from this information. The impact of drug doses in the output
signals is clearly important information to extract from the collected data, nevertheless
one must never forget that the impact of the drug dose in the output variables is intrinsically connected to the stimulation intensity the patient is subjected to, and to extract
information from these variables one needs to look at the problem for each stimulus
individually.
The methodology here described may be used to obtain periods of SS both on input
and output variables, and allow the construction of SS valid models (drug interaction
surface models), relating each drug dose to the observed eect, considering dierent
stimulation intensities.
Figure 4.9 presents the combined periods of input/output SS periods, for one of the
measured output signals (HR), showing the intra and inter patients variability in response to dierent drugs' combinations and noxious stimulus intensity. The information
gathered through the SS detection routines may be used not only in the assessment of
drugs' impact on the measurable output signals, but also to estimate noxious stimulus

127

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Figure 4.9: Combined periods of steady-state (SS) input/output in all data sets: relation
between drugs' eect-site concentrations (Ce) and average heart rate (HR). The stimuli
intensities are represented in dierent colors, with white representing periods without
stimulation (0 in the Rasch scale), and increasing gray intensity with increasing stimulus
intensity.

intensity, based on the drug dose impact observed in the output signals, conditioned
by the stimulation intensity. This is similar to the process described by Bouillon et al.
[167]. Although estimating noxious stimuli intensities through this method would be
more direct, similar to what was described for the pre-intensities estimation, more data
for all stimuli would need to be collected in steady-state conditions, which arises practical and ethical issues. This issue may be surpassed using the developed Rasch scale
in Chapter 3, taking into account anaesthesiologists' experience, that was shown to be
related to the estimated pre-intensities in the pharmacological studies (see Figure 3.12
and Table 3.3).

4.3.4 Discussion
In this section a new measure of patients' homeostasis state is proposed. This measure
may be viewed as a method to evaluate patients' Noc/ANoc balance in response to
changes in drug doses and noxious stimulation.

The variables found to be linked to

noxious activation in the previous section were in this section employed in the calculus

128

4.3 Phase II Maintenance Analysis - Steady-State Detection

of the combined index.
Individual SS indexes for each physiological signal considered were combined in the
translation of input and output SS, later implemented in a tool to translate on-line the
input and output changes continuously.
In addition to the use of the technique in the translation of patients' homeostasis,
it allowed the inspection of the protocol utility in the obtainment of periods in dynamic and SS conditions. Both contain information useful for the understanding of the
relations between the dierent physiological signals linked to nociception. It was observed that considerable periods in dynamic conditions were obtained, which brings useful
information for the following section.
The technique was employed in the search of combined periods of input and output
SS, which may allow the construction of surface models translating the relation between
hypnotic and analgesic drugs and impact on the output signals, under SS conditions, and
under the inuence of dierent stimuli intensities. The information gathered through
this process allows the quantication of drugs synergy eects on the measurable output
variables, similar to what has been done in previous pharmacological studies to estimate
stimuli intensities, for example. Nevertheless more data would have to be collected in
order to obtain information for all considered stimuli.
The proposed homeostasis index responded adequately to noxious stimulation, and
presents itself as a new multivariate method to assess patients' Noc/ANoc balance, when
analyzed within clinical context. Although the method allows the discern of patients'
stability or instability the main disadvantage of the method resides in the fact that
stability does not necessarily mean a good Noc/ANoc balance state. The diculty of
this method is to dene the optimum state of operation, since a patient may be stable
and nevertheless in distress. On the other hand the main advantage of the method is
the possibility of automatically detecting changes in the patients' condition, and alert
the anaesthesiologist for a possible change in drug dose to conduct the patient back to
stability, according to clinical assessment.
The SS index proposed contains important information regarding patient homeostasis state, nevertheless this information is not enough to fully describe the patient
Noc/ANoc balance. The next section focus on the Noc/ANoc balance assessment from
a dierent perspective, trying to explain variations in the signals linked to noxious activation, by a dynamic model estimating the perceived stimulus by the patient, and
therefore a location in the Noc/ANoc balance.

129

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

4.4 Phase II Maintenance Analysis - Physiological Modelling

Results presented in the previous section allowed an insight on the adequacy of the designed clinical protocol in the collection of data in dynamic conditions, and subsequent
search of dynamic relation models between the collected variables. This was shown in
Table 4.13. The SS valid models are useful to understand the continuous operation mode
of a system, but information is lost using this approach. As referred before, the dierent
physiological variables are intertwined in a process, conditioned by compensating mechanisms. The motivation of this section of the work is to determine dynamical models
which may be used in the description of the human system in response to both stimulus
and drugs' attenuation, and that produce an index of perceived stimulus (Noc/ANoc
balance), valid for the population of this study.

Figure 4.10: Representation of the attenuation and depression eects of the analgesic
drug on a stimulus of known intensity [78, 167], with the introduction of an additional
depression curve.

An hierarchical approach may be used in the translation of the opioid drug eect
in stimulus attenuation, followed by the hypnotic eect in hypnosis maintenance, under
the incidence of the unblocked stimulus, as noted before [167].

Figure 3.2 shows the

hierarchical model proposed, describing the analgesic action as an attenuation of the

stimulus incoming from the periphery for central processing, and later projected to the
frontal cortex, were it interferes with the conscious state of the patient. Although the
hierarchical model is an appealing approach to describe both hypnotic and analgesic
interaction, the proposed model is not taking into consideration the depressive eect
that the analgesic overdose may introduce.

In order to fully describe the Noc/ANoc

balance it would be optimal if there was a combination of both stimulus attenuation and

130

4.4 Phase II Maintenance Analysis - Physiological Modelling

physiological depression caused by excessive drugs' administration. This would result in
the construction not only of the stimulus block curve but also the depression curve, as
represented in Figure 4.10, being the optimum state of operation the point of maximum
stimulus attenuation, prior to depression (optimum point of Noc/ANoc balance). The
concept behind this approach is to better understand the impact of drugs' changes
in the natural compensation mechanisms, not only in response to dierent stimulus'
intensities, but also considering drugs' attenuation/depression, in the construction of a
Noc/ANoc balance index as proposed in Figure 1.19.
Although the approach represented in Figure 4.10 would be more adequate, data
collected in the scope of this thesis was not intended to produce depression data (interruption rules described in Chapter 2, Table 2.1), nor it would be ethically acceptable to
produce intentionally data to obtain the depression curves for each variable considered
in phase II of the study.
Previous studies [78, 167] use the estimation of stimulus pre-intensities based on
pharmacological studies, and analgesic needs to block precise stimulus. Estimated preintensities were employed in [78] to produce a surgical stress indicator, based on the
estimation of stimulation intensity (pre-intensities) and opioid dose. The same approach
was employed in this thesis.

The pre-intensities of the remaining stimuli were extra-

polated based on the results of the Rasch analysis, presented in Chapter 3, since the
relation between the estimated pre-intensities and the Rasch score was shown to be close
to linear (see Figure 3.12 and Table 3.3). Knowing the stimulus pre-intensities, conditions are gathered to assess the analgesic impact (attenuation) on each noxious stimulus.
Table 4.16 shows the extrapolated values of pre-intensities using the estimated Rasch
scale developed in Chapter 3. Following, estimated pre-intensities (preopioid intensity )
were employed in the assessment of the perceived stimulus, using Equation 4.21, were

opioid50 =1.01 ng/ml

tailed description).

of remifentanil, and

=0.72

(see subsection 3.2.3 for more de-

This step was achieved using both the registered noxious stimuli

throughout the procedures shown in Table 2.5 (scale), and the estimated remifentanil
eect-site concentration (opioid), to obtain a new variable, combination of these two,
the attenuated stimulus (postopioid

intensity ).

postopioid intensity=

preopioid intensity 1

opioid +(opioid

131

opioid

50 preopioid intensity)

(4.21)

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Table 4.16: Variable

scale identies each stimulus with a code number.

Stimulus inten-

sities estimated and translated in the Rasch scale, as well as extrapolated values for the
pre-intensities, considering the linear relation shown in Figure 3.12. Table entries presented in bold correspond to the stimuli that have pre-intensities estimations; original
values presented between parenthesis.

Code (scale)

Rasch Scale

Extrapolated Pre-Intensities

0,00

0.00

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

4,58
4,99
5,55
4,76
4,49
4,17
7,11
7,20
4,81
6,43
5,27
6,16
6,81
5,07
5,07
4,41
5,36
5,16
4,81
2,35
4,56
4,64
6,15
3,80
3,88
3,37
4,41
4,46
5,65
5,07
5,21
3,74
4,64
4,32

0,99
1,08 (1,25)

1,20
1,03 (0,80)

0,97
0,90
1,53
1,55
1,04
1,39
1,14
1,33 (1,57)
1,47 (0,83)

1,09
1,09
0,95 (1,00)

1,16
1,11
1,04 (0,80)

0,51
0,98
1,00
1,33 (1,50)

0,82
0,84
0,73
0,95
0,96
1,22
1,09
1,12
0,81
1,00
0,93

132

4.4 Phase II Maintenance Analysis - Physiological Modelling

It should be highlighted that the problem found in Chapter 3, regarding the dierentiation between no stimulus and the stimulus of the tube, without surgical stimulation,
was here corrected, using the proposed value in [78] of 0.25 to the pre-intensity of the
tube stimulus. This was done by replacing the periods were the stimulus was 0, between
intubation and extubation events, by the tube in pre-intensity. Huiku et al. [78] also
propose a method to the estimation of the patients' stress, subjected to noxious stimulation and opioid attenuation, the
of

preopioid intensity ,

Total Surgical Stress

(TSS), applying the estimation

but estimating the opioid depression as given by Equation 4.22.

T SS = preopioid intensity

Remif entanil Ce
3 ng/ml

(4.22)

where the relative weights of TSS were determined so that the clinical range of remifentanil matched the surgical range of stimulation in their study.
The estimated

postopioid intensity

and TSS are represented in Figure 4.11, with

the original Rasch scale stimulus intensity trend, and the remifentanil Ce in one of the
patients of the study. It may be observed that the new

postopioid intensity

and TSS

incorporate information on the stimulus intensity and attenuation provided by remifentanil. Although the trend is similar using both approaches, since the TSS is a simple
arithmetic rule based on a normalization of the remifentanil doses by the average expected, it allows the articial introduction of depression, evident when the administration
of remifentanil is initiated, and there is no stimulus.

Nevertheless, this is a forceful

introduction of depression, based on normalized remifentanil Ce, using an arithmetic

rule.

Both perceived stimulus indicators were employed in the following analysis, to

evaluate which one of these descriptors better correspond to the physiological signals
objectively evaluated in the patient.
Depression information is important if one imagines a structure to model and predict
the patient's state in a timespan (advisory system), trying to predict the physiological
variables future value (see Figure 4.12), and an estimate of the perceived stimulus,
which may be seen as an estimate of the Noc/ANoc balance, since the objective of the
anaesthesiologist is to obtain maximum stimulus attenuation, without depression.
The impact that the hypnotic drug has on the dynamical relations represented in
Figure 4.12 should be highlighted.

Although the hierarchical model considers that

the hypnotic only acts on the hypnosis state, suering interference of the attenuated
stimulus, the hypnotic and analgesic drugs exhibit a synergistic relation [167], and the

133

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

Figure 4.11: From top to bottom: stimulus intensity trend, using author's annotations
and intensities estimated by the Rasch analysis;

preopioid intensity

estimated based

on the Rasch scale extrapolation; estimated remifentanil eect-site concentration (Ce

in ng/ml);

postopioid intensity

estimates based on extrapolated

preopioid intensity

and remifentanil attenuation; total surgical stress. Data from one of the patients in the
study.

hypnotic has an important impact on the physiological variables related to noxious

activation (e.g. vasodilation caused by propofol depresses blood pressure, but it does
not directly mean that the Noc/ANoc balance has shifted). Knowing this, the hypnotic
impact must be considered in the modelling process, to incorporate the hypnotic as an

134

4.4 Phase II Maintenance Analysis - Physiological Modelling

Figure 4.12: Structure representing the impact of anaesthetic drugs and noxious stimulus on the physiological measurable eects.

interference in the measured variables.

The variables found to be linked to noxious activation in study phase I were cardiovascular and frontal EEG and EMG related. Huiku et al. [78] studied the capability
of these variables to capture the Noc/ANoc balance of the patient and found that the
performance of the proposed index (SSI) did not improve signicantly with the introduction of the EEG and EMG related variables, since the responses are usually of short
duration and transient. Both approaches were here considered, to better understand the
impact of these variables and discern if their incorporation on an index would provide
better results in the assessment of the Noc/ANoc balance.

Figure 4.13:

Structure of the physiological model, taking into account physiological

variables linked to noxious activation dynamical relations, hypnotic drug interference,

and output estimation of the perceived stimulus.

Obtaining the dynamical physiological model is the rst step to develop an estimator for the perceived stimulus, based on the measurable physiological variables linked
to noxious activation, and drug doses (see Figure 4.13). The approach followed in this
thesis was to consider dynamic relations between variables linked to noxious activation,
and provide a population model that may be used in the estimation of the expected

135

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

responses to noxious stimulation, providing an estimate of the Noc/ANoc balance. It
should be highlighted that the estimate of perceived stimulus, based both on the annotations and remifentanil eect-site concentrations, has weaknesses such as: an inability
to continuously and accurately describe stimulus intensity during the procedures (there
may be periods during visceral manipulation when the surgeon interrupts stimulation,
being these periods not taken into account), and also the remifentanil Ce are estimated, and may not correspond to the individual pharmacokinetic and dynamic processes.
Knowing this, perfect t statistics were not expected for the individually adjusted models, and most certainly not for the population model. Although perfect t statistics
were not expected, patients' perceived stimulus were well explained by the variables
used in the model construction, as it will be shown in the following subsections.
Due to signal lost during data collection, two patients were excluded from this
analysis, and a set of 29 patients was considered.

4.4.1 Individual Modelling

State-space models were individually adjusted to each data set, considering BIS, EMG,

u(t)), and both TSS (Model 1)

HR, SBP, PPGA and propofol Ce (PropCe) as inputs (

and

postopioid intensity

(Model 2) as outputs (y(t)), as given by Equations 4.23 and

4.24 [182].

x(t + Ts ) = Ax(t) + Bu(t)

(4.23)

y(t) = Cx(t) + Du(t)

(4.24)

where Ts is the sampling time (1 s). Matlab was used for system identication, with
automatic order selection.
To evaluate model tness, the mean square error (M SE ) and error of t (F it

Error)

statistics given in Equations 4.25 and 4.26, were used.

M SE =

n
X
(yh(i) y(i))2

i=1

F it Error =

n
X
|yh(i) y(i)|
i=1

136

(4.25)

(4.26)

4.4 Phase II Maintenance Analysis - Physiological Modelling

where

yh

is the modelled response to the input

u, y

is the original trend, and n is the

signal length.
To evaluate the adjusted models tness, correlations between the modelled TSS or

postopioid intensity ,

and the original TSS or

postopioid intensity

were also calculated.

It should be highlighted that the model structure here presented implies causality,
and the construction of the model as here described is anti-causal (output precedes the
input response).

Although this may be surpassed, altering the model structure, the

construction of anti-causal models does not allow on-line implementation.

Knowing

this, the models were adjusted assuming causality.

Table 4.17: Order 1 state-space models adjusted to the individual data, t statistics, and
correlation between output and modelled output. Model 1 corresponds to the output
TSS, and Model 2 to the output

postopioid intensity .

Three input sets were considered

using bispectral index (BIS), frontal electromyography (EMG), heart rate (HR), systolic
blood pressure (SBP), pulse photoplethysmography wave amplitude (PPGA), and propofol eect-site concentration (PropCe). Data presented as meanstandard-deviation.

Model 1 (TSS)
Input Variables Fit Error

MSE Correlation

BIS, EMG, HR, SBP, PPGA, PropCe

0,220,06

0,090,05

0,840,11

HR, SBP, PPGA, PropCe

0,240,07

0,110,06

0,810,11

HR, SBP, PPGA

0,280,08

0,140,08

0,730,19

Model 2 (postopioid intensity)

Input Variables Fit Error
MSE Correlation
BIS, EMG, HR, SBP, PPGA, PropCe

0,060,02

0,090,03

0,730,16

HR, SBP, PPGA, PropCe

0,070,03

0,100,03

0,700,13

HR, SBP, PPGA

0,070,03

0,100,03

0,660,12

Table 4.17 shows the values of the model tness, using dierent combinations of input
and output signals (Model 1 and Model 2). The best order selected for the state-space
model was one, both for Model 1 and Model 2. It may be observed that correlations
are higher using TSS as the output signal, indicating perhaps TSS is translating the
perceived stimulus better than

postopioid intensity ,

since the modelled response based

on the physiological signals linked to noxious activation, tends to provide better results
in following TSS rather than

postopioid intensity .

Figure 4.14 shows the results for the individually adjusted model, in one of the
collected data sets for both TSS and

postopioid intensity .

137

It may be observed that

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

(a) TSS

(b) postopioid intensity

Figure 4.14: Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil (output), and bispectral index, frontal electromyography, heart rate, systolic blood pressure, pulse plethysmography wave
amplitude, and propofol eect-site concentration as inputs. (a) Individually adjusted
state-space model of order 1 to TSS. (b) Individually adjusted state-space model of
order 1 to

postopioid intensity .

although both models follow the output trend the modeled TSS (correlation 0.90) tends
to give a more close relation to the output trend than
0.82).

postopioid intensity

(correlation

It should be highlighted that the output trend may only be regarded as an

indicator of the patient's position on the Noc/ANoc balance, based on the available
information of both stimulus intensity and analgesic dose.

As pointed before, this

information is incomplete, and it may be observed that although the model follows the
trend of the estimated perceived stimulus, in some portions it is not capable of adequate
adjustment to the output original signal. This is due to a variety of factors: patient
individual response to noxious stimulation and drug attenuation, which may vary due
to several intrinsic and external factors, or to the poor estimation of the output signal.
It should be highlighted that although the t error statistics is in some of the data
sets high, the correlation between modelled trend and the TSS trend is also high. Figure
4.15a shows the result for the data set with higher correlation (correlation of 0.97).
There are however cases in which the correlation is below 0.7 (2 cases), demonstrating
that the trend describing the perceived stimulus may not be accurately describing the
patients' state. Figure 4.15b shows the results for the data set with lowest correlation
between modelled and original TSS (correlation of 0.51).

138

4.4 Phase II Maintenance Analysis - Physiological Modelling

(a) High Correlation

(b) Low Correlation

Figure 4.15: Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil, using total surgical stimulus (TSS) as
output signal, and bispectral index, frontal electromyography, heart rate, systolic blood
pressure, pulse plethysmography wave amplitude, and propofol eect-site concentration
as inputs. (a) Individually adjusted state-space model of order 1, in the data set with
higher correlation to the output signal TSS (correlation of 0.97). (b) Individually adjusted state-space model of order 1, to the data set with lowest correlation (correlation
of 0.51).

Dierent sets of input signals were employed in the search of the dynamical model
translating the perceived stimulus: the rst set composed of BIS, EMG, cardiovascular
signals (HR, SBP, PPGA) and propofol Ce; the second composed of cardiovascular
signals and propofol Ce; and a third set composed only of cardiovascular signals. This
was done to evaluate the information contained in the signals found to be related to
noxious activation, in the continuous translation of the dynamic relations and Noc/ANoc
balance. EEG derived indexes, as referred previously contain information on noxious
activation, but this is usually found to be transient in time. In the adjusted models it
was found that incorporating information derived from the EEG monitor, cardiovascular
variables and propofol Ce interference provided the best results in terms of tness and
correlation to the original trend. Although a multivariate approach leads to the better
results, for practical reasons it should be highlighted that the fewer signals employed in
the determination of the Noc/ANoc balance, the more practical and general the method
is. It may be observed that correlation is decreased with the removal of the dierent
components of the input signals, and also thet the t error and MSE statistics increase.

139

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

The objective of this section of the work was to dene dynamical models, which may
be employed in the translation of the Noc/ANoc balance of the patient, nevertheless
also taking into consideration practical issues. For that reason, in the next subsection,
population valid models are explored, using information obtained in the individual
modelling phase. This includes the design of three models, using the tree sets of inputs
presented in Table 4.17.

This was done to obtain the best model for the Noc/ANoc

balance in the collected data, using the hospital's standard monitoring and drugs' TCI,
and two simpler models to be valid in dierent clinical setups (without BIS monitoring,
and without TCI administration).

Order 1 models were adjusted to the population

data, and TSS was used as the output signal, since it was demonstrated that better
results are obtained when comparing to

postopioid intensity .

4.4.2 Merged Modelling

Following identication of individually adjusted models, it is necessary to produce a
valid model for the population data, the model that better ts the total data les must
be adjusted. Two approaches may be considered to achieve this: a merged model, based
on all the individually adjusted models, or the model that best ts the merged data.
The latest is more advisable, so the total number of data sets were compiled in a merged
data set (several individual experiments), used to adjust a global model, and later test
if the adjusted model provided adequate performance in the explanation of the dynamic
relations observed in the individual data set.

Baseline normalized data were used in

this analysis.
To reduce inter-patient variability, and set the start point of every signal to the same
origin (0), each signal (signal(t)) had the baseline value removed (signal(t)). Also,
due to observed dierences in the amplitude responses, the baseline value was also used
to normalize the response amplitude (signalN (t)), as given by Equation 4.27.

signalN (t) =
where

t=30 s

signal(t) mean(signal(t))
mean(signal(t))

(4.27)

corresponds to the rst 30 data points in the beginning of the data set.

A dierent approach is proposed by Huiku et al. [78], that to reduce inter-patient

variability, all variables are subjected to an histogram transformation. To perform this
normalization knowledge of the data using the group of study, and individual data during collection, are used. This approach requires individual adjustment in the beginning

140

4.4 Phase II Maintenance Analysis - Physiological Modelling

of every data collection, and it was initially implemented. Nevertheless, the need to individually adjust the method in the beginning of each case, led to the application of
simpler methods, that provided similar results in the simulations.
Table 4.18: Order 1 state-space models adjusted to the merged data sets, t statistics,
and correlation between output and modelled output, considering baseline removed
data, and baseline normalized data. Three input sets were considered using bispectral
index (BIS), frontal electromyography (EMG), heart rate (HR), systolic blood pressure
(SBP), pulse photoplethysmography wave amplitude (PPGA), and propofol eect-site
concentration (PropCe). Data presented as meanstandard-deviation.

Model 1 (TSS)
Input Variables
Fit
Baseline Removed

MSE Correlation

BIS, EMG, HR, SBP, PPGA, PropCe

0,680,38

0,780,96

0,260,41

HR, SBP, PPGA, PropCe

0,690,37

0,800,89

0,190,41

HR, SBP, PPGA

0,710,50

1,021,83

0,280,37

Baseline Normalization
BIS, EMG, HR, SBP, PPGA, PropCe

0,670,37

0,750,87

0,280,41

HR, SBP, PPGA, PropCe

0,680,37

0,77085

0,240,41

HR, SBP, PPGA

0,710,49

1,001,75

0,320,36

The model adjusted using all collected data was simulated, and individual adjustment evaluated. The model did not provide adequate results when following the trend
of the individual data, nevertheless the model adjusted to the normalized data provided
better results. This is due to the inter-patient variability and inadequate translation
of the patients' state to the TSS, when regarding the amplitude responses that vary
between patients, even following normalization.

Table 4.18 provides the t statistics

and correlations obtained in the population data for each method here described, using
baseline removed data and normalized data by the baseline value.
Figure 4.16 shows the results using the adjusted model to the merged data. Figure
4.16a shows the best individual result to the merged data model, and Figure 4.16b
shows the worst individual adjustment, considering correlation between modelled trend
and the original estimated TSS. Although correlation is lower using the population
model than the individually adjusted model, as expected, a similar trend is observed
between modelled and estimated TSS in almost all data sets, similar to what is shown in

141

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

(a) High Correlation

(b) Low Correlation

Figure 4.16: Modelled perceived stimulus, based on the stimulus intensity and attenuation provided by the analgesic remifentanil, using total surgical stimulus (TSS) as
output signal, and bispectral index, frontal electromyography, heart rate, systolic blood
pressure, pulse plethysmography wave amplitude, and propofol eect-site concentration
as inputs. (a) Data set with higher correlation to the output signal TSS, considering
the merged data model. (b) Data set with lowest correlation, considering the merged
data model.

Figure 4.16b, with increasing and descending phases consistent with the TSS estimated,
nevertheless with drift, leading to low correlation between original and modelled trends.
Table 4.19 presents the average, minimum and maximum values observed for all data
sets considered in this phase of the analysis, for the merged data model, with the input
set of BIS, EMG, HR, SBP, PPGA and PropCe. The maintenance phase of anaesthesia
was considered in this analysis (Start 2nd Phase to Post-Operative Analgesia in

times

vector). The results presented in Table 4.19 correspond to the expected values of the
merged model. It was observed that the patients are kept with values of the perceived
stimulus below zero, meaning with a slight depression considering baseline values prior
drug administration.

This was expected and it demonstrates that the merged model

adequately responds to the real trends in the patient, since following induction the
variables linked to noxious activation are usually depressed, and in practice tend to
stay below these values, since an increase in these variables may be linked to noxious
activation and arousal. Knowing this, the values observed and shown in Table 4.19 may
be regarded as the typical values, prior to depression and prior to arousal, since the data
collected with this clinical protocol only allowed for data collection in these conditions.

142

4.4 Phase II Maintenance Analysis - Physiological Modelling

Table 4.19: Average, minimum and maximum values of the merged model output perceived stimulus during the maintenance phase, considering all data sets and input set
of BIS, EMG, HR, SBP, PPGA and PropCe: population trends.

Data presented as

meanstandard-deviation.

Minimum
AverageStandard-Deviation
Maximum

Minimum

Average Maximum

-2,41

-1,76

-1,36

-0,960,55

-0,690,49

-0,330,42

-0,08

0,35

0,59

Table 4.20: Parameters of the state-space models using the merged data, considering
the three input sets.

A
B
C
D
A
B
C
D
A
B
C
D

Input 1: BIS, EMG, HR, SBP, PPGA, PropCe

0,9998
4,75E-07

-6,68E-08

3,77E-07

-1,71E-06

2,60E-09

-9,59E-08

0,0760

-0,3194

0,6402

-0,0025

-0,0632

609,7999
0,0386

Input 2: HR, SBP, PPGA, PropCe

0,9998
4,95E-07

-1,66E-06

2,59E-09

-2,17E-07

0,7033

-0,0024

-0,0545

612,4551
-0,4651

Input 3: HR, SBP, PPGA

0,9998
-1,49E-06

9,69E-07

8,55E-10

0,9878

-0,0041

-634,0687
-0,1743

Table 4.20 presents the parameters estimated for the merged data models, considering the three dierent input data sets.

4.4.3 Perceived Stimulus Estimator

The merged data physiological models obtained in the previous section, may be regarded as general models describing the population trend, and possible estimators of the
patients' perceived stimulus, and therefore of the Noc/ANoc balance. The origin (zero)
corresponds to the patients' state prior to drugs' administration, the normal baseline

143

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

values; a decrease below zero may be interpreted as a depression, and an increase as
an increase in the noxious stimulus perceived by the patient, above the attenuation
provided by anaesthetic drugs, similar to what is represented in Figure 1.19.

Figure 4.17:

Designed software for on-line assessment of the developed steady-state

indexes, individually or for combinations of input and output signals (In and

Out ),

and the estimator of the perceived stimulus, based on the physiological model for the
population. Data presented in this example is from one of the patients of the study, for
a simulation of an on-line assessment.

The physiological models were implemented for an on-line assessment of the patients' perceived stimulus, in the previously described software for the homeostasis index
calculation. Figure 4.17 shows the nal version of the developed software that, based
on passive measures linked to noxious activation, provides homeostasis indexes, both
for input and output variables, and an estimate of the patients' perceived stimulus
(Noc/ANoc balance location).

4.4.4 Discussion
Dynamic models were adjusted to the individual data in the search for the best descriptors of patients' location regarding the Noc/ANoc balance. In order to be able to
estimate the patients' state, or perceived stimulus, the annotations kept by the author
during data collection, and translated in stimulus intensity trends as described in Chapter 3, were employed, using extrapolation to the

144

preopioid intensity

as described in

4.4 Phase II Maintenance Analysis - Physiological Modelling

pharmacological studies. Two approaches were used in the translation of the patients
perceived stimulus, the calculus of the

postopioid intensity

and described by Bouillon

et al. [167], and the TSS as described by Huiku et al. [78].

The two methodologies

provide dierent trends, since the rst approach is based on the attenuation provided
by the analgesic, and the second is a simple arithmetic rule.

The rst methodology

only provides information on the attenuation provided by the analgesic dose, while the
second method articially introduces the depression impact of the analgesic drug, as
shown in Figure 4.10. This may account for the superior results when modelling the
perceived stimulus using both approaches as output signals, indicating that the TSS
better translates the shift in patients state regarding the Noc/ANoc balance.
The model structure employed in this study implies causality, nevertheless the system under identication is anti-causal. Although this was the case, it was decided that
to produce models for on-line implementation the data were assumed to be causal, disregarding the evident anti-causal nature of the process. In fact, the modelled trend for
the individually adjusted models, followed satisfactorily the original output signal.
Regarding the order of the model, due to the characteristics of the output signal, the
best order estimated was of one, nevertheless, the output signal presents limitations,
since it only gives an indication of the patients state, disregarding the obvious variations
in stimulation intensity during the procedure. Also, although the TSS provided the best
results, the amplitude responses provided by stimulus and analgesic introduction may
not translate the real patients state.
Three dierent sets of input signals were considered in the analysis. This was done
to inspect the impact of dierent signals in the translation of the patients' Noc/ANoc
balance state. The best results obtained were with the model adjusted using all input
signals found to be related to noxious activation, and also the hypnotic impact. It was
observed that removing input signals deteriorates the simulation results, with higher
t error and MSE and lower correlation to the output trend. Nevertheless, to be able
to apply these results to more general clinical setups, the three approaches were here
considered.
The output TSS and order 1 model structure were chosen in the search of the merged
model to translate the patients' state regarding the Noc/ANoc balance, valid for the
population in the study. Due to the well known intra and inter-patient variability, different normalization methods were applied to improve comparability between patients,
in the search for the best model structure.

145

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

The model obtained for the population in this study was applied to all data sets, to
evaluate the response to noxious stimulus under dierent analgesic doses. It was found
that the method translates the introduction of noxious stimulus and analgesic attenuation, nevertheless the population model did not provide good results to translate the
individual trends, as expected and observed for the individual model. This is due to the
fact that dierent patients have dierent baseline and amplitude responses in response
to stimulus and anaesthetic drugs. Also, it should be highlighted that the values of the
estimated TSS are not comparable between subjects, in the sense that the maximum
values of remifentanil reached during the procedure are conditioned by the individual
patient tolerance (interruption rules described in Chapter 2). Further studies to answer
these questions should be conducted, in order to obtain population models that may
adapt to the individual responses to noxious stimulation and anaesthetic drugs attenuation/depression. The population models here presented may be regarded as estimators
of the Noc/ANoc balance trends, considering that the maximum and minimum values
of TSS are in all patients conditioned by the individual tolerance/sensitivity to the
analgesic drug.
The methodology here proposed provides complementary information to the patients' state, and was implemented for on-line assessment in the previously described
software, which evaluates the homeostasis state of the patient. As referred before, homeostasis of the patient is synonym of stability regarding the signals linked to noxious
activation, nevertheless it does not imply stability around the optimum state. For that
reason being able to locate the patient in the perceived stimulus continuum, provides important information to the anaesthesiologist conducting the general anaesthetics
administration.

4.5 Summary
Collected data under general anaesthesia, following the clinical protocol described in
Chapter 2, allowed the inspection of the most informative physiological variables on the
noxious activation process.

These variables were analyzed and processed in order to

reveal the information contained in them regarding the Noc/ANoc balance.

Of the list of physiological signals analyzed in this study, all part of the standard clinical monitoring, the variables found to be related both to noxious activation and analgesic drug attenuation in response to precise noxious stimuli were: HR, SBP, PPGA,

146

4.5 Summary
BIS, SD BIS, EMG and SD EMG. These variables responded adequately to the introduction of a noxious stimulus, and to the analgesic drug attenuation.
The variables linked to noxious activation, and drug attenuation, were used in the
search of possible measures translating the patients' Noc/ANoc balance. Two dierent
perspectives on this subject were employed, the rst an homeostasis index, based on
wavelet analysis, and the second a model translating the patients' perceived stimulus.
Wavelet analysis was employed to the physiological signals linked to noxious activation, allowing the calculus of individual steady-state indexes, calibrated for each signal
characteristics. It was demonstrated that the proposed homeostasis index adequately
responds to precise stimuli and analgesic drug dose. Besides providing an estimate of
inputs and outputs steady-state, with detection of arousal or depression episodes, the
technique was also applied to evaluate if the designed clinical protocol was ecient in
the production of steady-state and dynamic conditions. It was observed that the protocol was ecient in the production of data in steady-state conditions, ideal to produce
steady-state valid models, and in the production of data in dynamic conditions, ideal to
produce dynamic models relating the physiological variables linked to noxious activation, later used in the modeling process. The steady-state index described allowed the
detection of periods of input/output simultaneous conditions, that may be used, when
more data is available, to model the relations of drugs and stimulus intensity on the
physiological signals linked to noxious activation.
Dynamic models were adjusted to assess the perceived stimulus (Noc/ANoc balance
related), taking as output the perceived stimulus based on the analgesic drug dose, and
the estimate of the stimulus intensity registered by the author (Rasch analysis). The
model takes into account the hypnotic eect since it interferes with the physiological variables linked to noxious activation, and allows the estimation of the patient's perceived
stimulus. The model used to describe the Noc/ANoc balance, using as inputs signals
linked to noxious activation, adequately responded to precise dierent noxious stimulus
and analgesic doses in the individually adjusted models. The population models obtained in this study need further enhancements, in order to translate the individualized
response (adaptation to the individual tolerance/sensitivity).
A software was designed, for on-line application, taking as inputs the physiological
signals linked to noxious activation and the input drugs' doses, allowing the on-line
calculus of inputs and outputs steady-state indexes individually, and as a combination of
these variables. Besides information on the patients' homeostasis state, the physiological

147

4. DATA ANALYSIS - PASSIVE NOCICEPTION MEASURES

model to estimate the perceived stimulus intensity, linked to the Noc/ANoc balance,
was also implemented for on-line assessment. The two indexes (Homeostasis Index and
Perceived Stimulus Estimate) provide complementary information, and combined msy
allow the detection of arousal episodes and an estimate of the patient's location in the
Noc/ANoc balance continuum.

148

Chapter 5
Evoked Potentials - Active
Nociception Measures

If you would be a real seeker after truth,

it is necessary that at least once in your life
you doubt, as far as possible, all things.
Ren Descartes

This chapter proposes an active measure of the Noc/ANoc balance.

It starts with a

state of the art on evoked potentials, with special focus on the somatosensory evoked
potentials application.

Then, a clinical protocol conducted for the purpose of this

chapter is reported, describing its design, approval, informed consent management and
data collection in a volunteers' study. Finally, the collected data is analyzed to assess
the feasibility of this method to translate the Noc/ANoc balance in uncommunicative
individuals [183].

5.1 Introduction
Pain is one of the main concerns of human beings. It becomes known to all from a very
early age, and throughout life fear of pain, avoidance of pain, and dealing with pain
becomes a major concern experienced by all. Pain from a physiological point of view is

149

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

a very complex process that may be dened regarding its intensity, frequency, source of
stimulation, psychological background, attention, to name a few. Dierent brain areas
are activated, responsible for each pain sensing characteristics.

Understanding pain

perception and pathways from the sensing organs to processing in the brain is of great
importance when trying to develop tools to assess nociception.
Recent technological developments allowed researchers to literally see the active
brain areas using fMRI [184187], for dierent stimuli and interfering factors. Neuroimaging studies allowed the description of brain areas involved in the noxious stimulus
activation, and a better understanding of the processes triggered in the nociceptive
system.
The nociceptive system comprises several structural areas in the brain, responsible
for dierent processes related to pain perception.

Some of these structures routinely

activated are the primary and secondary somatosensory cortices (SI and SII), the posterior, mid and anterior insula, the anterior cingulate and the prefrontal cortices. Activity
in subsortical areas such as the periaqueductal gray, hypothalamus, amygdala, hippocampus, and cerebellum, have also been related to pain experience [187].
Nociceptive stimuli are conducted to the brain by C and A- bers, for later central
processing. The way this information is conducted and post-processed is the basis of
the use of evoked potentials to evaluate nervous conduction. This technique has been
used for many years by neurophysiologists for clinical diagnosis of neurological diseases.
In the 1970s, 1980s and 1990s, researchers explored the possibility of using somatosensory evoked potentials to study, and eventually quantify, pain/nociception. Studies
using somatosensory evoked potentials have linked evoked waves characteristics to the
intensity of stimulation or to the presence of anaesthetic drugs [188]. Understanding
the relations between wave characteristics, stimulation intensity, and drug eect, may
lead to a direct measurement of nociception.
Based on this principle, we decided to investigate, in adult volunteers, the use of
somatosensory evoked potentials to study pain/nociception, namely to see the eect
of dierent controlled stimulus intensities, and to evaluate the eect of anaesthetic
drugs on pain and evoked responses.

Evoked potentials may, in fact, be seen as a

straightforward method that allows the observation of pain/nociception and related

cortical phenomenons in their essence.

150

5.2 Evoked Potentials

5.2 Evoked Potentials

An evoked potential is the electrical potential generated in the nervous system in response to a known, precise stimulus. This is dierent from the spontaneous electrical
activity generated in the cerebral cortex (EEG), cardiac muscle (ECG) or skeletal muscle (EMG), because the subject is stimulated in a certain manner in order to observe
the response elicited by the stimulus. This stimulus may have dierent characteristics,

Visual Evoked Potential (VEP),

Somatosensory Evoked Potential (SEP). The

and there are mainly three types of evoked potentials:

Auditory Evoked Potential

(AEP) and

use of evoked potentials has been of great importance in understanding the sensing and
brain mechanisms in response to outer stimuli both in human and animal subjects, and
also for diagnostic procedures mostly in neurological diseases.
Evoked potentials are produced in response to a stimulus in the nervous system,
by sensory, electrical, magnetic or cognitive stimulation; sensory evoked potentials are
evoked potentials produced by stimulation of the sensory system.

These responses

arise from action potentials or graded polysynaptic potentials during the propagation
of an electrical impulse from the periphery to the central nervous system, and can be
recorded over the scalp, as well as at various sites along the anatomic pathway, using
surface electrodes (e.g. Ag/AgCl electrodes) or subdermal needles [189].

Figure 5.1: Schematic representation of the stimulation pathway from stimulus site to
higher processing [190].

The use of evoked responses in anaesthesia has been explored mostly using middle
latency auditory evoked potentials to evaluate its hypnotic component state [8]. The
use of evoked potentials during anaesthesia, namely technical aspects of the evoked

151

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

potentials monitoring, and the inuence of drugs used in the perioperative period on
measured responses is well known [189, 190] due to the extensive use of intra-operative
evoked potentials monitoring in surgical procedures where nerve damage may occur
(spinal and brain surgery). Somatosensory evoked potentials may be elicited in almost
any nerve trunk and at various levels (see Figure 5.1), the most frequently used being
the median and ulnar nerve stimulation. Somatosensory evoked potentials elicited by
cutaneous painful stimulus present dierent delays depending on the site and type of
stimulus.
A somatosensory evoked response (vertex potential) is labeled according to polarity
(P for positive, and N for negative) and latency; wave characteristics depend on the site
of collection, as shown in Figure 5.2. Components with latencies between 100-700 ms
are considered to coincide with secondary cortical processing, involved in stimulus recognition, localization, magnitude and painfulness characterization (late components),
dierent from the ultra-late components with latencies in the order of 1-2 s [190].
Thornton and Sharpe [190] described the equipment necessary to obtain and analyze
the evoked potentials response as reproduced in Figure 5.3. Observation, recording and
measurement of the evoked potential poses technical diculties, since the evoked potential is embedded in the EEG. Direct observation of the EEG does not allow the
detection of the evoked potential. The usually employed technique for evoked potentials extraction uses the averaging of successive EEG windows, in synchrony with the
stimuli.

The stimulus applied depends on the sensory pathway of interest, auditory,

somatosensory, or visual. Usually a large number of consecutive stimuli are required to

obtain reliable potentials.
In pain studies there is a main concern on applying stimulus which will only induce
responses due to pain sensation, so the stimulus should be exclusively painful (not
activating pressure and touch sensors), quantiable in terms of intensity and duration,
and also easily applied, removed and repeated without tissue damage or habituation
[190].
Electrical stimulus is one of the most used to produce pain, however it activates
dierent nerve types, and attempts have been made to modify the electrical stimulus
removing the small core epidermis from the skin, and applying the stimulus in the
pulp of the nger.

On the other hand, thermal stimulus is considered to be a more

natural sensation, producing a relatively short duration pain sensation, readily time-

152

5.2 Evoked Potentials

(a) Cortical

(b) Cervical
Figure 5.2: Somatosensory evoked potential representation with latency, amplitude and
respective vertex potential marks (positive and negative): (a) cortical site; (b) cervical
site.

locked to the EEG. This type of stimulus is easily quantied and varied, allowing the
determination of the subjects' pain threshold.
For the selection of the stimulus intensity required to obtain an evoked response,

153

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

Figure 5.3: Equipment to obtain somatosensory evoked potentials, and process data
with time-locked stimulus [190].

stimuli of increasing intensities are applied, resulting in a response with increasing

amplitude and decreasing latency; at some point a ceiling phenomenon is observed so
that increasing the stimulus intensity does not cause a further increase in the response.
The number of stimuli required to produce a clear averaged response depends on the
amplitude of the response in relation to the background noise, however high stimulation
rate may reduce the response since the stimuli may be introduced during the refractory
period, so in practice a compromise has to be reached between the clarity of the response
and the time taken to obtain it. Short latency waves such as in auditory (early cortical)
and somatosensory evoked potentials can be produced by fast stimulation rates. Some
authors propose that the stimuli should not be given at regular intervals in order to
desynchronize the response of regular sources of interference such as EEG and power
line. EEG signal ltering is useful to abolish frequencies which interfere with the signal
interpretation, however it should be made with caution, since it may distort the signal,
and it may need to be adapted. Some factors may interfere with the evoked responses
readings, such as temperature (hypothermia produces an increase in latency and may
also produce reduction in amplitude, and hyperthermia has the opposite eect). Skin
characteristics such as reectance and thickness may also interfere with the amount of
energy in the surrounding area of the nociceptor [191, 192].
It should be highlighted that to obtain evoked potentials the adequate specialized
equipment and software are necessary, namely the stimulus device with synchronized
EEG data processing.

154

5.3 State of the Art on Evoked Potentials and Nociception

5.3 State of the Art on Evoked Potentials and Nociception

SEPs have been studied in relation to pain and nociception, and in this section of the
thesis, a review of the studies on this subject is presented.
SEPs may be used to assess responses to painful stimuli and to characterize the
brain and body automatic response mechanisms to the specic stimulation [193]. The
way pain is perceived is related to the sensor grid and to the type of bers excited.
Several authors addressed the issue of pain perception using dierent types of noxious
stimulus to better understand the sensing process.
Revenko et al explored the way C-bers respond to cutaneous nociception in anaesthetized cats [71, 194]. The authors used injection of chemical substances (potassium
ions, acethylcholine and methacholine) to produce the stimulus, and the response impulses registered invasively from the C-bers. Their results suggest that the ring rate
of the bers activation could be used as an invasive method to assess the eect of local
anaesthetics in response to the stimulus. In a dierent study the authors used a dierent
animal model, the rat, to assess the eects of epidural anaesthesia with bupivacaine,
recording the spinal reexes. Three doses of analgesic were injected epidurally and the
plantar muscles and the sciatic nerve stimulated to evaluate the responses. The results
demonstrated that the use of electrophysiological methods could provide quantitative
evaluation of the ecacy of local anaesthetic agents [195].
Gender dierences in the nociceptive perception have also been explored by several authors. Walker and Carmody [73] used electrical induced pain on the earlobe to
evaluate nociception gender related dierences (10 women and 10 men volunteers) in
response to ibuprofen, an anti-inammatory drug.

It was observed that women had

lower stimulus' thresholds than men, and also that men were more tolerant to pain; this
suggests that gender may be one of the factors that condition the way pain is perceived.
Also ibuprofen only had signicant analgesic impact in men, increasing the stimulus
threshold.

Dahan et al. [74] reviewed gender-specic responses to opiates, or opioid-

mediated responses, in both animal and human subjects. It was found that there was an
increasing number of well-controlled experiments analyzing gender related dierences
with opioids administration, and also that gender dierences existed, but the direction
and magnitude of this dierence depended on many interacting variables related to the
drug (e.g. amount and pharmacology) and to the subject (e.g. genetics and age); for
example, Smith and Gray [196] explored the age-related dierences in response to opioid

155

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

analgesics in male rats, observing that opioid sensitivity increased with age, same as reported for human subjects. The authors also outlined that although gender dierences
were observed in opioid anaesthesia, the inpatient opioid analgesic sensitivity variability
should lead the clinician to titrate drug doses according to patients' individual needs.
This dierence is not only observed in opioid analgesics, but is also present with other
drugs like anti-inammatory drugs, as outlined before [73, 74].
Haslam [68] investigated the perception of size and pain considering the lateral
dominance of the subjects, a possible bias factor. The researchers studied a previous
nd [67] that perception of size would be dierent according to the dominant hand, when
holding objects of the same size, and also that the pain threshold would be greater in
the dominant hand, leading to biased estimates according to the lateral dominance.
Pain in this case was elicited by the use of a pressure algometer, and authors found that
right-handed subjects presented higher pain threshold in the dominant hand.
Pain sensing is conditioned by many factors, which should be taken into account
when evaluating somatosensory evoked potentials.

Site of stimulation is important

due to laterality eects, since after stimulation waves are the same size or larger on
the contralateral side of stimulation, and stimulation of both sides of the body lead
to increased wave amplitudes, and decreased latency. Age and gender impact on the
evoked responses is an issue of debate, suggesting that the eects are not large when
comparing to population variability, however the general trend for all responses is of
increasing latency with age.

Some diseases may aect the evoked responses such as

multiple sclerosis and other demyelinating diseases. Since these diseases have selective
action in the A- bers, C-bers remain intact, which may result in the absence of late
vertex potential and emergence of ultra-late response transmitted by C-bers [190].
The stimulus characteristics are also important in the evaluation of nociceptive
pathways. Carbon dioxide and argon lasers are used to induce thermal pain, however
the infra-red carbon dioxide laser causes a rapid increase in skin temperature, which
may result in supercial skin lesions, and the local inammation modies the nociceptive
pathway, making this kind of stimulus unfeasible for multiple stimulations. The visible
argon laser stimulus does not produce this kind of side-eects and can be applied for
cutaneous stimulation [190]. In order to develop a tool for pain assessment, the painful
stimulus intensity should be related to the amplitude of the evoked response, and also
the response should be attenuated by analgesic drugs. In previous studies researchers
evaluated the use of the late component of the vertex response (A- bers) and found

156

5.3 State of the Art on Evoked Potentials and Nociception

that there was a relation between the intensity of the stimulus and the amplitude of
the response both with electrical and thermal stimuli. Various studies have shown that
the evoked responses' amplitude is reduced by opioids, although discussion subsists on
the complex central depressant action of opioids on arousal/attention possibly being
responsible for that change rather than a direct nociceptive eect. Nevertheless studies
using local, oral analgesics and anti-inammatory drugs support the use of the vertex
potential to measure pain sensation [190, 197].
Thornton and Sharpe [190] stated that the laser pain evoked response is a great
promise as an objective measure of pain, however only the late vertex has been studied
(A- bers) leaving the ultra-late component as a possible correlate of C-ber activation,
which could bring great clinical impact in the study of chronic pain.
Kumar et al. [189] conducted a review on the use of evoked potentials monitoring
in anaesthesia, focusing on studies where anaesthetic drugs were used, and their impact
on the evoked potentials (somatosensory, auditory and visual). Figure 5.4 presents a
table extracted from [189], where results obtained in several studies using anaesthetic
drugs and the impact on somatosensory, audio, visual and motor evoked potentials, were
reviewed. Almost every drug produces increasing latency and decreasing amplitude of
the vertex potential, in exception of nitrous oxide, midazolam and ketamine that did
not present a change in the latency, and etomidate, ketamine and meperidine which
present an increase in amplitude eect (see Figure 5.4).
The use of AEP has also been explored to assess the impact of surgical stimulation,
showing that the surgical stimulation even at constant anesthetic doses, has impact
on the AEP. Bonhomme et al. [95] studied 23 patients, allocated randomly to receive
ropivacaine plus clonidine (Group R) or normal saline (Group S) epidurally, with BIS
guided sevourane anaesthesia. The authors found that AAI but not BIS signicantly
increased following incision in the S group. This may be explained due to the fact that
AEPs derived index is collected near the somatosensory cortex, prior to frontal cortex
processing, containing information on somatosensory excitation.
Stimulus characteristics, electrodes placement and technical aspects, such as stimulation duration and intensity, are of great importance when evaluating SEPs, with
several approaches described in the literature. Following, some techniques used in these
studies are reported, demonstrating the concerns dealt with when evaluating SEPs.
Carmon et al. [198] studied the evoked responses to thermal stimuli in the right
forearm, in four healthy adult volunteers, using a CO2 laser. EEG periods of 500 ms

157

158

potentials.

Figure 5.4: Table extracted from [189], summarizing a review of studies evaluating the impact of anesthetic drugs on evoked

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

5.3 State of the Art on Evoked Potentials and Nociception

following stimulation were averaged, with monopolar recording montage, using electrodes on the vertex Cz, on the contralateral somatosensory cortex C3, on the ipsilateral
somatosensory cortex C4 and one on the contralateral associative somatosensory cortex
P3. They observed that thermal pain could elicit evoked responses to noxious stimuli,
correlated with subjective pain experience. This was later conrmed in dierent studies
by the same author with seven and thirteen volunteers [199, 200].
Chen and Chapman [197] used electrical tooth stimulation and auditory stimulation
to observe the evoked responses in thirty healthy volunteers, and the impact of the use of
aspirin on the responses. They observed that the evoked response by dental stimulation
decreased in amplitude, and increased in latency, but not the auditory evoked potentials,
indicating that aspirin did not act as a brain depressant but only of pain perception.
Loughnan et al. [201] used visual, somatosensory and auditory evoked potentials
to evaluate the impact of diazepam and fentanyl on the evoked potentials, in thirteen
patients scheduled for elective surgery.

The stimulus for the SEPs was an electric

stimulation of the median nerve, with a 100 s constant current pulse generator, set
to 150% of motor threshold at 2 Hz for 256 stimuli. Collection electrodes were placed
over the contralateral hand area and over the fth cervical vertebra, both referrred
to a midfrontal reference. The authors did not nd dierences between latencies and
amplitude responses after drug administration in any of the collected evoked potentials,
stating that these drugs may allow the preservation of evoked responses when it is
necessary to assess ichaemia, nevertheless other studies demonstrated dierences on
evoked potentials' amplitude and latency, as shown in Figure 5.4.
Arendt-Nielsen and Bjerring [202] compared the use of topical and local injection of
analgesic, using laser-induced pain and SEPs. The stimulus was applied to the dorsum
of the left hand using an argon laser 200 ms duration, power adjustable 0.005-2.9 W
and wavelengths 0.488 m (blue) and 0.515 m (green). The stimulus was applied to
dierent sites to avoid sensitization and the time intervals between stimuli randomized
10-30s.

The EEG records were from the vertex to earlobe.

The authors found that

lidocaine produced a total sensory block almost immediately following injection (absence
of response), and following topical cream application both sensory and painful thresholds
were increased with evoked responses reecting the changes in perception.
Arendt-Nielsen et al. [203] evaluated alfentanil ecacy using SEPs in six healthy
volunteers. The stimulus used was an argon laser, with adjustable output power from

159

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

50 mW to 3.5 W, comprising both blue (488 nm) and green (515 nm). The evoked potentials were recorded from a platinum needle electrode, inserted over the vertex with
reference linked to ear lobe electrodes. The EEG was amplied, ltered by a secondorder lter (0.5-15 Hz) and sampled. The authors found that the potentials recorded
over the vertex were large, and that 16-24 potentials were enough to obtain a good
signal-to-noise ratio. The peak-to-peak amplitude and latency of N1 complex was used.
The authors observed that after alfentanil injection, the amplitude of the N1 complex
was signicantly lower and also the latency increased, with no signicant changes in
respiratory and cardiovascular measurements.
Thornton et al. [204] explored the use of auditory evoked potentials and SEPs in
eight anaesthetized patients, before surgery, to evaluate the hypnotic eect of isourane
comparing to the hypnotic eect of nitrous oxide, which is recognized as a weak hypnotic
drug but a good analgesic, using equi-MAC concentrations. The somatosensory evoked
potentials were recorded using scalp and cervical montages, and the stimulus was applied
to the median nerve at a rate of 2.2 Hz and sucient current to produce a motor
response.

The evoked responses were averaged over 2.8 minutes.

SEPs amplitude

was reduced signicantly by the nitrous oxide, when compared to the isourane. The
cervically recorded waves had no signicant dierences between drugs, meaning that
nitrous oxide does not appear to antagonize nociceptive transmission at that level.
Granovsky et al. [205] used a

Contact Heat Evoked Potentials Stimulator

(CHEPS),

stating that this stimulus modality provides a larger area stimulation than the laser
stimuli; nevertheless to avoid habituation it has to be moved around during the tests.
In this study the evoked responses by CHEPS were evaluated using only EEG derived
signals, 24 healthy volunteers. The stimulus temperatures were applied randomly, with
dierent latencies according to the stimulus temperature (time to heat and time to
return to baseline values). Each stimulus block was applied to the same body region,
and then a 15 minutes interval, and the thermode was moved in a random manner in
a 68 cm area. Evoked potentials were recorded from Fz, Cz, Pz, C3, C4, T7 and T8
(Easycap), the average of all selected channels was used as the reference. The authors
found positive correlation between evoked potentials amplitude and both stimulus intensities, and pain magnitude, being the second the main contributor to the evoked
brain response.
Ohara et al. [207] explored electrical activity of the brain using subdural electrodes
in response to laser evoked stimuli. The stimulus was provided using a Thulium YAG

160

5.3 State of the Art on Evoked Potentials and Nociception

Figure 5.5:

Brain schematic representation with sensory, motor and auditory areas

outlined (extracted from [206]).

laser, with dierent energy levels (300-800 mJ), applied to the dorsum of the hand, recording the patients evaluation of pain perception in a numerical rating scale 0-10. The
electrical activity recorded from the subdural electrodes (

Electrocorticogram,

ECoG),

was amplied, ltered with a bandpass 0.1-300 Hz for the laser evoked potentials and
30-300 Hz for the somatosensory evoked potentials, with sampling rates above 1000 Hz.
The researchers found evoked responses over the SI, parasylvian and the medial frontal
cortex, and also the subjects' pain ratings were correlated to the intensity of the stimulus. The evoked peaks amplitude were also correlated to the intensity of the stimulus,
meaning that these areas may be involved in the stimulus intensity encoding.
In more recent studies further improvements in technology allowed the production
of functional images of the brain [187].

These studies aim at discerning the brain

areas responsible for pain processing, pain sensing, pain intensity discrimination and
attention.

Bornhvd et al. [184] used laser stimulation to apply computer controlled

brief radiant impulses of thulium near infra-red radiation (wavelength 1.96 m, diameter
5mm, pulse duration of 1 ms). This laser allows a penetration depth of 360 m into the
human skin, without damaging the epidermis or aecting the subcutaneous tissue, and
causing temperature to rise in the supercial skin fast enough to elicit activation of
thinly myelinated A- and unmyelinated C-bers. The stimuli were applied to the back
of the hand in random intervals (8-12 s) to avoid the anticipation of stimulus, and the site

161

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

was changed to avoid sensitization, habituation and tissue damage. Also the stimulus
intensity was randomized (300-400-500-600 mJ). The objective was to observe the active
brain areas, and the relation between the intensity of the stimulus with the

Oxygen Level-Dependent

Blood

(BOLD) signals in the fMRI images, and the individual's pain

perception. Response in the SI contralateral to the stimulus showed a linear relation to

the stimulus intensity, and in SII/posterior insula and anterior insula a nonlinear relation
was observed. This non-linearity is expected in areas responsible for coding pain rather
than discerning stimulus intensity. SI - stimulus related; SII/posterior insula and insula
- pain related;

Dorsolateral Prefrontal Cortex

(DLPFC) and

Posterior Parietal Cortex

(PPC) - attention/working memory-related (see Figure 5.5).

Other works [185, 186] also used fMRI but with simultaneous measurement of EEG.
Christmann et al. [186] used electric painful stimulation in six healthy volunteers. The
subjects were all right-handed (Edinburgh Handedness Inventory), and stimulated in
the right thumb for 20 minutes (20 blocks, 32 monophasic constant current stimuli of
200 s duration each). In order to average out alpha activity, the inter-stimulus interval
varied between 500 ms and 700 ms (around 1.7 Hz). After determination of the subjects
sensory and pain thresholds as well as pain tolerance, the stimulation current was set
individually to a level of 50% between pain threshold and pain tolerance. Roberts et al.
[185] used CHEPS in ten healthy volunteers, also evaluating BOLD signals.

In both

studies BOLD signals and SEPs were correlated to pain evaluations.

Kochs et al. [208] studied other parameters linked to nociceptive responses such as
HR and BP, in comparison to the use of evoked potentials. The authors explored the use
of parameters extracted from the EEG (percentile frequencies), HR, MAP and PETCO2
for three dierent groups (placebo, ketamine injection of 0.25 mg/kg and 0.50 mg/kg,
10 adult healthy volunteers in each group). SEPs but not EEG frequency percentiles
were highly correlated to pain ratings. The results seemed to demonstrate that the late
components of cerebral potentials evoked by somatosensory stimuli correlate with the
pain ratings and may be used as indicator of pain, even when the patient is unable to
provide an evaluation of the pain rating. They also observed that although there was
a dierence in HR and MAP after treatment, these variables maintained values inside
the physiological range. These results suggested that although HR and MAP contain
information about nociceptive responses, they may not be as much informative as the
somatosensory evoked potentials.

162

5.3 State of the Art on Evoked Potentials and Nociception

Recent studies [119, 120, 209, 210] used a similar approach to the evoked potentials,
but analyzing a dierent response to stimulation, a

nociceptive withdrawal reex

(RIII).

Micalos et al. [119] found a relation between RIII threshold and the painful threshold,
indicating that the identication of the RIII threshold may be a valid reliable methodology in experimental pain studies. France et al. [210] proposed the use of EMG to
identify nociceptive exion reex, and von Dincklage et al. [209] designed a tool to continuously track the RIII threshold, that was demonstrated to be related to anaesthetic
doses, and response to noxious stimulation, with comparable prediction probability to
the BIS index in a study with twenty male volunteers [120].

Nevertheless this infor-

mation is obtained prior higher processing, for spinal nociceptive processes.

Possible

limitation of the method is the use of muscle relaxants during general anaesthesia,
blocking EMG responses used in the assessment of the RIII.

In summary, SEPs have been used with success to assess nociception. Electrodes'
placement, stimulus' characteristics and site of stimulation diered among the dierent
studies. SEPs' stimulus should be reproducible, easy to quantify, well dened in time,
with little adaptation, only activating bers responsible for the pain processes, without
causing tissue damage. The stimuli most used were electrical, laser and contact heat.
The use of electrical stimulation activates motor responses, not reecting a pure reaction
to pain [211], nevertheless some studies refer the use of intradermal needles to apply the
stimulus, decreasing the motor interference in the evoked responses [190, 211]. Electrical
stimulation is an easy and repeatable stimulus to apply when studying evoked responses,
and poses as an attractive stimulus option. Laser stimulation has the advantage of only
activating C and A- bers, although there is the problem of adaptation; some authors
propose the change of site of stimulus between stimulations, preventing skin lesions and
habituation. Recent reports on the use of somatosensory evoked potentials refer the use
of contact-heat stimulators, to apply the stimulus. This stimulus has the advantage of
only activating bers responsible for the pain conduction, not evoking motor responses.
The attention factor may also interfere with SEPs, since when the subject is expecting
the stimulus there is a bias in the response due to stimulus anticipation. This may be a
problem in awake subjects, but it does not seem to pose as a problem for unconscious
subjects, nevertheless, changing the site of stimulation and randomizing stimulation
timing may be the solution for this issue.

163

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

After reviewing the state of the art on the use of somatosensory evoked potentials
during anaesthesia, we decided to investigate the relations between stimulation intensity,
and anesthetic drugs' eect on SEPs. The use of SEPs during anaesthesia so far, seems
to be focused only on monitoring nervous conduction during high risk procedures, to
prevent lesions of the nervous bers and adverse consequences.

5.4 Clinical Protocol Design

A clinical protocol was designed for a volunteers' study on SEPs and nociception, receiving hospital's approval.

Available guidelines for somatosensory evoked potentials

monitoring were used in the protocol design [212214].

5.4.1 Design Goals

The clinical protocol was designed to evaluate the following items:

Identify sensitive, motor and painful thresholds with correspondent sensory evoked
responses, in each volunteer.

Determine intensity relations between stimulus and amplitude of evoked response,

in order to obtain comparable measurements between individuals.

Identify the most adequate EEG channel (electrodes positioning) to obtain the
somatosensory evoked potentials, in the translation of noxious stimulus intensity
and drugs impact.

Corroborate the use of somatosensory evoked potentials during anaesthesia in

the translation of the Noc/ANoc balance, identifying the impact of changes in
stimulus intensity and dierent drug doses.

5.4.2 Selection Criteria

Healthy volunteers without pre-medication, lled the study requirements. The volunteers did not receive any kind of monetary compensation. For each volunteer, a folder
referenced by a code number, was created to store study les collected during the experiment and register demographic data (age, weight, height and gender).

164

5.4 Clinical Protocol Design

5.4.3 Clinical Setup

The study was conducted in the neurosurgical operating room, during three Saturday
mornings in order to avoid disturbing clinical activity.

An anaesthesiologist and an

anaesthesiology nurse were present throughout the entire procedure, on a volunteer

basis.
Volunteers presented themselves in the operating room with fasting over 6 hours,
and signed the informed consent form (Appendix D). Standard monitoring (ECG, SpO2 ,
non-invasive arterial pressure) and BIS were used. A hand vein was catheterized with
a venous catheter G20 and saline infusion started at 400 ml/h, and nasal oxygen at
3 l/min.
Anaesthetics administration was conducted using a TCI system with two electrical infusion pumps, present in the operating room, and the TCI and synchronization

c Waves (Demed), already installed and functional, in a dedicated

software RugloopII

computer, also present in the operating room.

After connecting all monitors and infusion pumps to the computer, it is possible to
control the infusion pumps through the TCI software, and determine the theoretical drug
concentrations in the volunteer (pharmacokinetic models of Schnider [50] for propofol,
and Minto [39] for remifentanil).
continuously by the author.

All stimuli and important events were registered

Collected data were later exported to Excel les using

c (Demed).

Labgrab2.03

The material used in data collection was:

Computer with data collection and TCI software installed;

EEG and evoked potentials monitor - Endeavor

Needle electrodes for EEG collection;

ECG electrodes for electrical stimulation of the wrist (3volunteer)

General Electric Aisys

ECG electrodes (5volunteer);

Nasal cannula;

BIS Vista

T M CR (Nicolet-Vyasis);

R

- monitoring and automatic ventilation;

T M monitor;

165

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

BIS sensors;

System for administration of intravenous drugs;

Perfusion pump, B.Braun

Propofol 1% Lipuro

Remifentanil Ultiva

Two infusion pumps Alaris

R

Infusomat;

R
R
R



/Fresenius
in B.Braun
Perfusor syringe 50 ml;

R
R


in B.Braun
Perfusor syringe 50 ml;
R

Asena GH MKIII.

5.4.4 Methods

Figure 5.6: International 10-20 system for electroencephalogram collection. The letters
used stand for: F-frontal, T-temporal, C-central, P-parietal, O-occipital, and Z referring
to the electrodes positioned in the midline.

Volunteers were maintained in a comfortable position, in a calm environment during

the experiment. Electrodes for electrical stimulation were placed in the volunteers wrist,
for median nerve stimulation, and EEG electrodes were positioned according to the
monitoring standards in order to obtain two dierent EEG channels, and subsequent
somatosensory evoked potentials collection. Montage according to Figure 5.6:

Channel 1: Scalp contralateral to the stimulation site (C3' or C4' - 2 cm posterior

to C3 and C4, respectively) - Scalp (Cz);

166

5.4 Clinical Protocol Design

Channel 2: Neck (C5S, cervical on the spinous processes of the cervical vertebra
C5) - Scalp (Cz).

EEG data were collected continuously, with stimulus synchronization and evoked
potentials trends extracted. The monitor used to stimulate and obtain the SEPs was an
Endeavor CR (Nicolet-Vyasis), with stimulation rate of 3.1 Hz (around 180 stimulus/min),
with 0.5 ms duration per stimulus. The monitor allows the adjustment of the electrical
stimulus current, so that individual tuning is possible.
Before drugs' administration, all baseline variables were registered.

Then drugs'

administration was conducted using TCI of propofol, with Schnider PKPD model [50]
and remifentanil, with Minto PKPD model [39].

EXPERIMENTAL GUIDE:
1. Start stimulation according to recommendations, adjusting electrical stimulation
in each volunteer in current steps (0.1 mA) until reaching sensitive, motor and
painful thresholds for each individual. After threshold detection, a battery of 5
tests should be taken according to the scheme:

Sensitive Threshold (ST);

Motor Threshold (MT);

Painful Threshold (PT);

30% increment of Painful Threshold (1.3PT);

2. Execute the stimulus battery and obtain baseline SEPs, accompanied by volunteer's pain evaluation in a numerical rating scale (NRS) varying from 0 (no pain)
to 10 (worst possible pain).

Register latencies and amplitudes of the obtained

SEPs.
3. Start remifentanil infusion using eect-site concentration steering. Set remifentanil eect-site to 0.5 ng/ml, after reaching the target, stimulate with the PT and
1.3PT stimulus, obtaining SEPs and volunteer self-evaluation in the NRS (2
minutes interval between stimulation).
4. Increase remifentanil eect-site concentration in 0.5 ng/ml steps and repeat stimulation; terminate increasing steps of remifentanil if arterial pressure or heart
rate decrease 20% of the baseline, or by clinical recommendation.

167

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

5. Decrease remifentanil eect-site concentration to 1.0 ng/ml.

6. Initiate propofol infusion, using eect-site concentration steering, and setting

the target to 1.2 g/ml; start increasing propofol, rst to 2.0 g/ml, followed by
0.5 g/ml steps, obtain SEPs and volunteers pain evaluation. Propofol increasing
steps should terminate as soon as the volunteer loses response to verbal stimulus,
keeping response to mechanical stimulation (OAAS=2).

7. Decrease propofol dose to a light sedation level (1.2 g/ml) and obtain SEPs and
patient pain evaluation.

8. Restart remifentanil increasing steps of 0.5 ng/ml, obtaining SEPs and NRS evaluations in each step. Terminate the study as soon as the volunteer loses response
to verbal stimulus, but still responds to mechanical stimulation (OAAS=2).

9. Cease drugs' infusion.

Following the procedure above described, volunteers rested for a period of one hour
in the recovery room, receiving express indication not to drive a car, operate machinery
or precision mechanisms during that day.
SEPs waves' amplitudes and latencies were obtained as follows:

for the cortical

SEPs, the amplitude considered was the dierence between the positive and negative
peaks, and the latency the dierence between the latencies of the positive and negative
peaks (interpeak latency), this was done hypothesizing that wave morphology suered
an attenuation and expansion in time, besides an increase in latency in response to the
stimulus; for the cervical SEPs the amplitude considered was the amplitude of the only
peak and respective latency, as represented in Figure 5.2.
Remifentanil doses used in the study were dened according to recommended dose
ranges for spontaneous breathing patients as described in Appendix C.
In order to analyze the collected data and assess relations between stimulus intensity,
SEPs wave characteristics, pain evaluations in the NRS and drug doses, non-parametric
Spearman rank correlation coecient and paired sample Wilcoxon rank sign test were
used. P<0.05 was considered statistically signicant.

168

5.5 Results

5.4.5 Institutional Approval

The designed clinical protocol was submitted for institutional approval following the
same procedures as described earlier (N/REF.

a 073/10(053-DEFI/069-CES)) and de-

picted in Figure 2.2. However, this study was conducted in volunteers, and therefore,
beyond the reach of usual clinical practice with increased costs for the hospital. Due to
the use of hospital's resources the clinical protocol was also evaluated by the Hospital's
Financial Department to assess study expenses.

5.5 Results

Figure 5.7: Clinical setup used in the volunteer's study data collection: A) SEPs monitor and stimulation device Endeavor

T M CR (Nicolet-Vyasis); B) Electrical stimulus

electrodes placed on volunteer median nerve; C) Aisys monitor and ventilator; D) BIS
sensor and SEPs needles montage.

After obtaining hospital's approval, data were collected during three Saturday mornings. All volunteers arrived with fasting over 6 hours, and following the experiment

169

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

were maintained under observation for one hour, advised not to operate machines following discharge. Figure 5.7 presents the clinical setup and monitoring devices used in
the experiment. Data presented as meanstandard-deviation.
Table 5.1: Volunteers' demographic data (data as meanstandard-deviation).

Volunteers' Demographic Data (n=10)

Gender (M/F)

5/5

Age (years)

29.59.1

Weight (kg)

63.08.9

Height (cm)

171.47.2

Table 5.2: Propofol and remifentanil eect-site concentration targets and interruption
rules applied in the study: the study began with remifentanil administration only, in increasing steps as described in the text, until one of the interruption rules occurred; after
interruption the drug doses were reduced to the rst step considered in the next phase,
and drug doses were increased again according to the scheme.

The study had three

sequent schemes, rst remifentanil only, followed by constant remifentanil and propofol
increasing steps, and nally constant propofol and remifentanil increasing steps.

Remifentanil Only

Remifentanil and
Propofol

Propofol and
Remifentanil

Propofol

Remifentanil

Propofol

Remifentanil

Propofol

Remifentanil

Ce (g/ml)

Ce (ng/ml)

Ce (g/ml)

Ce (ng/ml)

Ce (g/ml)

Ce (ng/ml)

1.2

1.2

1.2

1.2

2.5

2.5

2.5

1.2

Remifentanil 0.5 ng/ml

Propofol 0.5 g/ml

Remifentanil 0.5 ng/ml

increasing steps

increasing steps

increasing steps

STOP increasing drugs' concentrations when BP or HR decrease below 20%

baseline values, by clinical indication, or on OAAS=2 (loss of response to verbal
command)

Ten healthy adult volunteers (see Table 5.1 for demographic data), were enrolled

170

5.5 Results
in the study. One subject was excluded from the analysis due to the impossibility of
obtaining SEPs evaluation for the PT stimulus.

(a) Well Dened

(b) Not Well Dened

Figure 5.8: Collected cortical somatosensory evoked potential waves, with dened peaks
by the neurophysiology technicians.

SEPs waves were punctually collected for each drug combination, and delimited by
experienced neurophysiology technicians (Neurinbloc). Due to the nature of the process,
and being the volunteers forcefully fully awake and freely moving, data were prone to the
occurrence of outlier values, specially during painful stimulation. Figure 5.8 presents a
pair of cortical SEPs waves collected during the experiment; observing the waves while
5.8a is very well dened, similar to the wave outlined in Figure 5.2, 5.8b is not as well
dened (contamination).
Remifentanil and propofol administration was done by stepwise increments as outlined in Table 5.2, using TCI. Interruption rules described in Table 5.2 include a 20%
drop in BP and HR from baseline values, loss of response to verbal command with
response to mechanical stimulation (OAAS=2, Table 1.1), or interruption by clinical
indication.
Drug increments were stopped every time the volunteer was unable to respond to
verbal command, while being capable of responding to mechanical stimulation; this was
achieved in all volunteers. Brief periods of apnea occurred, all spontaneously reversed
requiring only neck extension and drugs' administration interruption.
BIS minimum values reached were 52.38.6 (see Figure 5.9).

Pain ratings in the

NRS varied from 0 to 8, and 1.3PT stimulation currents varied between 6 mA and
52 mA (see Figure 5.10).

171

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

(a) BIS Trends

(b) BIS Individual Trend

Figure 5.9: BIS trends for all volunteers during the experiment (a), and one case in
detail (b), centered according to propofol start (time=0). The study was interrupted
as soon as the volunteer lost response to verbal command, maintaining response to
mechanical stimulation, or by clinical advice.

5.5.1 Electrical Stimulus Currents and Evoked Responses Prior to

Drugs' Administration
Individually adjusted stimulus currents for ST were 2.91.3 mA, for MT 8.02.8 mA,
for PT 20.39.6 mA and for 1.3PT 26.512.4 mA. Increasing variability in stimulation
currents from the sensitive threshold to the painful threshold was observed and is shown
in Figure 5.10.
Volunteers exhibited dierent painful thresholds, meaning dierent sensitivities to
electrical stimulation; also, the SEPs wave characteristics varied among volunteers, in
spite of stimulation with the individually adjusted painful stimulus (see Figure 5.11).

172

5.5 Results

Figure 5.10: Sensitive (ST), motor (MT), painful (PT) and 1.3 painful (1.3PT) threshold currents of stimulation for each volunteer.

Volunteers evaluated pain in a NRS both when the PT was achieved, and for every
1.3PT stimulus applied subsequently. As soon as the stimulus was enough to produce
pain, the NRS evaluations were higher or equal to 5 in all volunteers except for one.
Figure 5.12 presents the individual evaluations for the PT and 1.3PT stimuli applied
prior to drug administration. From the PT to the 1.3PT stimulus, the evaluations in
the NRS were maintained or increased.

5.5.2 Numerical Rating Scale Evaluations and Remifentanil and Propofol Concentrations
Remifentanil administration, at increasing concentrations, resulted in a reduction in
pain evaluations in the NRS. The same happened when propofol was administered. Table 5.3 presents the NRS evaluations for the maximum drugs' combinations achieved
in each administration scheme (no drugs, increasing remifentanil and no propofol, increasing propofol with constant remifentanil, and nally increasing remifentanil with
constant propofol). Although a decrease in the NRS evaluations was observed for all
administration schemes, in the last two combinations, the volunteer either rated pain as
0, or was unable to rate pain due to becoming unresponsive (lack of verbal response).

173

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

(a) Painful Threshold

(b) 1.3 Painful Threshold

Figure 5.11: Relation between the painful stimulus (a) and 1.3 painful stimulus (b)
intensity and the SEPs' latency and amplitude responses prior drug administration.

5.5.3 Normalized Somatosensory Evoked Responses Ratio

The interindividual variability in current stimulus (see Figure 5.10), and corresponding
SEPs amplitudes (see Figure 5.11), led us to the decision to normalize the SEPs amplitude in response to the 1.3PT stimulus, by the observed SEP amplitude prior to drug
administration for the PT stimulus individually adjusted, as given in Equation 5.1.

SEP AmpN orm =

174

SEP Amp
SEP Amp0

(5.1)

5.5 Results

Figure 5.12: Numerical rating scale pain evaluations (0-10) for each volunteer both for
the painful threshold (PT) and 1.3PT stimuli.

Table 5.3: Numerical rating scale (NRS) evaluations in each volunteer for the baseline
and maximum drug doses combination achieved in each administration scheme (NE no evaluation).

NRS Evaluations

Volunteer

Increasing

No

Increasing

and Constant

Remifentanil and

Drugs

Remifentanil

Remifentanil

Constant Propofol

(1.0 ng/ml)

(1.2 g/ml)

5
NE
0
NE
NE
NE
NE
NE
NE

NE
0
0
NE
0
NE
NE
2
0

1
2
3
5
6
7
8
9
10

where

Increasing Propofol

SEP Amp

8
6
8
8
8
8
9
9
7

2
4
3
6
1
2
0
2
0

is the observed SEP amplitude and

SEP Amp0

is the SEP amplitude

for the PT stimulus without drugs in the system.

The knowledge that increased latencies and decreased amplitudes are related to drug
SEPs attenuation [188, 189], led us to decide to evaluate the ratios

175

and

RN orm

as

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

given in Equations 5.2 and 5.3.

R=
where

SEP Lat

(5.2)

SEP AmpN orm

SEP LatN orm

(5.3)

is the SEP latency, and

RN orm =
where

SEP Amp
SEP Lat

SEP LatN orm

is the normalized latency by the observed latency to the PT prior

drug administration (SEP Lat0 ) and given by Equation 5.4.

SEP LatN orm =

SEP Lat
SEP Lat0

(5.4)

5.5.4 Cortical Somatosensory Evoked Responses and Numerical Rating Scale Evaluations
Table 5.4: Spearman correlation coecient (rho) for each volunteer between the normalized cortical somatosensory evoked potentials ratio (RN orm ), numerical rating scale
evaluations, and remifentanil eect-site concentration in the rst administration scheme,
where remifentanil is administered in ascending steps, without propofol (N is the number of points used to determine the correlation coecient, and * stands for signicant
correlation with P<0.05).

Volunteer

NRS vs RNorm N Remifentanil vs RNorm N

0,4564

14

-0,6347

0,4343

13

0,1078

0,5934*

18

-0,5593

10

0,4543

-0,7885*

0,2592

11

-0,0904

0,5465*

21

-0,2325

16

0,2285

16

-0,1734

15

0,3898

18

-0,4869

13

10

0,5007*

17

-0,6242

10

MeanStandard-Deviation

0,42920,1214

176

-0,38690,3002

5.5 Results
Table 5.5:

Spearman correlation coecient () for the population data between the

cortical somatosensory evoked potentials normalized ratio (RN orm ) and the numerical rating scale evaluations, and drug's eect-site concentration in each administration
scheme (where N is the number of points used to determine the correlation coecient).

N Variables

0,1678

0,0500

137

-0,1762

0,0860

96

Remifentanil vs

-0,1251

0,4076

46

Propofol with Remifentanil=1.0 ng/ml vs

-0,0231

0,9000

32

Remifentanil with Propofol=1.2 g/ml

NRS vs

RN orm
RN orm
RN orm
vs RN orm

The relationship between SEPs and reported pain is shown in Figures 5.13 and 5.14
as the result of the ratio and normalization of SEPs characteristics (1.3PT stimulation)
versus the NRS evaluations. A decrease in cortical SEPs amplitude with decreasing pain
ratings in the NRS (see Figure 5.13) was observed.

After normalization volunteers'

inter-variability was reduced, while the relations were maintained (see Figure 5.14). All
volunteers exhibited the same direction of relation, conrmed by the Spearman rank
correlation coecients: in all volunteers the correlation coecients were positive, but
signicant in only three, the ones with the larger number of evaluation points (see Table
5.4).
Since after normalization the obtained ratios were comparable between subjects and
dimensionless, data from all volunteers was aggregated to obtain a population correlation coecient (see Table 5.5). The correlation coecient between SEPs normalized
ratio and the evaluations in the NRS exhibited a positive trend, although not statistically signicant.

5.5.5 Cortical Somatosensory Evoked Responses and Remifentanil

and Propofol Concentrations
Results for all volunteers regarding the three administration schemes (remifentanil only,
increasing propofol and constant remifentanil, and increasing remifentanil and constant
propofol) are shown in Figure 5.15. In general, and for each administration scheme in
each volunteer, a decrease in the ratio with increasing drug doses was observed, although
the relation was not always evident due to the reduced number of points for each

177

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

(a) Original SEP Amplitude

(b) Normalized SEP Amplitude

Figure 5.13: Numerical rating scale evaluations (NRS) versus original (a) and normalized (b) SEP amplitude.

volunteer and possible outlier interference (in fact the remifentanil only administration
scheme has the larger number of evaluation points).
Spearman rank correlation coecients between cortical SEPs normalized ratio and
the remifentanil eect-site concentrations were calculated for each volunteer. Almost all
volunteers exhibited the same direction in the relation between increasing remifentanil
doses and the SEPs ratio, however statistical signicance was achieved in only one
volunteer (see Table 5.4). Table 5.5 presents the Spearman rank correlation coecients
for the population data between the cortical normalized ratio and drugs' eect-site
concentration in each administration scheme (decrease of the ratio with increasing drug

178

5.5 Results

(a) SEP Ratio

(b) Normalized SEP Ratio

Figure 5.14: Numerical rating scale evaluations (NRS) versus: (a) SEP amplitude/SEP
latency; (b) normalized SEP amplitude/normalized SEP latency.

doses, although not statistically signicant).

5.5.6 Cervical Somatosensory Evoked Potentials

Data were collected from both cervical and cortical sites to assess the use of the two sites
of collection is the comparison between evoked potentials prior and posterior central
sensory processing.

The evoked potential in this case was dierent from the evoked

potentials in the cortical site (negative peak only).

The ratio between cervical SEP amplitude and latency were analyzed with original
and normalized values. Cervical SEP's ratio and corresponding numerical rating scale

179

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

(a) Remifentanil Only

(b) Propofol & Remifentanil of 1.0 ng/ml

(c) Remifentanil & Propofol of 1.2 g/ml

Figure 5.15: Anaesthetic drugs impact on somatosensory evoked potentials normalized
ratio SEP Normalized Amplitude/SEP Normalized Latency versus: (a) Remifentanil
only; (b) Propofol and remifentanil eect-site concentration of 1.0 ng/ml; (c) Remifentanil and propofol eect-site concentration of 1.2 g/ml.

180

5.5 Results

(a) Cervical SEP Ratio

(b) Cervical Normalized SEP Ratio

Figure 5.16: Original (a) and normalized (b) ratio between cervical SEP amplitude and
latency, versus evaluations in the numerical rating scale.

ratings did not show a direct relation (see Figure 5.16); also, a relation between these
variables and the drugs eect-site concentrations was not observed (see Figure 5.17).

Correlations were obtained, demonstrating that the relations are not regular: for
example, correlation between NRS evaluations and the SEPs ratio obtained for each
volunteer presented dierent directions, with average value of 0.130.25.

The same

happened for the SEPs ratio versus remifentanil Ce, with average value of -0.090.40.

181

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

Figure 5.17: Normalized cervical SEP ratio versus remifentanil eect-site concentrations
for each volunteer.

Figure 5.18: Boxplot of the normalized SEP amplitude/normalized SEP latency ratio
(RN orm ) for the maximum drug doses combination achieved in each volunteer in each
administration scheme: 1 - No drugs; 2 - Remifentanil only; 3 - Propofol increasing steps
and constant remifentanil (1.0 ng/ml); 4 - Remifentanil increasing steps and constant
propofol (1.2 g/ml).

182

5.5 Results

5.5.7 Maximum Drug Doses Cortical Eect

The relation between SEPs normalized ratio and the maximum drugs doses achieved
in each volunteer for each administration scheme (no drugs, remifentanil only, propofol increasing steps with constant remifentanil, and remifentanil increasing steps with
constant propofol) was examined. Figure 5.18 presents the normalized SEPs ratio dispersion for each administration scheme, considering the ratio values obtained for the
maximum drug combination achieved in the individual volunteer. There was a decrease in the ratio, for every administration scheme in comparison to the baseline prior
to drug administration. This dierence was signicant between administration scheme
1, no drugs, and 4, increasing remifentanil and constant propofol, using a Wilcoxon
paired sample comparison (P=0.0273). Administration scheme 4, increasing remifentanil doses, for a constant propofol dose of 1.2 g/ml, is the more resemblant to general
anaesthesia, since a propofol sedation dose was chosen and the analgesic dose increased
according to volunteer needs/concession.
Although a tendency in decreasing

RN orm was observed for the maximum drug doses

achieved in administration scheme 2 and 3, more data is needed to obtain statistically

signicant results for these administration schemes. The incidence of outliers still has
an important eect on the hypothesis tests since punctual measurements of SEPs in
conscious non-paralyzed volunteers were used.

5.5.8 Nonlinear Nature of Pain Sensation an Dynamic Modelling

As described in Chapter 1, pain perception is conditioned by many factors, the most
direct being the intensity of the stimulus. The way sensed pain relates to the intensity
of the stimulus used to be described by the Weber-Fechner law, that stated that the
magnitude of the sensation was proportional to the logarithm of the intensity of the
stimulus; actually Equation 1.1 is used instead to describe this relation.

The non-

linearity of the sensing process, led us to the inspection of logarithmic transformations

of the collected data, also suggested by Crabb et al. [188].
Using a semi-logarithmic scale demonstrated to be appropriate to reduce interpatient variability, as shown in Figure 5.19.

The use of logarithmic transformation

may lead to the production of linear relations between the analyzed variables.
The assessment of the normalized ratio, possibly with a logarithmic transformation
seems to be an adequate method to assess the Noc/ANoc balance.

183

It was related

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

(a) Normalized SEP Ratio

(b) Normalized SEP Ratio (log10 scale)

Figure 5.19: Normalized SEP cortical ratio in the original linear scale (a) and logarithmic scale (b), versus evaluations in the numerical rating scale.

to changes in drug doses and to the reported pain by the volunteers in the NRS. It
produces comparable measures between subjects, dimensionless and normalized for the
individually adjusted SEP in the painful threshold.
Figure 5.20 presents the original data points of each combination of propofol and
remifentanil, versus the SEPs normalized ratio (RN orm ) obtained during each combination, for all volunteers.

This gure represents the dynamic relation between drug

doses (eect-site concentration, that in the case was in equilibrium with the plasma
concentration) and the measurable eect

RN orm ,

Noc/ANoc balance.

184

that is hypothesized to translate the

5.6 Discussion

Figure 5.20: Collected data points from all volunteers considering propofol and remifentanil eect-site concentrations (Ce) as inputs, and the normalized SEPs ratio (RN orm )
as output measurable eect.

More data, with a dierent study design (during general anaesthesia, with unconscious patients) is needed to adjust a dynamical model to be used to assess the relation
between propofol, remifentanil and output measurable eect

RN orm ,

in other words to

assess the Noc/ANoc balance.

5.6 Discussion
In this study a new method, combining both amplitude and latency of cortical SEPs,
is proposed as an objective indicator of the Noc/ANoc balance.
Propofol and remifentanil increasing doses lead to cortical SEPs wave depression
with amplitude reduction, increased latency, and simultaneously to decreased NRS ratings. These observations corroborate results from previous studies [188, 189], indicating
a depression in wave amplitude, and increase in latency, with increasing doses of propofol and remifentanil. In the present study the information contained on SEPs' wave
morphology was assessed, in cooperative volunteers, in response to dierent stimulus intensities, and in response to hypnotic and analgesic drugs, allowing for a more complete
analysis of SEPs applicability in the assessment of Noc/ANoc balance.

185

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

Sensory evoked potentials wave characteristics are aected by several factors [66,
189192], including genetic predisposition [215].
A wide interindividual variability both on the individually adjusted stimulus thresholds (specially in the painful threshold), and in the evoked responses was observed,
conrming that pain mechanisms vary between subjects, and any tool capable of objectively assessing the Noc/ANoc balance must take this into account. To overcome this
limitation, normalization of the SEPs wave characteristics was performed by the observed wave characteristics in response to the PT prior to drug administration. This way,
observed responses could be normalized by the same factor, producing measurements
comparable between subjects.
Due to wave characteristics, experimental observations, as well as the trends described in the literature, an increase in complexity, and a decrease in the number of
variables was obtained by combining the SEP's latency and amplitude in a single index. The idea was to take advantage of the information contained in both parameters,
knowing that with drug increase the amplitude of the wave decreases and the latency
increases [189]; this inverse proportionality was explored using a ratio between the two
variables, which enhances the information contained in both wave characteristics.

It

was observed that this information was additive, and that the relation between NRS,
drug doses and the ratio were maintained. Following the same approach as before, the
ratio was improved by using the normalized SEPs amplitude and latency, which led to
the construction of a dimensionless measure, comparable between subjects.
SEPs may be used in the future to translate the nociceptive/drug attenuation balance not only in conscious subjects, but most importantly in individuals not capable
of communicating verbally or in other ways, due to disease, altered mental or consciousness states, or young age, since the results demonstrate that the wave depression is
accompanied by a decrease in conscious pain perception (decreased NRS ratings).
No relations between cervical SEPs wave characteristics and both NRS ratings and
drug doses were found.

This may be due to the fact that the cervical potential is

collected prior higher processing and therefore, is only representative of the stimulus
conduction before higher processing.

Also the fact that stimulation intensity was si-

milar in all drug steps, may indicate that propofol and remifentanil do not antagonize
nociception transmission at this level, and similar wave characteristics are exhibited
[204].

186

5.6 Discussion
It must be highlighted that the pain evaluations are in a NRS, and therefore have
limited application. These scales do not guarantee linear relations between entries, they
are just ordinal scales [142, 144], and therefore the linear regressions presented in the
gures are just indicative of the direction of the relations. The correlation coecient
used to express the relation was the non-parametric Spearman rank correlation coecient, which does not imply a linear relation in the data, solely that the direction of
increase is maintained (increase in

is followed by successive increase/decrease in

y,

not necessarily in a linear fashion). Also note that the number of points to determine
the correlation coecients individually is, in many cases, low and therefore with limited
meaning. To strengthen the results presented in this study the number of volunteers
should be increased.
Although statistical signicance was reached in the collected data, the eect of outliers may not be disregarded. Besides the inter-volunteer variability, volunteers were
conscious throughout most of the procedure, and were not paralyzed, so that movement
was present and a possible cause of artifacts. Although minimization of the incidence
and eect of outliers was a concern, the SEPs measurements obtained correspond to
punctual moments, and therefore may be contaminated or biased. This may dicult
data interpretation, and more data should be collected to minimize this eect.

Ne-

vertheless, signicant results were obtained, leading to the assumption that the SEPs
may translate the nociception/anti-nociception balance.
The way pain is perceived is conditioned by many factors, eliciting noxious responses with known precise stimulus, adjusted to individual needs, and observing evoked
responses after central processing is a direct way of measuring and understanding the
nociceptive phenomenon. SEPs amplitudes are depressed by the use of analgesic and
hypnotic drugs, but more important the SEPs amplitudes are related to the pain ratings
of the volunteers, showing that this may congure an objective method of measuring
nociceptive balance in uncommunicative patients.
The fact that the hypnotic drug had an impact on the SEPs wave amplitude, as
well as the analgesic drug, should be reinforced.

Although one may think that the

analgesic drug should be the only one blocking the stimulus, in fact observations show
a decrease in wave amplitude accompanied by a decrease in the NRS ratings provided
by the volunteers, corroborating also the decrease in the SEPs amplitude in response to
propofol, which may be interfering with pain perception. It was clearly observed that
the addition of propofol resulted in SEPs depression accompanied by a decrease in pain

187

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

ratings, both reversed with propofol withdrawal. Some studies have addressed the issue
of propofol possible analgesic eects [46, 216, 217] and it is described that propofol may
inhibit the response to noxious stimuli [34, 45].
It was also observed that pain is sensed in a non-linear fashion, as described in the
literature [23, 188], and that nonlinear transformations, such as logarithms, may be
necessary in data interpretation and comparison between patients.
It is also interesting to notice that when adjusting the stimulus intensity, as soon
as the stimulus is considered to be painful the evaluations in the numerical rating scale
are chosen in the upper half of the scale, which may indicate that the pain process is a
switch process, and from the moment the stimulus crosses from discomfort to painful,
the evaluations in the NRS tend to be higher.
In conclusion, in this study the inuence of an analgesic and a hypnotic drug on
SEPs wave characteristics and corresponding pain as assessed by the subject were shown
to be related. A new method is proposed combining SEPs wave amplitude and latency,
which may be used as an dimensionless indicator of the Noc/ANoc balance with special
interest in subjects not capable of communicating.

5.7 Future Developments

The results obtained in the volunteers study leave an open door to the direct and
objective measurement of the Noc/ANoc balance in uncommunicative patients.

The

fact that the SEPs waves characteristics are related both to stimulus intensity and drug
attenuation, responds to the basic requirements of a Noc/ANoc index.
The proposed index, based on the ratio of the normalized SEPs amplitudes and
latencies, presents itself as an easy to obtain measure, since the PT and 1.3PT may
be easily obtained to calibrate and produce individually adjusted measures, normalized
to an dimensionless index (RN orm ).

Also the logarithmic transformation, increases

comparability between subjects, since it was observed that the inter-volunteer variability
decreased and extreme values were brought together to the main direction observed in
the population data.
In order to continue exploring the potential value in this method, studies in patients
under general anaesthesia with BIS guided and controlled hypnosis should be conducted,
in order to evaluate the impact both of dierent outer stimulation intensities, and of
dierent analgesic levels on the

RN orm

in the anaesthetized patient.

188

5.7 Future Developments

A dynamic model translating the analgesic plasma concentration into eect-site concentration (ke0 estimation), and eect-site concentration into measurable eect (structure unknown, but may be translated in a sigmoidal relation) should be adjusted, taking
into account possible synergistic relations between the hypnotic and analgesic drugs (as
in Equation 1.7).

Although data were obtained with simultaneous administration of

propofol and remifentanil, as represented in Figure 5.20, the number of data points is
small, as well as the dispersion and number of drugs' combinations. This was the case
due to protocol features, which was designed to evaluate the impact of dierent drugs'
doses on pain sensation and SEPs wave characteristics, to inspect if SEPs translated
the Noc/ANoc balance. Now, with the results in the volunteers study pointing in the
direction of SEPs as an objective tool to assess the Noc/ANoc balance, a new study
should be designed to improve technical aspects of data collection and processing, clinically validate the tool and identify and adjust the dynamical model relating Cp to Ce
and Ce to measurable eect.

Figure 5.21: Representation of a possible patch to place on the arm for example, to
administer the somatosensory stimulus, with multiple sources of contact heat or lasers,
that would be activated randomly, and irregularly in time.

One issue that in this phase of the study did not assume great importance, since the
objective was to inspect the feasibility of the method, was the fact that the SEPs waves
were obtained punctually, not in continuous mode.

The problem with this method

is that it requires a continuous stimulus application in order to be able to record the

evoked potential and correspondent wave characteristics, which may lead to habituation
and SEPs bias depression. Also the stimulus source is important, the best stimulus as
described in the introduction, would be a heat source, either CHEPs or laser.
problem with this approach is the possible occurrence of skin lesions.

The

A solution for

this problem would be to have dierent sites of stimulation, that would be activated

189

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

randomly, in unevenly spaced in time sequences in order to prevent also the anticipation
eect (see Figure 5.21). The signal processing technique used in the SEPs averaging
would have to take this into consideration, using a time-locked search of the SEP wave.
In a next phase of the technique development, a continuous measurement should be
obtained, with automatically detection of the wave features without the need of human
interference, suchlike what happened in this study [131].

5.8 Summary
In this chapter the use of evoked potentials with special focus on somatosensory evoked
potentials, has been reviewed. Evoked potentials are elicited responses to a well dened
and precise in time stimulus. The use of evoked potentials in clinical practice is fully
described with special interest in the evaluation of sensory conduction in high risk
procedures, and as diagnostic tests in some pathologies aecting sensory conduction.
Sensory evoked potentials wave characteristics are aected by several factors including genetic predisposition, leading to wide interindividual variability in sensing thresholds and amplitude responses. Anaesthetic drugs interfere with evoked potentials'
waves, with dierent impacts according to the drug site of action and eect.
The use of evoked potentials in anaesthesia is not a new concept, since auditory
evoked potentials have been used to assess the hypnosis state of a patient during general
anaesthesia, with a commercially available monitor translating the patient's hypnosis
state; some authors have initiated the study of somatosensory evoked potentials in
the search for a tool to assess nociception, nevertheless these studies were made with
dierent technological means and never proceeded to the nal step of designing a tool,
fully describing the balance between nociceptive activation and drugs attenuation.
Understanding the way noxious stimulus are processed by the human body allows a
more direct assessment of nociceptive responses, since the stimulus conduction and evoked responses are directly related to the stimulation intensity, and drugs' attenuation.
The use of SEPs in the assessment of noxious stimulation and drugs attenuation in
healthy adult volunteers was investigated. A clinical protocol was designed and submitted for institutional approval. Following institutional approval and written informed
consent, data were collected and subsequently analyzed.
It was observed that both remifentanil and propofol had impact on SEPs amplitude
and latencies, with a correspondent change in the NRS evaluations. The results suggest

190

5.8 Summary
that the SEPs waves contain information on the balance between noxious stimulation
and drugs' attenuation.
A great interindividual variability in stimulation intensity and response to individually adjusted thresholds were also observed. After analyzing the data the amplitude
and latencies were normalized using the observed SEP in response to the PT with no
drugs in the system, which led to a decrease in interindividual variability.
A new method combining both the amplitude and latency of the SEPs is proposed, taking in advantage the known inverse proportionality of the two measures with
increasing drug doses. Normalization also introduces improvement since the ratio interindividual variability is decreased.
Somatosensory evoked potentials wave amplitude and latency were related to the
individual nociception and drug attenuation balance, and may be used to translate the
individual Noc/ANoc balance in uncommunicative individuals.

191

5. EVOKED POTENTIALS - ACTIVE NOCICEPTION MEASURES

192

Chapter 6
On Nociception Control

It's not what happens to you,

but how you react to it that matters.
Epictetus

This chapter addresses the Noc/ANoc balance control, and discusses the benets and
drawbacks of automatic administration of anaesthetic drugs, as well as possible methodologies to approach this problem's specic demands. It is also proposed a nonlinear
control technique to be applied to the hypnosis control, since it poses as a parallel problem to the Noc/ANoc balance control, and it already has validated monitors
[55, 218, 219].

6.1 Introduction
The use of automation in medicine is gaining increasing interest due to the potential
improvement it may bring to patient care. Although the application of these techniques
in life sciences is not recent nor a novelty [220, 221], technological advances have facilitated its application in clinical environments such as an operating room, to improve
performance in complex tasks [220].

193

6. ON NOCICEPTION CONTROL

Figure 6.1: Schematic representation of the analogy between the automatic pilot systems
in the airplane, and the anaesthesiologists' action during general anaesthesia: induction,
maintenance, noxious stimulation regulation and recovery.

Anaesthesiology is one of the medical specialties that has taken an interest in this
issue, specially for the automatic control of drugs' administration during general anaesthesia, in the search of the optimum drug dose titration, which may lead to cost
reduction and shortened post-operative recovery phases [222]. The introduction of new
monitoring devices, new faster acting and excreted anaesthetics, as well as the development of fast and controllable actuators for the administration of these drugs (e.g.
remotely controlled infusion rate in electronic syringe pumps), has allowed the development of automatic control systems for the administration of anaesthetic drugs [121, 127].
Researchers have long addressed the issue of closed-loop control in anaesthesia, proposing dierent techniques for the administration of muscle relaxants [124126], hypnotics [55, 121123, 127] and analgesics [21, 127]. Nevertheless, the introduction of these
techniques carries safety and ethical issues, that need to be analyzed with care before
a fully automatic system may take over the control of drugs' administration [129].
The idea of closing the loop to provide a more safe and accurate anesthesia is
appealing, both for health professionals and patients, but machines are blind to a variety
of unexpected events that may occur, and often do, in the daily clinical practice, and
may not be disregarded. Anaesthesiologists' action is usually compared to piloting an
airplane with the automatic pilot systems. This is so mostly because of the known fact
that both airplane piloting and anaesthesiology are at the top when it comes to safety
regulations (check lists), safety protocols, and error issues, because both imply heavy
technology and complex decision processes taking in account many variables.

Also

because of the similarities between a ight and anaesthesia: take o of a plane is like
induction of anaesthesia, the lack of control over one's life in a ight being comparable

194

6.1 Introduction
to loss of consciousness; as an airplane passenger knows of the risks, however remote
of crash of the aircraft, so the patient submitted to anaesthesia is aware of the risks
of never regaining consciousness.

The ight itself is like the maintenance phase of

anaesthesia, with many variables potentially causing turbulence and requiring action to
maintain stability. Landing is like the recovery phase of anaesthesia and regaining of
consciousness, a smooth landing being comparable to rapid and stable ROC. Airplane's
pilots are responsible for the lives of many people on each ight, anaesthesiologists are
responsible for the life of just one.

As passengers are strangers which the pilots do

not get to know, patients subjected to anaesthesia develop a one-to-one relationship

with the anaesthesiologist.

The version of anesthesiologists taking the airplane into

the sky (loss of consciousness), maintaining the plane in ight height (maintenance of
anaesthesia), leading the safe landing of the airplane (recovery of consciousness), may
be extended by introducing the air 'bumps' during ight to translate noxious activation
in response to noxious stimulation, or unexpected events such as blood loss, which act
as interferences in the human system altering pharmacokinetics for example (see Figure
6.1). Figure 3.11 in Chapter 2 is representative of these interferences, varying both in
intensity and duration; the anaesthesiologist needs to anticipate/react in accordance,
to keep patient's homeostasis.

Automation has taken an important role a long time

ago and is currently widely accepted. Although researchers have developed and used
experimentally automatic administration of anaesthetics in the clinical setting, not a
single automation device has been commercialized.

As of today only a few hundred

patients worldwide have received anaesthesia using automated devices.

When developing a system to automatically administer drugs, one needs to understand the implications of such a system, and the required specications for it to be safely
used in clinical practice.

The rst question that arises in this discussion is: will the

anesthesiologist be replaced by a machine in the future? As in all applications, the road

from the rst automation techniques applied to solve a problem, to the replacement of
the original control agent, is long. One must guarantee that the use of the automatic
control is by far more ecient than the original control agent, in this case the anaesthesiologist. Clearly a machine may respond more rapidly or with more accuracy in
drug adjustment, nevertheless currently machines are unable to predict and take into
account dierent variables that need to be assessed through patient's observation and
environment control [123]. O course one may imagine the possibility of implementing
the anaesthesiologist knowledge and common sense rules in an articial intelligence

195

6. ON NOCICEPTION CONTROL
network, that could reproduce the anaesthesiologist's decision processes responding to
every possible scenario with the best solution.

The problem in this approach clearly

resides in the consideration of every possible scenario, and in the response to particular
conditions, not considered in the design phase. Nevertheless, this is a problem both with
automation and with the original human controller, since anaesthesiologists rely on the
experience and learning to deal with similar cases, and are therefore prone to (human)
error. The problem with a full automatic control resides in the danger of ignoring important factors that may be present in the scenario and, to achieve the minimum error
possible, to over or sub-dose the patient in that specic condition. A straightforward
solution is to never allow the system to exceed the maximum or minimum drugs' recommended doses, nevertheless this solution may not be sucient to guarantee patient
safety, due to wide variability in individual responses to drug administration.
From manual control to fully automatic drug administration, dierent levels of utility
and feasibility may be considered:

1.

Advisory system:

the anaesthesiologist manually controls the administration of

a drug, in accordance to the recommended doses provided by the advisory system,

which uses the monitored physiological data and drug consumption to indicate the
appropriate drug dose in that specic situation.
2.

Advisory/automatic system:

system controls drug administration based on

the physiological monitored data and drug consumption, but the anaesthesiologist
may override the system, to manually control the drug doses, and switch back (or
not) to automatic control.
3.

Automatic system:

the system automatically controls drug administration,

without the clinician's intervention, based solely on the information gathered by

the monitoring devices and administered drugs/concentrations.

These three levels dier in concept and control methodologies. The Advisory System
will only serve as an advisor to recommend drug adjustments and direction's to take to
the anaesthesiologist; it does not require the anaesthesiologist to follow the advice, and
it is prepared to simply continue on predicting the best solution to lead the patient into
optimum conditions (reference target).
The Advisory/Automatic System level is conceptually dierent, since it starts as an
automatic control system, but it may be overriden by the clinician if he/she nds it

196

6.1 Introduction
necessary; the concept is to allow the clinician to adjust drug doses according to his/her
clinical perception/decision, with the system working during this period as an advisory
system.

Resulting from this intervention, some technical problems may arise: when

the clinician takes over control, no problem should arise nevertheless, the system must
be robust enough to admit the return to the automatic control mode.

The problem

is that conceptually the anaesthesiologist considers that the automatic control was not
satisfactory enough (changes in the patient conditions not considered in the controller
design and patient modelling), and he/she overrides the system to lead the patient to
a new optimum condition. This is feasible, the problem residing in the return to the
automatic control, since the controller is blind to the conditions update and may present
an abrupt reaction to this change, over or sub-dosing the patient to guarantee a rapid
return to the target conditions. This problem may be solved using three strategies:

not allowing the return to the automatic control, and solely allow for manual
control from that point on;

limiting the control action when returning to the automatic control, in order to
avoid sudden control actions (this solution has limited application, since the controller will eventually lead the patient to the previously dened state of operation);

changing the state of operation (controller target), guaranteeing that the the system will not overshoot or undershoot when returning to the automatic control
mode, maintaining the new route dened by the anaesthesiologist.

The Automatic System would be regarding automation as the nal objective, a

controller considered to be robust and ecient enough to deliver general anaesthetics
autonomously, without human intervention.

Such a system must be shown to be by

far more ecient than the original control agent, the anaesthesiologist, before it can be
implemented in clinical practice.
So far the generic problems with automatic control systems in the context of this
application were discussed:

nevertheless there are some conceptual issues in general

anaesthesia genesis that need to be approached in the design of a robust controller regarding the patient as a whole. The next subsection will focus on the concept of general
anaesthesia, the proposed methodologies to control the administration of anaesthetic
drugs, and the future of control systems in the eld of anaesthesia.

197

6. ON NOCICEPTION CONTROL

6.2 General Anaesthesia Triad Control

The more favored approach in the scientic community has been to control each component of the general anaesthesia triad separately [21, 121, 223225], in the search for the
optimum drug dose for each component (see Figure 1.2). However, if one looks at the
interactions between the administered drugs to produce a determined eect, and their
impact on each other, either directly or indirectly (see Figure 1.18), a multi-variable
control approach seems more reasonable in the control of anaesthetic drugs' administration, corresponding to a

Multiple-Input and Multiple-Output

(MIMO) control problem.

Some authors have followed this route in the search for a multi-variable control of general
anaesthesia [127, 221, 222, 226].
First there is the known dynamic relation between the analgesic and hypnotic drugs,
and their synergism, leading to lower hypnotic consumptions when in presence of an
analgesic (see Equation 1.7). Next, the impact that the hypnotic drug may have in the
noxious responses, which was observed in the volunteers study, with the decrease in
the pain NRS evaluations in response to increasing propofol doses. This has also been
referred in some pharmacological studies [34, 46, 216, 217].

Finally, the muscle rela-

xant, which may be controlled independently from the remaining general anaesthesia
components, but may interfere in the data collection quality for the monitoring of the
remaining components, as for example, in a paralyzed patient, where the risk of outliers
occurrence in data collection is diminished, meaning the risk of obtaining biased erroneous values due to muscle activity and artifacts is decreased (e.g. frontal EEG derived
indexes contamination by frontal EMG activity) [7]. Figure 6.2 is a schematic representation of the possible interactions between the considered anaesthesia components and
corresponding drugs, be it on each other (direct interference) or as indirect interference,
through the introduction of erroneous measurements in the monitored physiological
variables used to assess the patient's state.

Observing the gure, muscle paralysis is

the only component that may be controlled independently from the other two, since
those two do not interfere directly or indirectly in the monitoring and control of muscle
paralysis. Analgesia and hypnosis, may both be contaminated by movement artifacts,
depending on the variables used to asses the patient's state concerning these general
anaesthesia components, and also they interfere with each other. Attempts to control
the hypnosis component independently from the analgesia have been done, using spinal
analgesia, and guaranteeing independence from noxious activation interferences [227].

198

199

variables related to noxious activation used to monitor the analgesia component.

hypnosis and analgesia components, and nally dash-dot arrow represents the impact that the hypnotic drug may have on the

dashed arrows represent the impact that the paralysis state of the patient has on the monitoring devices used to assess the

indirect interference on each other: line arrows represent the synergy relation between the hypnotic and the analgesic drugs,

Figure 6.2: Schematic representation of the general anaesthesia triad, with corresponding anaesthetic drugs, and their direct or

6.2 General Anaesthesia Triad Control

6. ON NOCICEPTION CONTROL
Another way to minimize outer interferences is to have the patient totally paralyzed.
This was unpractical due to longer recovery times, leading to unnecessary drug and
time consumption, but recently a new drug capable of rapid full reversal of deep levels
of muscle paralysis has been introduced (sugammadex). It may be used in the future in
daily clinical practice to allow maintaining patients fully paralyzed until the end of the
procedure, and rapidly reverse paralysis. This might allow more reliable hypnosis and
analgesia measurements, and improvements on the way these components are assessed
and controlled.
The control of each component of anaesthesia separately is motivated by the lack
of an adequate measure for the analgesia component, and also of the description of the
dynamic interactions between the three components, which needs more research to be
fully explained, although some attempts have been made in the multi-variable control
of anaesthesia [127, 220, 221].

Monitoring devices are widely available and perform

well to assess muscle paralysis and hypnosis. For muscle paralysis, EMG, mechanomyography (MMG) and acceleromyography (AMG) are used to translate the patients state
in response to an electrical stimulus, EMG being the less prone to mechanical interference in signal collection [222], which is important from the control engineer's point
of view.

For hypnosis, monitors using EEG and AEP are currently widely accepted

into clinical practice, and validated in the translation of the patient's state regarding
this component. Several studies used these monitors to close the loop in the hypnotic administration. For the analgesia component research in the monitoring devices is
currently active, and the search for a tool to adequately translate the patient's state is
on. Nevertheless, as referred before, variables associated to the Noc/ANoc balance have
been used to assess the patients state, and even control the analgesic administration
[21, 127, 224, 228]. However much work needs to be done: rstly in the validation of
such a tool, and secondly in the search of the dynamic relations between drug dose and
eect, as well as the dynamical relations between the general anaesthesia components.
In Chapter 1 a controller structure to control the general anaesthesia triad is proposed in Figure 1.18, taking into consideration the possible direct or indirect interferences
between components. Although conceptually this approach is appealing, there is much
work to be done in the description of the system dynamics, and of the relations between
the components and of possible monitoring interferences.
The Noc/ANoc balance control is conditioned by events of noxious stimulation that
the anaesthesiologist usually anticipates by adjusting drugs' doses, minimizing respon-

200

201

multiple-input and multiple-output (MIMO) approach for the general anaesthesia triad.

Figure 6.3: Schematic representation of an automatic control system for the administration of general anaesthetics, using a

6.2 General Anaesthesia Triad Control

6. ON NOCICEPTION CONTROL
ses associated with the introduction of the stimulus.

The introduction of dierent

stimulation phases may also be taken into account, using some predened stimulus intensities for the usually expected surgical and anaesthesia related noxious stimuli (see
Table 3.2). This information may be manually introduced into the system, much like
what the anaesthesiologist does in current practice, altering the conditions and guaranteeing anticipated response; otherwise the analgesic adjustment will only be done
a posteriori, in response to the noxious activation, which is in many cases undesirable
and may lead to arousal episodes (intense noxious stimulation).

Figure 6.3 presents

the MIMO controller approach, taking into account the dynamic relations and interference between components. This controller administers drugs and maintains patient
homeostasis, considering the general anaesthesia triad, and possible inputs provided by
the anaesthesiologist (e.g.

PK alterations due to blood loss, or predicted changes in

stimulation intensity).
To develop a fully automated system for general anaesthesia administration and
control, the rst task is to understand and model the relations from drug dose to nal
eect, taking into consideration possible drugs' interactions. Pharmacokinetic models
are currently available for most of the currently used muscle relaxants, hypnotics and
analgesics. Such models describe the relation between administered drug dose and drug
concentration on the organ of eect. However, dynamic models describing the relation
between the eect-site concentrations and the expected monitored value (e.g. Hill equation), were addressed in only few studies. Taking into account possible drugs' synergy
(bivariate Hill equation), what lacks is the in depth study of the analgesia component,
translating the possible dynamical relations and interferences in the monitored variables (see Figures 1.13 and 1.18). Once the analgesia component may be translated by
a good enough model of the population, the controller may be designed inside a MIMO
framework.

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

Although this thesis focus mostly on nociception measurement and control in anaesthesia, control of the hypnotic component of anaesthesia is here addressed [55, 218, 219].
This is done because the two components have some common characteristics, but also

202

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

because control of the hypnosis component is much more developed and some can be learned from it. Part of the research work of this thesis was dedicated to the development
of a control simulator, in a study using the hypnosis component index Entropy. The
aim of this study was to present a robust technique to the control of the administration
of an anaesthetic drug, taking into account the inter-patient variability, and expose a
parallel problem to the nociception control, discussing the points of contact between
the two and the implications of the proposed approach.
The main objective of the anesthesiologist is to provide the correct amount of drug
for each component (hypnotic, analgesic and muscle relaxant), to achieve an adequate
state in each component.

As described in Chapter 1, several monitors are currently

available for the anaesthesiologist to asses the patient's hypnosis state. These monitors
use the EEG and derived signals, to transform a signal that is dicult to analyze, into a
simple index varying from 0 (isoelectric EEG) to 100 (awake patient) bringing a useful
tool in preventing over-dosage, that might be associated with long-term consequences,
and in preventing traumatic awareness episodes [13].
The hypnosis index used in this study was the SE, from the Entropy Module (Datex-

R

) [18], described in Chapter 1.

Ohmeda

Several models have been used to predict the drugs' eect on the human body
[39, 50, 51, 229], but because of the known interindividual variability, the use of average
models as predictors of the drug eect has limited performance.

Knowing this, the

use of sliding-mode control techniques seems suitable for the control of the drug administration, maintaining robustness to patient deviations from the average population
model.
The objective of this study was to apply sliding-mode control techniques to the model
structure of the hypnotic drug in the patient system, and consequent repercussion in the
hypnosis index used [128, 230], using a

Singe-Input and Single-Output (SISO) approach.

6.3.1 State Entropy Modelling

In this study the relationship between administered drug (propofol) and the body was
modelled considering three phases: the pharmacokinetic phase (relation between infused
dose and plasma concentration), the pharmacodynamic phase (plasma concentration to
eect-site concentration), and the translation between eect-site concentration and the
measurable eect (SE). The three phases are addressed in the following sections.

203

6. ON NOCICEPTION CONTROL
6.3.1.1 Pharmacokinetic and Dynamic Modelling
Over the years studies have been carried out to describe the drugs' course and metabolism in the human system [39, 4951].

Experiments with drug administration and

subsequent plasma collection and eect analysis, were made to adjust and obtain the
best model structure and parameters, trying to predict the plasma and eect-site concentrations. One of the key studies was performed for the hypnotic drug propofol, by
Schnider and co-workers [50].

This model used a three-compartmental model struc-

ture (see Figure 1.13) to describe drug plasma concentrations (pharmacokinetic phase).
This study presented the relations between individual patient demographics (gender,
age, height and weight) and the ow parameters of the model, as shown in Table 1.8.
The average and the standard-deviation of the estimated parameters

i (i = 1, ..., 11)

are shown in Table 1.9.

The dynamic phase model is also shown in Figure 1.13.

The

ke0

parameter was

obtained from the relation between the point where the maximum eect is achieved,
corresponding to the equilibrium between plasma and eect-site concentrations.

6.3.1.2 Hill Equation

To translate the relation between the eect-site concentration and the observable clinical
eect (see Figure 1.13), in this case the SE value, a Hill equation (Equation 1.3) was
used.
The parameters of the Hill equation were chosen according to observed clinical practice and SE characteristics.

The

ke0

parameter was obtained by Ellerkmann and co-

workers, in a study with SE monitoring, taking in consideration the delay associated

with EEG processing, characteristic of the monitor used (see Table 6.1) [229].

6.3.1.3 Model Representation

The system's dynamic equation may be written using the information described in the
previous sections. The complete system is represented by Equations 6.1 and 6.2.

k12 k13 k10 k21
k31
0
m 1 (t)
m 2 (t)
k12
k21
0
0

m 3 (t) =
k13
0
k31
0
ke0

0
0
ke0
Ce(t)
V1

m1 (t)
1
m2 (t) 0

m3 (t) + 0
Ce(t)
0

r(t)

(6.1)

204

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

Table 6.1: Estimated parameters for the Hill equation using State Entropy monitoring
[229].

State Entropy
Average Standard-Deviation

Hill Equation
E0
EC50 (g/ml)

86.89

1.5

1.9

1.5

4.5

0.48

0.37

Pharmacodynamics (Ellerkmann et al)

ke0 (min1 )


E(t) = E0 1
where

r(t)

Ce(t)

Ce(t) + EC50


(6.2)

is the infusion rate. See Chapter 1 for more detailed description.

6.3.2 Feedback Linearization

Most of the chemical and physical phenomenons observed are of non-linear nature, and
even though linear control techniques are well developed, in practice the predened conditions can not be met. One example of this non-linearity is the way of drug processing
and eect in the human body. Knowing this, several non-linear control techniques were
developed to meet each specic problem demands.
Feedback linearization is a non-linear control approach which central idea is to transform a non-linear system into a fully or partially linear system, in such a way linear
control techniques may be applied. This is achieved by simplifying system's dynamics
transforming the original system into a simpler equivalent. The input-output linearization may be done in three steps: dierentiate the output
choose

y,

until the input

appears;

such it cancels the non-linearities and that guarantees convergence; study

internal dynamics [128].

Let
of

denote the system's relative order, meaning the number of dierentiations

until the input

appears.

Considering our system, the system model may be

represented by Equations 6.3 and 6.4.

x = f (x) + G(x)u

205

(6.3)

6. ON NOCICEPTION CONTROL

y = h(x)
with

x, u, f , G

and

h(x)

(6.4)

dened by Equations 6.5, 6.6, 6.7 and 6.8.

m1 (t)
m2 (t)

x=
m3 (t) , u = r(t)
Ce(t)

(6.5)

k12 k13 k10 k21

k31
0

k12
k21
0
0
x
f (x) =

k13
0
k31
0
ke0
0
0
ke0
V1

(6.6)

1
0

G(x) =
0
0

h(x) = E0 1

(6.7)

x(4)

x(4) + EC50

Let us generically dene the Lie derivative

La b

of


(6.8)

with respect to

by Equation

6.9.

La b =

b
a
x

(6.9)

The system under study has one input, one output, four state variables, and relative
order

r = 2; after the system feedback linearization comes the relation given by Equation

6.10.

y (2) = L2f h + LG (Lf h) u

(6.10)

The system may be represented by Equation 6.11

y (2) = f (x) + G (x)u

where

and

(6.11)

are given by Equations 6.12 and 6.13.

G (x) = LG (Lf h) =
x

206

h
f
x


G

(6.12)

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

f (x) =

L2f h

=
x

h
f
x


f

(6.13)

This method may be applied for dierent non-linear control problems, nevertheless
there are some important limitations: it can not be employed to all non-linear systems;
the state must be measured; and robustness is not guaranteed in the presence of parameter uncertainties or dynamics not incorporated in the model. For the last issue,
robust control techniques and adaptive control techniques have been introduced, such
as sliding control, which will be presented in the following section.

6.3.3 Sliding-Mode Control

Despite the eorts in the systems' modelling process, theoretical models used to describe
and approximate reality, are rarely capable of incorporating the eective full dynamics
of the system, or contain uncertainties regarding the model parameters.

In the case

of the drug pathway in the human system, this uncertainty is evidenced by the known
interindividual variability observed in response to drug administration, conditioned by
patients' demographics (age, weight, height and gender), but also by the clinical background, and associated diseases.
Imprecisions found during the modelling process may be classied in two distinct
groups:

Structured or parametric uncertainties, derived from uncertainties of the terms

included in the model;

Non-structured uncertainties or unmodelled dynamics, derived of system's order

underestimation.

The approach methodology of sliding-mode control comes from the perception that

th

it is much easier to control rst order systems than n

order systems.

A notation

simplication is introduced, in such a way the n-order systems may be in fact replaced
by rst order systems, demonstrating that in principle it is possible to obtain optimum
performance. Nevertheless, such performance carries high costs when it comes to control
action, being necessary to nd a compromise between performance and control action
energy.
Sliding-mode control techniques were applied in this simulation study to the hypnosis
component of general anaesthesia assessed by SE. The model parameters were obtained

207

6. ON NOCICEPTION CONTROL
from clinical studies, and the uncertainties described earlier introduced in the controller
design [128, 230].
The problem of sliding-mode control consists in the construction of an attractive
surface, guaranteeing that all trajectories outside the surface point towards it. Dening
the sliding surface by Equation 6.14.


s=
with

d
+
dt

r1
e

(6.14)

e dened by Equation 6.15, a positive scalar specifying the bandwidth, and r=2.

e = y yd
with

yd

(6.15)

being the desired trajectory for the SE value.

We may write Equation 6.14 as given in Equation 6.16.

s = y (1) yr(1)
where

yr

is computed from

and

(6.16)

yd .

The objective is to track the desired

yd , assuming that the system has the parametric

uncertainties given by Equations 6.17 and 6.18, associated to the uncertainties of the
model, where

and

are the estimated

and

f ,

respectively.

On the input

associated G the parametric uncertainty is considered to be in a multiplicative form,

and on

in an additive form.

(x), || D
G (x) = (1 + )G

(6.17)





f (x) f (x)

(6.18)

In the presence of parametric uncertainty, the system control law is given by Equation 6.19.


1 h
i
(x)
u= G
yr(2) f (x) k sgn(s)
with

G (x)

and

(6.19)

given by equations 6.12 and 6.13, the sign function dened by

Equation 6.20, and the switching gain

given by Equation 6.21.

208

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

1, s < 0
0,
s=0
sgn(s) =

1,
s>0

(6.20)



 (2) 
(1 D)k = + D yr f (x) +

(6.21)

.
In order to avoid excessive control action, the sign function may be substituted in
the control law by the saturation function given by Equation 6.22, where
boundary layer width around the surface

s;

denes the

considering the signal characteristics

was

dened as 5.

1, s
s/, < s <
sat(s/) =

1,
s

(6.22)

To obtain the complete control law some of the constants need to be determined.

was assumed to be 0.1, to include possible deviations from the modelled

associated), considering the multiplicative form as described in Equation 6.17.

alizing from a dened point (s

|s(t = 0)| / ,

with

6= 0)

(input
Initi-

the surface is reached in a nite time inferior to

strictly positive, and then converges exponentially to the desired

point, with a time constant given by

(r 1)/.

Considering that the patients are fully

awake at the beginning of the infusion (y=86.89), and knowing that for a slow induction
of 200 ml/h loss of consciousness occurs around 2 minutes following infusion start (120s
to reach y=50), we have
Regarding the

= 0.3.

parameter, the adjustment of this constant may be made experi-

mentally, nevertheless there are values that limit its value: in mechanical systems it
must be lower than the resonance frequency of the system; it depends of possible neglected delays in time
rate

vS

TA

as given by Equation 6.23; and it is dependent of the sampling

(Equation 6.24), being the bandwidth

the lower of the three.

1
3TA

(6.23)

1
S vS
5

(6.24)

Assuming that the neglected delays are in the order of

rate

vS = 1/5Hz , = 1/30Hz . TA

TA = 10s,

and the sampling

was set to 10s to include possible delays in the

209

6. ON NOCICEPTION CONTROL
monitoring device output (corrupted signal with extended EEG window of analysis),
and regarding the fact that the

ke0 ,

described in Table 6.1, already takes into account

the monitoring device characteristics.

was approximated on-line using the deviations

described before and depending on the system's state.

6.3.4 Methods

Figure 6.4: System general structure and feedback for the sliding-mode controller.

The model and corresponding controller were implemented in Simulink, MatlabR2007a

(see Figure 6.4).
The controller was simulated using both the sign and the saturation functions, in
order to evaluate the response to a step and a sinusoidal references.
Given the desired application, the reference signal was a step set to 50, the central
value of the desired target range (Equation 6.25). A sinusoidal reference was also used
to test the controller robustness to changes in reference target, varying between 87 and
0.


yd (t) =

86.89, t < 50
50,
t 50

(6.25)

The controller activity was initialized by the usual infusion rate of propofol used
for slow induction of anesthesia (200ml/h or 556 g/s), and restricted to the interval
[0-556] g/s, in consistency with the system characteristics and limitations.
To test the controller robustness, the designed controller was applied to three theoretical patients, considering a nominal patient, male, 170 cm, 70 kg and 20 years old:

Nominal Patient:

The patient with the average expected values of the model

parameters, used in the controller design.

Minimum Patient: Theoretical patient with all the minimum model parameters
expected, with the parameter deviations considered in the controller design.

210

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

Maximum Patient: The same as before, but with the maximum model parameters
expected.

These patients were dened in this way in order to take the controller to the extremes of the modelled uncertainties, and to verify if it would be robust enough, while
guaranteeing the achievement of the reference target (for more information refer to
Appendix E).

6.3.5 Results
Results of the application of the designed controller are divided according to simulated
patient, and reference target.

Figure 6.5: Control activity for the nominal patient using the sign function: on top SE
and SE reference, below the error sequence, and on bottom the infusion rate (ml/h).

Because of the excessive control action, the sign function was replaced by the saturation function in the control law. The control action is smoothed, which is preferable
for the clinical application. Figure 6.5 presents an example of the control activity for
the nominal patient and a sinusoidal reference.

6.3.5.1 Nominal Patient

The controller response is shown in Figure 6.6a, with a smooth control action achieving
the step reference in an acceptable time (slow induction).
As response to the sinusoidal reference the control action is adjusted to obtain the
lowest possible error, while conditioned by the patient's own response. It can be seen

211

6. ON NOCICEPTION CONTROL

(a) Step Reference

(b) Sinusoidal Reference

Figure 6.6: Control activity for the nominal patient using the saturation function: on
top SE and SE reference, below the error, and on the bottom the infusion rate (ml/h).

that for the recovery phases (ascending reference), the SE is not able to follow the SE
reference, due to the patient modelled excretion (see Figure 6.6b).

6.3.5.2 Minimum Patient

(a) Step Reference

(b) Sinusoidal Reference

Figure 6.7: Control activity for the minimum patient using the saturation function: on
top SE and SE reference, below the error, and on the bottom the infusion rate (ml/h).

The control action for the minimum patient had a satisfactory behaviour with small
error and small overshoot. The overshoot may have been caused by the patient deviation

212

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

(minimum parameters) from the nominal patient. This may be the same reason why it
takes a little more time to achieve the reference signal with the maximum infusion rate
delimited (see Figure 6.7a).
The results showed (see Figure 6.7b) a good performance while the reference signal
was decreasing, and higher error on the ascending phase due to patient's metabolism
modeled parameters, causing a slower excretion. Nevertheless this increased error following the reference is due to the modeled patient characteristics, not depending on the
control action, as it may be observed.

6.3.5.3 Maximum Patient

(a) Step Reference

(b) Sinusoidal Reference

Figure 6.8: Control activity for the maximum patient using the saturation function: on
top SE and SE reference, below the error, and on the bottom the infusion rate (ml/h).

Figure 6.8a shows the controller action and performance for the maximum patient.
The reference is achieved with a very small error and with no overshoot.
Figure 6.8b represents the response to a sinusoidal reference with small error. Unlike
the nominal and minimum patients, the maximum patient has a high excretion rate,
which allows a good performance following the reference ascending phases.

6.3.6 Discussion
A robust controller was designed for the administration of the hypnotic propofol, using
the SE as feedback variable, taking into consideration the inter-patient variability observed in the population during the modelling process.

213

6. ON NOCICEPTION CONTROL
The excessive control action was smoothed by the saturation function in exchange
from perfect performance, which may in the future allow the implementation in an
automatic syringe pump for hypnotic administration. The control technique performed
satisfactorily showing robustness to deviations from the nominal patient, which is the
most attractive characteristic due to the known interindividual variability.

Although

with the minimum patient a small overshoot occurred when following the step reference,
this overshoot was insignicant when trying to achieve a target range of 40 to 60.
As for the sinusoidal reference response, and the error associated with the ascending
phase of the sinusoid, this exercise was done only to demonstrate the robustness of the
method, showing that depending on the patient's characteristics a longer time may be
required for recovery, independently of the control action. Indeed, the control action
was able to follow the reference SE very well in the descending phases, which is the
most interesting feature from the clinical application point of view.

Figure 6.9: Representation of the patient modelled system, with possible error inputs,
controller, and state observer, for an on-line implementation of the automatic control
of the hypnotic.

Results showed that sliding-mode control presents a ne choice to control the hypnosis component state, possibly resulting in a more secure and stable anaesthesia administration. It was evident that although the controller presented a good performance,
even in the considered extreme patients, other drugs and variables should be introduced, taking into consideration the interactions previously described (e.g.

interaction

with the analgesic), and controlling the remaining components of anesthesia.

Exter-

nal interferences like surgical stimulus may also be introduced in the model, in order

214

6.3 Hypnosis Control - A Parallel Problem to the Nociception Control

to achieve the best drug combination for patient stability, preventing intense noxious
stimulation to result in arousal episodes.

Besides extending the model to take into

consideration the possible interactions between general anaesthesia components, a state

vector sliding mode observer should be added, for possible on-line implementation and
clinical applicability of the controller, as shown in Figure 6.9. It should be highlighted
that the system and controller were implemented in continuous mode, in order for them
to be implemented in real time, the system and controller should be discretized.

Figure 6.10:

Schematic representation of the necessary steps to model the relation

between the administered analgesic drug dose and the measurable eect, when a tool
for this eect is fully validated in the translation of the nociception/anti-nociception
balance.

Although the controller was designed for the administration of the hypnotic propofol, this was done to demonstrate the technique applicability in anaesthesia.

The

presented technique may also be implemented for the control of the administration of
the analgesic remifentanil, taking into consideration the observed interindividual variability in the PKPD model adjustment (see Table 1.7) [38, 39], and of the Noc/ANoc
balance, as soon as a monitor is fully validated to translate the patient's state concerning this general anaesthesia component, and proven to be related to changes both
in stimulation intensity, and analgesic drug dose. Only when such a monitor becomes
validated to be capable of translating patient's Noc/ANoc balance, it will be possible to
construct the dynamic model from the plasma concentration to the measurable eect,
taking into account population deviations from the average patient model. Although
for remifentanil the translation from Cp to Ce is described by Minto et al [38, 39], the

ke0

parameter was obtained using an EEG derived measure to translate the opioid ma-

215

6. ON NOCICEPTION CONTROL
ximum eect. This may well not be the best measure to translate the opioid maximum
eect in the Noc/ANoc balance. Actually, recent research has deviated from the EEG
for the assessment of the Noc/ANoc balance. Moreover, dierent measures or monitors
incorporate dierent delays in the patient's state assessment, which should be taken
into account in the modelling phase. Figure 6.10 presents the steps necessary to obtain
a model describing the relations between administered drug and measurable eect for
the analgesia component.
Because this simulation study was only performed to evaluate the robustness of
sliding-mode control techniques in anaesthesia, the controller was not further developed,
nevertheless it should be highlighted that in order for the technique to be applied in the
administration of anaesthetic drugs an observer of the patients state must be designed.

6.4 Prediction and Advisory System

In Chapter 4 two dierent approaches for the Noc/ANoc balance measurement, based on
physiological variables linked to noxious activation, and part of standard monitoring,
were proposed.

The rst, an homeostasis index that allows identication of pattern

responses to noxious activation in variables shown to be related to noxious stimulation;

the second a physiological model describing the balance between drugs' administration
and stimulation intensity, able to estimate the perceived stimulus by the patient, and
translate the balance Noc/ANoc.
The proposed approaches look at the same problem from dierent perspectives,
allowing the identication of arousal episodes linked to noxious activation.

The two

methodologies were implemented in a tool that may be used on-line by anaesthesiologists

for early detection of arousal episodes and patient discomfort. The developed software
(see Figure 4.17) allows the identication not only of changes in the physiological signals,
but also the detection of the direction of the signal shift. This may be very useful to
determine the nature of the compensatory mechanisms occurring in the patient, for
example, if BIS value presents a decreasing trend, and drops below 40, than the system
could advise the anaesthesiologist to decrease the hypnotic dose.
There is a set of rules, that are the base of decision making by anaesthesiologists
in their daily activities, and that may be implemented in such a system, advising the
clinician in advance of the best action to take, taking into account the patient as a whole
and anaesthesiologists experience in similar cases. The idea is to take into consideration

216

6.5 Summary
clinical expertise to dene a set of rules, that would reproduce the anaesthesiologist's
action, and aid the clinician in the fast identication of the patient state, advising on
the action to take. Such tool would be in the context of an advisory system, as described
earlier in this chapter.

In future developments, this tool will be further enhanced to

include decision rules.

In Chapter 5 a dierent approach to the measurement of the Noc/ANoc balance
based on active measures is proposed. The described method allows the identication
of the individual basal sensitivity (painful threshold), the identication of the expected
baseline response, and the production of a normalized dimensionless index.

Besides

the dierences in the genesis of the proposed methods, passive or active, the last differs from the others because in this case it is possible to dene the baseline state of
operation, and measure the response to a known intensity stimulus.

This allows the

construction of comparable measures, to individually adjusted thresholds, that may be

used in the search of dynamic relations between drugs' doses and Noc/ANoc balance.
As described earlier a Noc/ANoc index must respond adequately to dierent noxious
stimuli intensities and drugs' attenuation, which was observed in Chapter 5. Although
much work is necessary in the development of a such a tool, it was shown that there is
important information contained in this signal, directly linked to the sensing process.
This may therefore be the key to a direct measurement of the Noc/ANoc balance, that
could bring a way of dening dynamic models as represented in Figure 6.10, and enable
the loop closing of the analgesia component.

6.5 Summary
The automatic control of general anaesthetics, have been under the anaesthesiology
scientic community attention since the early 1980's, with proposed strategies for the
hypnotic and muscle relaxant administration.
The application of automatic control to general anaesthesia must not be seen as
an attempt to substitute the anaesthesiologist, but rather as an assistant tool to the
clinician, which may allow him/her to direct his/her attention to other demanding tasks
in patient care, as well as an improvement in patient care.
The future of automatic control in anaesthesia lies on a MIMO approach, considering
the relations between drug's eects, and the possible monitoring interferences associated
with the patient's state. The approach is to regard the patient's state as a whole, while

217

6. ON NOCICEPTION CONTROL
taking into account each component individually and possible outer interferences such
as blood loss.
From the present to the development of a fully automatic control of the general
anaesthesia triad, more studies on the subject are required.

Models describing the

population trend and the assessment of the impact of possible synergistic eects, and
monitoring interferences, need to be developed.
A robust controller was developed for the hypnosis component.

Control of the

hypnotic component may be considered to be a parallel problem to the Noc/ANoc

balance control problem. This was developed as an academic exercise to demonstrate
the applicability of nonlinear robust control techniques to the automatic control of
an anaesthetic drug, which has already validated monitors. Robust non-linear control
techniques may be applied in the future to the control of the analgesic, similar to
what was demonstrated in this study, once a monitor is validated and dynamic models
available, with the inter-patient variability taken into account.
The homeostasis index and the perceived stimulus estimator developed in Chapter
4 were implemented in a software to monitor on-line the patient's physiological signals
related to noxious activation, individually and as a combination of these. A set of rules
may be implemented in the future to aid the clinician in the decision making.

This

system may be regarded as an advisory and alarm system, alerting the anaesthesiologist
to a change in the patients' state, improving detection of arousal or discomfort of the
patient, anticipating a possible unwanted scenario, and advising a drug change to lead
the patient back to homeostasis.

218

Chapter 7
Conclusions and Future Work

One never notices what has been done;

one can only see what remains to be done.
Marie Curie

Anaesthesia has three main components: hypnosis, analgesia and paralysis (immobility).
From these three components, analgesia is the only one for which an objective assessment
is not available.

Researchers have recently directed eorts towards the search of an

objective tool to measure the nociception/anti-nociception balance state of the patient.

To measure nociception, it is of major importance at rst to understand the entire
sensing process from the sense organs (the nociceptors) to the higher processing in the
brain. Moreover, tool capable of translating the nociception/anti-nociception balance
state of the patient must reect both stimulation intensity and drugs' attenuation.
The original research work of this thesis started with a clinical study in surgical patients under general anaesthesia. As part of this study, physiological variables usually
monitored in routine clinical anaesthesia practice like heart rate, blood pressure, peripheral pulse wave amplitude, EEG, and EMG were collected. Also, recently proposed
EEG (CVI) and cardiovascular (ANSSI) based indexes were analysed.

Patients were

subjected to dierent noxious stimuli (anaesthesia related and surgery related) as well
as to stepwise changes in analgesic drug concentrations (remifentanil).
In a second study, a survey was designed to obtain a tool to be used in the analysis
of the collected data in the clinical study with patients. This survey was conducted with

219

7. CONCLUSIONS AND FUTURE WORK

anaesthesiologists and allowed the construction of a scale to measure noxious stimuli
intensity. In this study fty seven anaesthesiologists rated a total of thirty ve dierent well dened noxious stimuli, using a numerical rating scale. The responses were
analysed using Rasch analysis, and a measure was constructed. This was an innovative
approach which resulted in the development of a new tool that may be useful in the
clinical and research settings. The developed method takes advantage of the anaesthesiologists experience in drug titration and stimuli anticipation, to provide an independent
linear measure of noxious stimuli intensity. It was compared with results presented in
pharmacological studies, where stimulus intensity was assessed by the amount of drug
necessary to block noxious activation.
Collected data during general anaesthesia allowed the inspection of passive physiological signals associated to noxious activation. Heart rate, blood pressure, photoplethysmography wave amplitude, BIS, BIS standard-deviation, EMG and EMG standarddeviation, were associated to noxious activation responses.

For the analysis of the

relationship between the measured physiological variables, the intensity of the noxious
stimuli, and the eect of the anaesthetics propofol and remifentanil, a new multi-variable
method was developed. This method diers from the existing, because it allows an objective assessment of the patient's global state and of each variable taken individually.
The new method consists of a homeostasis index: if the drugs are in steady-state, a
change in the homeostasis index reects a change in the nociception/anti-nociception
balance. This method was applied to data collected in the clinical study for 31 patients,
to extract the periods of simultaneous steady-state for both the input and the output.
This allowed a distinction between situations when the patient was in steady-state or in
a dynamic condition, leading to a new way of assessing the nociception/anti-nociception
balance. With this method a new index of anaesthesia homeostasis was developed.
The next step in the work done in this thesis was the development of a model of the
perceived stimulus during anaesthesia and surgery, dened as any event leading to a
change in the patient. This model related the input drugs and stimulus intensity to the
observed responses in the physiological variables associated to noxious activation. It allows for an estimation of the patients perceived stimulus (nociception/anti-nociception
balance), considering the physiological variables connected to noxious activation as inputs.
The two methodologies responded adequately to the introduction of precise noxious
stimuli, and dierent analgesic levels.

The homeostasis index and the perceived sti-

220

mulus estimator provide complementary information considering the nociception/antinociception balance of the patient, and were implemented in a system for on-line assessment.
The objective assessment of nociception using evoked potentials was addressed in
this thesis and a new index was developed. A clinical study in healthy volunteers measured the same physiological variables of the study in patients, and also somatosensory
evoked potentials (SEPs), while volunteers were subjected to painful stimuli before and
during dierent combinations of anaesthetic drugs propofol and remifentanil. It was observed that wave amplitude decreases and latency increases with increasing drug doses,
accompanied by a decrease in reported pain. A new index, based on wave amplitude and
latency values normalized for the individual painful threshold, was proposed as a new
objective active method of nociception/anti-nociception balance, producing normalized
measures, comparable between subjects.
The nal work in the thesis was the development of a new controller for the administration of anaesthetic drugs.

Technological and drugs' advancements have both

increased the interest in the scientic community for the automatic control of each component of general anaesthesia. Regarding to the analgesia component, some attempts
have been made for the resolution of this problem, nevertheless this is dependent on
the existence of a fully validated monitor for the analgesia component. Robust control
techniques should be applied in the future, taking into account both the known inter
and intra-individual variability observed in the population. It was demonstrated that
robust control techniques may be implemented in the control of general anaesthesia
components, by comparison to the parallel problem of the hypnosis component control,
incorporating the known patient model deviations observed in the population.
The search for objective methods for nociception assessment is still an open research
eld. The development of a tool capable of translating the nociception/anti-nociception
balance during anaesthesia is of great importance in order to provide the patient with
the best care, adequately titrating the dierent anaesthetic drugs. Hypnosis monitors
are nowadays validated and available for clinical use, aiding in the administration of
the hypnotic, preventing awareness and over-dosing and possible adverse eects.

If

one imagines that the patient may be conscious during the surgical procedure, the
scenario may become much worsened if the patient is also perceiving pain.

In this

case, like in many others, a tool to translate the nociception/anti-nociception balance

would allow an adequate drug titration for the three general anaesthesia components,

221

7. CONCLUSIONS AND FUTURE WORK

balancing the eects with individual information for each component, from the dierent
monitors.

The methods developed in this thesis aim at answering some of the open

questions in nociception objective assessment.

The validation of the methods here

developed may aid in a more adequate control of anaesthesia, leading to better patient
care, better recovery and possibly better long term outcomes (e.g.
of post-operative chronic pain).

lower incidence

Both passive and active methods developed in this

thesis adequately responded to dierent stimulus intensities and drugs' eect, necessary
premises to be considered as nociception/anti-nociception balance indicators.

Also a

software was developed with the passive nociception measures, providing information
on the patients homeostasis and location regarding the nociception/anti-nociception
balance, which may aid anaesthesiologists in a faster detection and better control of
nociceptive responses.

7.1 Future Research Directions

The future work arising from this thesis is closely related to the validation and extensive assessment of the developed measures of the nociception/anti-nociception balance
during anaesthesia.
As a rst study, the methodologies developed using passive measures of noxious
activation should be validated in a randomized controlled clinical trial, to assess their
ecacy in the detection of inadequate analgesia. Drugs' consumption and anaesthesia
global stability should be assessed to determine if the use of these indexes allow for
a more rapid detection of noxious activation and analgesic titration, in comparison to
standard clinical practice.
Detection of combined input/output steady-state periods may be used in the construction of surface synergy models, and allow the objective assessment of stimulus intensity, similarly to the stimulus assessment provided in previous pharmacological studies.
The new SEPs based index developed in this thesis was demonstrated to be related both to pain sensation, and anaesthetic drugs' attenuation. In this way, another
study that may be envisioned is the implementation of this method for an on-line and
continuous assessment of the nociception/anti-nociception balance during anaesthesia.
The characteristics of the noxious stimulus applied to obtain somatosensory evoked
responses should be addressed in the search for the best stimulus modality.

222

7.1 Future Research Directions

Besides validation of the tools developed in this thesis during anaesthesia, a pharmacodynamic model for remifentanil based on this indicator should be developed, taking
into account intrinsic index characteristics. If the developed index shows to adequately
respond to drugs' attenuation and changes in noxious stimulation, than it could be used
to close the loop for the analgesia component of anaesthesia.

223

7. CONCLUSIONS AND FUTURE WORK

224

List of Publications

The research done, during the course of this doctoral program, resulted in a number of
scientic communications and publications, listed below.

Awards
The developed work was recognized in the scientic community, and received two distinctions:

Zaldiar-Astor award, by the Grnenthal Foundation, Portugal 2010

Second best presentation in the Jornadas de Iniciao Investigao Clnica

(Introduction to Clinical Research Workshop, 2nd edition), by the Departamento
de Ensino, Formao e Investigao of the Centro Hospitalar do Porto, 2010

Publications

Ana Castro, Pedro Amorim, Catarina S. Nunes, Modelling State Entropy of the
EEG and Auditory Evoked Potentials - Hypnotic and Analgesic Interactions, 29th
Annual International Conference of the IEEE Engineering in Medicine and Biology
Society (EMBS), New York, NY, USA, IEEE. (ISBN: 978-1-4244-0787-3), 2007

Ana Castro, Catarina S. Nunes, Pedro Amorim, Fernando G. Almeida, Hypnotic

Administration for Anesthesia Using Sliding-Mode Control, 30th Annual Inter-

225

. LIST OF PUBLICATIONS
national Conference of the IEEE Engineering in Medicine and Biology Society
(EMBS), New York, NY, USA, IEEE. (ISBN: 978-1-4244-1814-5), 2008

Ana Castro, Fernando Gomes de Almeida, Pedro Amorim, Catarina S. Nunes,

A Wavelet Based Method for Steady-State Detection in Anesthesia, 31th Annual
International Conference of the IEEE Engineering in Medicine and Biology Society
(EMBS), New York, NY, USA, IEEE. (ISBN: 978-1-4244-3296-7), 2009

Diana Afonso, Ana Castro, Eduarda Amadeu, Pedro Amorim, Avaliao objectiva
da dor: condutncia elctrica cutnea, Dor 2010; 17: 5-8

Publications Undergoing Review

Ana Castro, Fernando Gomes de Almeida, Catarina S. Nunes, Pedro Amorim,

A review on objective methods for nociception/anti-nociception balance measurement and control, European Journal of Pain

Ana Castro, Pedro Amorim, Catarina S. Nunes, Fernando G. Almeida, Development of a scale to assess noxious stimuli intensity during anesthesia and surgery
using Rasch analysis, European Journal of Pain

Ana Castro, Pedro Amorim, Catarina S. Nunes, Fernando G. Almeida, Somatosensory evoked potentials may objectively translate nociception/anti-nociception
balance with anesthetics, Journal of Clinical Monitoring and Computing

Communications
Abstracts of scientic communications presented in international conferences are listed
here. Most were published in supplements of the Anesthesiology, European Journal of
Anaesthesiology, and the Journal of Neurosurgical Anesthesiology.

226

Ana Castro, Eduarda Amadeu, Ftima Martins, Paula S Couto, Pedro Amorim, Catarina S. Nunes, Propofol and Remifentanil Drug Interaction on the State
Entropy od the EEG: Surface Model, Total Intravenous Anesthesia - TCI World
Congress, Oral Presentations, p. 98, 2007

Ana Castro, Eduarda Amadeu, Paula S Couto, Pedro Amorim, Catarina S. Nunes, Bispectral Index and State Entropy Simulation with Propofol and Remifentanil Eect-Site Steering, Journal of Neurosurgical Anesthesiology, 20(4):

340,

Lippincott Williams & Wilkins, October 2008

Ana Castro, Eduarda Amadeu, Paula S Couto, Pedro Amorim, Catarina S. Nunes, Educational Software for Bispectral Index and State Entropy Simulation,
Anesthesiology A686, 2008

Ana Castro, Catarina S. Nunes, Pedro Amorim, Fernando G. Almeida, Propofol

Automatic Administration - Sliding-Mode Control, Total Intravenous Anesthesia
- TCI World Congress, Oral Presentations, 2009

Ana Castro, Catarina S. Nunes, Pedro Amorim, Fernando G. Almeida, Automatic

Control of Propofol's Eect on the State Entropy using a Sliding-Mode Technique
(a simulation study), European Journal of Anaesthesiology, Volume 26, Supplement 45, p.34, 2009

Ana Castro, Fernando G. Almeida, Eduarda Amadeu, Pedro Amorim, Catarina

S. Nunes, A Wavelet Based Algorithm for Steady-State Detection: Application in
BIS Guided Anesthesia, Anesthesiology A1293, 2009

Ana Castro, Fernando G. Almeida, Eduarda Amadeu, Pedro Amorim, Catarina

S. Nunes, Steady-State Detection Applied to BIS Guided Anesthesia - A Wavelet
Based Method, Journal of Neurosurgical Anesthesiology, 21(4), 2009

227

. LIST OF PUBLICATIONS

Carmen Gonzalez, Nadja Bressan, Ana Castro, Joaquim Mendes, Catarina S.

Nunes, Francisco Lobo, Pedro Amorim, Use of the fall in BP during induction
to model the reduction of propofol eect-site concetrations that would attenuate
the hemodynamic eects of anesthesia, Journal of Neurosurgical Anesthesiology,
21(4), 2009

Diana Afonso, Ana Castro, Maria Eduarda Amadeu, Pedro Amorim, Avaliao
objectiva da dor: Conductncia Elctrica Cutnea, 8 Convnio da ASTOR, 2010

Ana Castro, Catarina S. Nunes, Pedro Amorim, Fernando Gomes de Almeida,

Anaesthesia Steady-State Index: A Combined Measure of BIS, Blood Pressure
and Heart Rate, European Journal of Anaesthesiology, Volume 27, Supplement
47, p.64, 2010

Ana Castro, Pedro Amorim, Catarina S. Nunes, Fernando Gomes de Almeida,

SEPs may objectively translate nociception/anti-nociception balance with anesthetics, Journal of Neurosurgical Anesthesiology and Anesthesiology 2011 (accepted
for publication)

Ana Castro, Pedro Amorim, Catarina S. Nunes, Fernando Gomes de Almeida, Development of a scale to assess noxious stimuli intensity during anesthesia and surgery using Rasch analysis, Journal of Neurosurgical Anesthesiology and Anesthesiology 2011 (accepted for publication)

Manuel Pereira, Benno Rehberg-Klug, Ana Castro, Pedro Amorim, Inuence of

Diuse Noxious Inhibitory Control, Temporal Summation and Expectation on
Pain Perception in Healthy Volunteers, Journal of Neurosurgical Anesthesiology
and Anesthesiology 2011 (accepted for publication)

228

Appendix A
Remifentanil Recommendations in
Portugal

Remifentanil is an approved drug by INFARMED (Autoridade Nacional do Medicamento e Produtos de Sade, I. P. - National Authority of Medicines and Health Products) for clinical use in Portugal. The following text presents a summary of the guidelines of remifentanil use in Portugal, with the recommended dosages and potential side
eects [34, 35, 37].
Remifentanil is indicated as an analgesic for induction and maintenance of general
anaesthesia, and also as an analgesic for patients in the Intensive Care Unit, under
mechanical ventilation.
Because this is a very potent drug it must be only administered according to the
recommendations, in facilities equipped with cardiovascular and respiratory functions
support and monitoring, and by trained personnel in the use of anesthetic drugs, and
cardio-respiratory resuscitation maneuvers.
Remifentanil administration may be with a continuous perfusion of remifentanil,
with the infusion velocities according to the patients' body weight and the procedure.
The administered doses should be titrated according to the patients' response. Remi-

229

A. REMIFENTANIL RECOMMENDATIONS IN PORTUGAL

fentanil may also be administered using a Target Controlled Infusion. The TCI model
proposed to use in the drug specications is the Minto pharmacokinetic model, that
uses the age and the lean body mass as covariates in the model, with a recommended
dilution of 20 or 50 g/ml.

Table A.1: Recommended doses of remifentanil in co-administration with other anaesthetic agents for adult patients [35, 37].

Indication

Bolus Injection

Continuous Perfusion (g/kg/min)

(g/kg)

Anaesthesia Induction

1 (administered during a

Initial Rate

Limits

0.5-1

period of 30 s or more)
Maintenance of anaesthesia in ventilated
patients
N2O (66%)

0.5 to 1

0,4

0.1 to 2

Isourane (Initial dose

0.5 to 1

0,25

0.2 0.05 to 2

0.5 to 1

0,25

0.3 0.05 to 2

0.5 MAC)
Propofol (initial dose
100 g/kg/min)

Tables A.1, A.2 and A.3 show the recommended doses of remifentanil for a coadministration with other anaesthetic agents in adult patients, the infusion velocity and
the equivalent plasmatic concentrations by approximations of several stabilized manually perfusion of remifentanil when using target controlled infusion. When administered
as bolus, it should be administered in a minimum period of 30 seconds.
In the recommended doses, remifentanil decreases signicantly the hypnotic agent
need in order to maintain adequate anaesthesia. Knowing this the concomitant administration of hypnotic drugs and remifentanil must be balanced.

The available data

presented in the remifentanil recommendations only describe the recommended doses

230

Table A.2: Recommended infusion rates of remifentanil according to body weight [35,
37].

Infusion Rate

Patients Weight (kg)

(g/kg/min)

10

20

30

40

50

60

0.0125

0.188

0.375

0.75

1.125

1.5

1.875

2.25

0.025

0.375

0.75

1.5

2.25

3.75

4.5

0.05

0.75

1.5

4.5

7.5

0.075

1.125

2.25

4.5

6.75

11.25

13.5

0.1

1.5

12

15

18

0.15

2.25

4.5

13.5

18

22.5

27

0.2

12

18

24

30

36

0.25

3.75

7.5

15

22.5

30

37.5

45

0.3

4.5

18

27

36

45

54

0.35

5.25

10.5

21

31.5

42

52.5

63

0.4

12

24

36

48

60

72

for simultaneous administration of propofol or isourane. Due to the rapid start and
short duration of the remifentanil action, the administration should be titrated in increases of 25-100%, or decreases of 25-50% of the actual dose, every 2-5 minutes, in oder
to obtain the desired level of -opioid response.
Remifentanil decreases the need of other medications such as benzodiazepines, inhaled agents or hypnotic drugs. Because of the short duration of action, the post-operative
analgesia should be prepared in time with a long duration analgesic. Remifentanil is
not recommended for the post-operative analgesia in spontaneous breathing patients,
only for the immediate post-operative period until the long duration analgesia reaches
its maximum eect.
The bolus injection of remifentanil, due to its properties and side eects, is not
recommended in patients in spontaneous breathing.

231

A. REMIFENTANIL RECOMMENDATIONS IN PORTUGAL

Table A.3:

Infusion velocity of remifentanil and equivalent plasmatic concentrations

by approximations of several stabilized manually perfusion of remifentanil when using

target controlled infusion [35, 37].

Infusion Rate

Plasmatic Concentration

(g/kg/min)

of Remifentanil (ng/ml)

0.05

1.3

0.1

2.6

0.25

6.3

0.4

10.4

0.5

12.6

25.2

50.5

The administration of remifentanil in children is similar to the administration of the

drug in adult patients, with a correction factor for the drug dose by patients' weight,
although due to the lack of data in children with age below 1 year its' administration
is not recommended.
Remifentanil should be used by experienced personnel in the recognition of adverse
side eects of potent opioids in a totally equipped environment. Side eects of remifentanil are described in the following section. Remifentanil Side Eects

A.1 Remifentanil Side Eects

In the recommended doses remifentanil may cause

muscle rigidity.

Like in other

opioids this eect is related with the dose and administration velocity. Knowing this a
bolus should be administered for a period of 30 seconds at least. Muscle rigidity observed
should be treated according to the clinical state of the patient, if it occurs during the
induction of general anaesthesia, this condition should be treated using muscle relaxants

232

A.1 Remifentanil Side Eects

or more hypnotic.
Like all potent opioids, deep analgesia is accompanied by

sion,

respiratory depres-

what demands the use of remifentanil only in places with the right equipment

to monitor and treat this condition. Special care should be taken with patients with
respiratory insuciency. Signs of respiratory depression may be abolished with a temporary suspension of the drug administration or with a decrease of 50% of the infusion
rate. Remifentanil did not display recurrent respiratory depression, however many factors may aect the post-operative recovery, and it is important to make sure that the
patient has fully recovered and is in spontaneous breathing before taking the patient to
the recovery room.
The

cardiovascular risks associated with the administration of remifentanil like

hypotension and bradycardia that rarely may evolve to assystole and cardiac arrest,
may be reduced by decreasing the remifentanil infusion velocity, decrease the dose of
the concomitant anesthetics or by administrating intravenous uids, vasopressure agents
and anticholergenics, according to the situation.

Hypovolemic, hypotensive, weak or

older patients may be more sensitive to the cardiovascular side eects of remifentanil.
As in all opioids, remifentanil may cause

dependency.

Residual drug in the death space of the intravenous catheter may be sucient to
cause respiratory depression or apnea, and/or muscle rigidity, if the same catheter is
used to administer other drugs or solutions.

This may be avoided by the use of a

rapid ux catheter or an exclusive catheter for the administration of remifentanil that
is immediately removed after the discontinuation of the drug administration.
Like other opioids, remifentanil decreases the need of inhaled or intravenous anesthetics, and benzodiazepines to maintain general anaesthesia. If the central nervous system
depressants doses are not reduced there is an increased risk of incidence of adverse side
eects associated with these drugs. The adverse eect of remifentanil may also be exacerbated by the concomitant use of cardiac depressor drugs such as beta blockers and

233

A. REMIFENTANIL RECOMMENDATIONS IN PORTUGAL

calcium channels blockers.
During

pregnancy and breast feeding the use of remifentanil is only recommen-

ded when the potential benet is greater than the potential malecent. It is unknown if
remifentanil is excreted in the mother milk, however substances related to remifentanil
were found in female rat milk experiments, and the mother should be advised t not to
breast feed in the next 24 hours after the remifentanil administration. During labour,
remifentanil crosses the placenta barrier, and may cause respiratory depression in the
child.
The

more common adverse eects of remifentanil derive directly from the phar-

macology of the -opioids agonists, these adverse eects disappear minutes after the
interruption or decrease in the administration rate of the drug. The following list presents the adverse eects incidence, classied as very frequent (>1/10), frequent (>1/100
and

<10/10), little frequent (>1/1000 and <1/100), rare (>1/10000 and <1/1000) and

very rare (<1/10000).

Immune system diseases Rare:

allergic reactions, including anaphylaxis in asso-

ciation to one or more anesthetic agents.

Nervous system diseases Very frequent:

musculo-skeletical rigidity. Rare: seda-

tion (during the recovery period of general anaesthesia).

Cardiopathy Frequent:

bradycardia. Rare: assystoly/cardiac arrest, usually pre-

ceded of bradycardia, and generally in conjunct with other anesthetic agents.

Vasculopathy Very frequent:

hypotension. Frequent: post-operative hypertension.

Respiratory, thoracic and of mediation Frequent:

acute respiratory depression,

apnea. Little frequent: hypoxia.

Gastrointestinal diseases

Very frequent:

nausea and vomits.

Little frequent:

obstipation.

Cutaneous and subcutaneous tissue conditions Frequent:

itching.

General perturbations and alterations in the local of administration Fre-

234

A.2 Pharmacologic Properties of Remifentanil

quent: post-operative tremor. Little frequent: post-operative pain.
In the case of

overdosage, like every potent opioids, the predicted pharmacologic

actions of remifentanil will be extended.

Due to the very short action duration of

remifentanil, the side eects are limited to the immediate period after administration.
The drug interruption leads to the return to the initial state in 10 minutes. In the case
overdosage occurs, the drug administration should be interrupted, the airway should
be accessible for possible need of assisted or controlled ventilation, and to keep the
adequate cardiovascular function. In the case respiratory depression is associated with
muscle rigidity, the administration of neuromuscular blockers may be needed to facilitate
patients ventilation. Other uid and vasopressors may be administered intravenously
to treat hypotension.
The intravenous administration of an opioid antagonist like naloxone may be given
as a specic antidote to control serious respiratory depression and muscle rigidity. The
respiratory depression duration after the overdosage is unlikely to exceed the opioid
antagonist action.

A.2 Pharmacologic Properties of Remifentanil

Remifentanil is a selective

-opioid

-opioid

agonist with rapid eect and short duration. The

action is antagonized by narcotic agonists like naloxone. After the adminis-

tration of the recommended doses of remifentanil its eective half-life time is of 3 to 10

minutes. The average clearance in adult healthy patients is of about 40 ml/min/kg, the
distribution central volume of 100 ml/kg and the volume of distribution in the dynamic
equilibrium is of 350 ml/kg. Plasmatic concentrations of remifentanil are proportional to
the administrated doses along the recommended doses. For each 0.1 g/kg/min increase
in the infusion rate, the plasmatic concentration is increased in 2.5 ng/ml. Remifentanil
connects to the plasmatic proteins in about 70%.

235

A. REMIFENTANIL RECOMMENDATIONS IN PORTUGAL

Remifentanil is metabolized by sterases, and it is susceptible of metabolization by
plasmatic and non-specic tecidular sterases. Remifentanil metabolism results in the
formation of a metabolite

acid carboxylic, which in the dog has 1/4600 of remifentanil

potency. In man, the research indicates that all pharmacological activity is associated
with the original compound. The activity of this metabolite has no clinical consequence.
The metabolite half-life in adult healthy patients is of about 2 hours. In patients with
normal renal function the time for 95% of renal elimination of the primary metabolite
of remifentanil is of about 7 to 10 hours. Remifentanil is not a substrate for plasmatic
cholinesterases.
The rapid reversion of the post-analgesia action of remifentanil is not aected by
the state of the

renal function.

Clearance of the metabolite acid carboxylic is reduced

in patients with renal insuciency, and the metabolite concentration is expected to

reach 100 times the remifentanil concentration in the equilibrium state in patients with
moderate and severe renal insuciency in the intensive care. Clinical results show that
the accumulation of the metabolite does not result in -opioid relevant clinical eects,
even after the administration of remifentanil for three days. There is no data regarding
the metabolite pharmacokinetics in treatments of longer duration.
At least 30% of the metabolite acid carboxylic is removed during hemodyalise.
Pharmacokinetics of remifentanil is not altered in patients with acute

hepatic in-

suciency waiting for hepatic transplant, or during the anhepatic phase of the surgery
of leaver transplant. Patients with hepatic insuciency may be more sensitive to the
eect of respiratory depression of remifentanil.
In

pediatric patients

the clearance and distribution volume in the equilibrium

phase are increased in younger children, reaching the values for adult healthy patients
at the age of 17. The pharmacokinetics of the metabolite acid carboxylic in pediatric
patients age 2 to 17 years, is similar to the observed in adults, after correction of the
body weight dierences.

236

A.2 Pharmacologic Properties of Remifentanil

In

elderly patients the remifentanil clearance is slightly reduced (>65 years), when

compared to young patients. The pharmacodynamic activity of remifentanil increases

with age.

Elderly patients have an EC50 of remifentanil to form delta waves in the

electroencephalogram that is 50% of the tonger patients dose, therefore the initial dose
of remifentanil should be reduced in 50% in elderly patients and then carefully adjusted
according to the patient needs.

237

A. REMIFENTANIL RECOMMENDATIONS IN PORTUGAL

238

Appendix B
Informed Consent Form

239

Estudos de investigao

Consentimento Informado

CONSENTIMENTO INFORMADO

Anlise e Superviso da Nocicepo Durante a Anestesia

Eu, abaixo-assinado _____________________________________________, fui informado de que
o estudo de investigao acima mencionado se destina a analisar a anestesia geral a que serei
submetido, de modo a obter informao para o desenvolvimento de um ndice clnico que permita ao
anestesista avaliar melhor se a anestesia/analgesia so eficazes. Sei que neste estudo me ser
administrado no incio uma de trs doses possveis de medicamento analgsico remifentanil. Sei
tambm que no decorrer do estudo a dose deste medicamento ira sendo alterada de modo a
permitir a anlise de alguns sinais clnicos, ficando assegurado que no sentirei qualquer dor.
Foi-me garantido que todos os dados relativos identificao dos participantes neste estudo so
confidenciais e que ser mantido o anonimato.
Sei que posso recusar-me a participar ou interromper a qualquer momento a participao no estudo,
sem nenhum tipo de penalizao por este facto.
Compreendi a informao que me foi dada, tive oportunidade fazer perguntas e as minhas dvidas
foram esclarecidas.
Aceito participar de livre vontade no estudo acima mencionado.
Tambm autorizo a divulgao dos resultados obtidos no meio cientfico, garantindo o anonimato.
Os dados, sem qualquer identificao do participante, sero partilhados com a Aspect Medical
Systems, Inc., garantindo a confidencialidade dos dados.
Nome do participante no estudo: ________________________________________
Assinatura ---------------------------------

Data [Ano/ms/dia].

Nome do mdico responsvel: ________________________________________

Assinatura ---------------------------------

Data [Ano/ms/dia].

Anexo: Documentao de informao sobre o projecto.

DEFI Modelo de Consentimento Informado para Estudos de Investigao Edio 1 / Verso 1 Aprovado em CA em 15-03-2007

Estudos de investigao

Consentimento Informado

Documentao de informao sobre o projecto, anexo ao consentimento informado.

Na anestesia geral administra-se um medicamento que faz o doente dormir, isto
permanecer inconsciente, e um medicamento que anula as respostas do organismo a todos os
estmulos que pudessem causar dor, principalmente aos procedimentos relacionados com a cirurgia.
No doente anestesiado, como est inconsciente, falamos, no em dor, mas sim em nocicepo.
Hoje em dia, ao realizar uma anestesia geral, dispomos de um monitor que permite avaliar
se a anestesia adequada no que diz respeito sua componente hipntica, isto , saber se o doente
est mais ou menos adormecido. Todavia, no existe actualmente, nenhum monitor que permita ao
anestesista saber se o doente tem a quantidade do medicamento analgsico adequada, isto ,
avaliar o grau de nocicepo. A nossa investigao pretende obter informao de modo a poder
desenvolver um novo ndice clnico para medir a nocicepo/analgesia no doente anestesiado. Tal
importante para o avano da anestesiologia e poder permitir o desenvolvimento de sistemas
inteligentes para a administrao da anestesia.
Numa pessoa acordada, este tipo de estmulos provocam aquilo a que chamamos dor. No
doente anestesiado, como est inconsciente, no pode sentir dor; todavia h tendncia a que
ocorram manifestaes em resposta a estes estmulos s quais chamamos nocicepo. Tais
manifestaes so por exemplo um aumento da tenso arterial e dos batimentos do corao,
produo de suor ou lgrimas, ou um aumento da actividade elctrica das clulas do crebro ou dos
msculos da face.
Neste estudo vamos administrar aos doentes a anestesia que fazemos no dia-a-dia,
introduzindo um protocolo segundo o qual a dose do medicamento analgsico vai aumentando ou
diminuindo, ao mesmo tempo que analisamos os sinais clnicos de nocicepo, isto , de resposta
aos estmulos cirrgicos.
O medicamento analgsico que vamos usar chama-se remifentanil e caracteriza-se por ter
uma aco muito eficaz e rpida, permitindo, em menos de dois minutos, atingir um novo nvel mais
alto, qualquer que seja esse nvel.
Os sinais que vamos analisar, so sinais que recolhemos sempre em todos os doentes
anestesiados, mesmo que no se realizasse o estudo: tenso arterial, frequncia cardaca, tamanho
da onda do pulso arterial e ainda o BIS e EMG, que so sinais elctricos resultantes da actividade
das clulas do crebro e do msculo da testa, captadas atravs de um sensor colado na testa e que
permitem avaliar se o nvel de anestesia (mais ou menos adormecido) adequado. Ao aumentar
ou diminuir a dose do medicamento analgsico, logo que se observe uma alterao significativa
(20%) no valor desses sinais clnicos, diminui-se ou aumenta-se a dose do medicamento de modo a
obter a normalizao dos sinais clnicos. Na fase em que a dose do medicamento diminuda, no
se ultrapassa um limite considerado de segurana Com a medio do referido BIS fica assegurado
que o doente se mantm inconsciente.
DEFI Modelo de Consentimento Informado para Estudos de Investigao Edio 1 / Verso 1 Aprovado em CA em 15-03-2007

Estudos de investigao

Consentimento Informado
O estudo ter uma primeira fase antes do incio da cirurgia, mas j depois de o doente estar
anestesiado. Nessa primeira fase usaremos um procedimento habitual que usamos com frequncia
no dia-a-dia e que consiste na aplicao de um estmulo elctrico, cutneo ao nvel do pulso
(estmulo tetnico), analisando a resposta dos sinais clnicos atrs referidos. Se houver alterao de
algum dos sinais, a dose do medicamento analgsico ser aumentada; no havendo alterao, d-se
incio cirurgia.
Neste estudo, as nicas diferenas na anestesia em relao ao que seria a anestesia, caso
no fosse aplicado o protocolo do estudo, so as j descritas variaes na dose do medicamento
analgsico e a escolha da dose inicial de analgsico. Para esta dose inicial os doentes sero
distribuidos ao acaso para receber uma de trs doses pr-definidas. Qualquer uma dessas doses
est dentro das doses clnicas recomendadas, sendo que os doentes que recebem a dose mais
baixa, naturalmente recebero uma dose mais elevada do medicamento hipntico.
Aps a cirurgia, o doente ser monitorizado de acordo com os procedimentos habituais,
sendo registados os consumos de medicamento analgsico, bem como a avaliao do doente numa
escala de dor, durante o perodo em que se encontrar no recobro.

DEFI Modelo de Consentimento Informado para Estudos de Investigao Edio 1 / Verso 1 Aprovado em CA em 15-03-2007

Appendix C
Remifentanil Recommended Doses

Appendix A presents in Table A.1 the recommended doses of remifentanil in co-administration

with propofol. In the study, infusions were controlled by a TCI system, and it is therefore necessary to demonstrate equivalence of theoretical eect-site concentration doses
and the recommended doses of remifentanil in co-administration with propofol.
To evaluate comparability between recommended doses and the doses administered
in the study protocol, some simulations using RugloopI

c software were made. Eect

site concentration targets with correspondent bolus dose and infusion rates were evaluated, considering a male patient, 70 kg, 170 cm, 60 years old. Recommended remifentanil
doses in co-administration with propofol are:

bolus 0.5-1 g/ml

infusion rate 0.05-2 g/kg/min

35-70 g;

3.5-140 g/min.

Simulating the induction targets for each patient using TCI, the following bolus are
administered, in consonance with the recommended doses:

2.0 ng/ml

bolus of 34 g;

3.0 ng/ml

bolus of 51 g;

243

C. REMIFENTANIL RECOMMENDED DOSES

4.0 ng/ml

bolus of 68 g.

Simulating the maintenance doses of remifentanil administered using TCI, one must
consider the minimum 1.5ng/ml eect-site concentration allowed in the study protocol,
for the second phase of the study. Considering the simulated patient this target is translated into a continuous administration of remifentanil at an infusion rate of 3.84 g/ml,
and therefore within the recommended range.

244

Appendix D
Informed Consent Form Volunteers' Study

245

Estudos de investigao

Consentimento Informado

CONSENTIMENTO INFORMADO

Anlise e Superviso da Nocicepo Durante a Anestesia

Potenciais Evocados Sensitivos
Eu, abaixo-assinado _____________________________________________, voluntariei-me e fui
informado de que o estudo de investigao acima mencionado se destina a avaliar a utilizao dos
potenciais evocados sensitivos para diferentes nveis de estmulo sensitivo e de analgsico/hipntico.
Sei que neste estudo me sero administradas doses de analgsico e hipntico, para verificar o
impacto destas drogas nos potenciais evocados, bem como uma bateria de estmulos sensitivos de
intensidade crescente at ultrapassar o limiar de sensibilidade dolorosa. Tomei conhecimento dos
riscos associados administrao destas drogas, nomeadamente o de depresso respiratria, para o
qual poderei necessitar de ventilao assistida e intubao traqueal.
Foi-me garantido que todos os dados relativos identificao dos participantes neste estudo so
confidenciais e que ser mantido o anonimato.
No receberei qualquer compensao pela participao no estudo, e sei que posso recusar-me a
participar ou interromper a qualquer momento a participao, sem nenhum tipo de penalizao por
este facto.
Fui informado de que aps o estudo, nesse dia, no deverei conduzir automvel, nem operar
mquinas ou mecanismos de preciso.
Compreendi a informao que me foi dada, tive oportunidade fazer perguntas e as minhas dvidas
foram esclarecidas.
Aceito participar de livre vontade no estudo acima mencionado.
Tambm autorizo a divulgao dos resultados obtidos no meio cientfico, garantindo o anonimato.
Nome do participante no estudo: ________________________________________
Assinatura ---------------------------------

Data [Ano/ms/dia].

Nome do mdico responsvel: ________________________________________

Assinatura ---------------------------------

Data [Ano/ms/dia].

Anexo: Documentao de informao sobre o projecto.

DEFI Modelo de Consentimento Informado para Estudos de Investigao Edio 1 / Verso 1 Aprovado em CA em 15-03-2007
Pgina 1 de 2

Estudos de investigao

Consentimento Informado

Documentao de informao sobre o projecto, anexo ao consentimento informado.

O presente estudo destina-se avaliao do uso de potenciais evocados sensitivos para
avaliar estmulo doloroso e o impacto que as drogas anestsicas intravenosas, propofol (hipntico) e
remifentanil (analgsico), tm nos potenciais registados.
Os potenciais evocados sensitivos so uma ferramenta habitualmente utilizada como
diagnstico, para avaliar patologias que afectam a conduo nervosa. Estudos sugerem que a
latncia e amplitude dos potenciais sensitivos esto relacionadas com a intensidade do estmulo
sensitivo, bem como com a dose de analgsico administrada para bloquear a resposta a esse
estmulo. Assim, neste estudo o objectivo pretende testar a hiptese de que diferentes estmulos
tm diferentes amplitudes de resposta, e avaliar qual o impacto do analgsico, do hipntico e da
combinao destas drogas, nos potenciais evocados sensitivos, de modo a estabelecer separao
entre efeitos, e se este mtodo pode de facto traduzir informao relativamente ao balano
analgesia/nocicepo.
As drogas anestsicas administradas no estudo podem provocar depresso respiratria, pelo
que pode ser eventualmente necessrio iniciar ventilao assistida e proceder a uma intubao
traqueal.
No dia em que se realiza o estudo os voluntrios devem apresentar-se com jejum superior a
seis horas, e aps a realizao do trabalho, no devero conduzir automvel, ou operar mquinas
ou mecanismos de preciso.
Os voluntrios que aceitem participar neste estudo no recebero compensao monetria.

DEFI Modelo de Consentimento Informado para Estudos de Investigao Edio 1 / Verso 1 Aprovado em CA em 15-03-2007
Pgina 2 de 2

D. INFORMED CONSENT FORM - VOLUNTEERS' STUDY

248

Appendix E
Simulations and Control

The PKPD models of Schnider[50] and Minto[38, 39], for propofol and remifentanil
respectively, were implemented in Simulink, Matlab

R

.

The referred PKPD models

have a three compartments structure (Figure 1.13) and the block diagram is shown in
Figure E.1.
Each pharmacokinetic and dynamic model was obtained from a set of individuals,
and adjusted to build the average patient model[38, 39, 50]. Pharmacokinetics and
dynamics vary between patients according to age, weight, height and gender.

These

variables were introduced in the modelling process, and model parameters depend on
them either directly (age, height and weight), or through LBM (gender, height, weight).
To assess inter-patient variability considering the patient data, PKPD models were
simulated for dierent combinations of age, weight, height and gender as follows:

age from 20 to 85 years in steps of 2 years;

weight from 40 to 110 kg in steps of 5 kg;

height from 150 to 200 cm in steps of 5 cm;

LBM limited between 35 and 50.

249

E. SIMULATIONS AND CONTROL

Figure E.1: Simulink block diagram for the three compartments pharmacokinetic and
dynamic model.

The

kij

constants are transfer rates between compartments,

is the drug mass in each compartment and

(i, j

ce

massi

the eect-site concentration estimated

= 1, ..., 3).

Figures E.2 and E.3 present the results of the simulations for dierent combinations
of age, weight, height and gender, with LBM limitation, for Schnider and Minto models

250

(a) Male
Figure E.2:

(b) Female

Step response simulations for the Schnider model, considering dierent

combinations of age, weight, height and gender.

respectively, and an unitary step input.

Besides the variation in modelled parameters according to the patient personal data
(age, weight, height and gender), the model parameters also vary even for a patient with
the same characteristics considering demographic data. This variability was reported
in the studies were PKPD models were adjusted, and presented in Chapter 1, Tables
1.7,1.8 and 1.9. The parameters and reported dispersion were introduced in order to
reproduce the observed inter-patient variability (Table E.1), and be able to produce the
minimum and maximum patients described in Chapter 6.
The Schnider PKPD model implemented was part of the designed controller in
Chapter 6, and the inter-patient variability taken into account in the test of the controller robustness. Figures E.4 and E.5 present the block diagrams for the controller
used in the simulations for the nominal, minimum and maximum patients, and a propo-

251

E. SIMULATIONS AND CONTROL

(a) Male

(b) Female

Figure E.3: Step response simulations for the Minto model, considering dierent combinations of age, weight, height and gender.

sed structure to control on-line the infusion rate based on the State Entropy monitored
values in the patient, producing an adequate infusion rate (ml/h) to maintain or lead
the patient to the reference target concerning the control variable, in this case the EEG
derived index translating the patient's hypnosis state.
Inter-patient variability observed is resultant both from dierences in the patient's
data, age, weight, height and gender, and due to variations from the population modelled
trend, seen in the variability reported in the modelled parameters. Much of the variability observed in settling times of the simulations is consequence of the non-symmetry
between ow rate constants. We also observe that compartments 2 and 3 may be seen as
perturbations of the central compartment (parallel dynamics), and translation between
Cp and Ce presents itself as a series dynamic.

252

253

controller, used in the simulations for the nominal, minimum and maximum patients.

R

, with the system representation and feedback for the sliding-mode

Figure E.4: Block diagram implemented in Simulink Matlab

254

Representation of the diagram block for a possible on-line implementation, with the input incoming from the

actuator.

monitored values of State Entropy (SE) in the patient, and with the output infusion rate (ml/h) directed to the syringe pump

Figure E.5:

E. SIMULATIONS AND CONTROL

Table E.1: Considered variation intervals in the simulations, regarding the reported parameters and dispersion measures of the pharmacokinetic and dynamic model, reported
in [50].

Minimum

Maximum

V1

3,714

4,826

V2

If age>53: 14.240-0.531*(age-53) Else:

If age>53: 23.560-0.251*(age-53) Else:

14.240-0.251*(age-53)

23.560-0.531*(age-53)

168,2

307,8

[1.772+(weight-77)*8 +(LBM-

[2.008+(weight-77)*8 +(LBM-

59)*9 +(height-177)*10 ]/4.826

59)*9 +(height-177)*10 ]/3.714

Where: 8 =0.0636 if weight>77 Else

Where: 8 =0.0276 if weight>77 Else

8 =0.0276; 9 =-0.1021 if LBM>59

8 =0.0636; 9 =-0.0341 if LBM>59

Else 9 =-0.0341; 10 =0.0084 if

Else 9 =-0.1021; 10 =0.0444 if

height>177 Else 10 =0.0444

height>177 Else 10 =0.0084

If age>53:

If age>53:

[1.066-0.034*(age-53)]/4.826 Else:

[1.514-0.014*(age-53)]/3.714 Else:

[1.066-0.014*(age-53)]/4.826

[1.514-0.034*(age-53)]/3.714

k13

0.748/4.826

0.924/3.714

k21

If age>53:

If age>53:

[1.066-0.034*(age-53)]/V2max Else:

[1.514-0.014*(age-53)]/V2min Else:

[1.066-0.014*(age-53)]/V2max

[1.514-0.034*(age-53)]/V2min

0.748/307.8

0.924/168.2

V3
k10

k12

k31

255

E. SIMULATIONS AND CONTROL

256

References

[1] R. D. Miller, editor.

Miller's Anesthesia.

Elsevier, 6th edition, 2005.

(Cited on

page 1.)

[2] S. P. Desai, M. S. Desai, R. Maddi, and G. E. Battit. A tale of two paintings:

Depictions of the rst public demonstration of ether anesthesia.
106(5):10461050, 2007.

[3] F. D. Moore.

(Cited on page 2.)

John collins warren and his act of conscience - a brief narrative

of the trial and triumph of a great surgeon.

1999.

Anesthesiology,

Annals of Surgery,

229(2):187196,

(Cited on page 2.)

[4] J. Wawersik. [history of chloroform anesthesia].

tion, 22(6):144152, 1997.

Anaesthesiologie und Reanima-

(Cited on page 2.)

[5] J. Enderle, S. Blanchard, and J. Bronzino, editors.

Engnieering.

Introduction to Biomedical

Elsevier Academic Press, 2nd edition, 2005.

(Cited on pages 3

and 39.)

[6] L. C. Jameson and T. Sloan. Using eeg to monitor anesthesia drug eects during
surgery.

Journal of Clinical Monitoring and Computing, 20:445472, 2006.

(Cited

on pages 4, 5 and 6.)

[7] I. Rampil. A primer for eeg signal processing in anesthesia.

Scandinavica, 89(4):9801002, 1998.

Acta Anaesthesiologica

(Cited on pages 4, 5, 7 and 198.)

257

REFERENCES
[8] J. Bruhn, P. S. Myles, R. Sneyd, and M. M. R. F. Struys. Depth of anaesthesia
monitoring: what's available, what's validated and what's next?

of Anaesthesia, 97(1):8594, 2006.

British Journal

(Cited on pages 5, 72 and 151.)

[9] A. E. Guedel. Inhalation anesthesia: A fundamental guide.

American Medical Association, 109(2), 1937.

The Journal of the

(Cited on page 5.)

[10] D. A. Chernik, D. Gillings, H. Laine, J. Hendler, J. M. Silver, A. B. Davidson,

E. M. Schwam, and J. L. Siegel. Validity and reliability of the observer's: Assessment of alertness/sedation scale: Study with intravenous midazolam.

Clinical Psychopharmacology, 10(4), 1990.

Journal of

(Cited on page 6.)

[11] T. Gan, P. S. Glass, A. Windsor, F. Payne, C. Rosow, P. Sebel, and P. Manberg.

Bispectral index monitoring allows faster emergence and improved recovery from
propofol, alfentanil, and nitrous oxide anesthesia.
1997.

Anesthesiology, 87(4):808815,

(Cited on page 7.)

[12] A. Ekman, M. L. Lindholm, C. Lennmarken, and R. Sandin. Reduction in the

incidence of awareness using bis monitoring.
48(1):2026, 2004.

Acta Anaesthesiologica Scandinavica,

(Cited on pages 7, 22 and 113.)

[13] P. S. Myles, K. Leslie, J. McNeil, A. Forbes, and M. T. Chan. Bispectral index

monitoring to prevent awareness during anaesthesia:
controlled trial.

Lancet,

363(9423):17571763, 2004.

the b-aware randomised

(Cited on pages 7, 22, 42,

113 and 203.)

[14] S. S. Kelley, editor.

during anesthesia.

Monitoring Consciousness - Using the Bispectral IndexTM

Covidien, 2nd edition, 2010.

(Cited on page 7.)

[15] USA Aspect Medical Systems. Monitoring for pain - new insight from the eeg,
2007.

(Cited on pages 7, 35 and 36.)

258

REFERENCES
[16] S. D. Greenwald and C. Rosow. Bis and emg variability increase before somatic
responses during surgery.

Anesthesiology, 105:1027, 2006.

(Cited on pages 7, 35,

36, 90 and 111.)

[17] C. V. Seshagiri, C. E. Rosow, and S. Greenwald. Bis(r) variability in response to

skin incision was greater opposite to the side of surgery.
2007.

Anesthesiology, page 807,

(Cited on pages 7, 36, 98, 111 and 112.)

[18] R. K. Ellerkmann, V. M. Liermann, T. M. Alves, I. Wenningmann, S. Kreuer,

W. Wilhelm, H. Roepcke, A. Hoeft, and J. Bruhn.

Spectral entropy and bis-

pectral index and measures of the electroencephalographic eects of sevourane.

Anesthesiology, 101(6):12751282, 2004.

(Cited on pages 8 and 203.)

[19] H. Vierti-Oja, V. Maja, M. Srkel, P. Talja, N. Tenkanen, H. Tolvanen-Laakso,

M. Paloheimo, A. Vakkuri, A. Yli-Hankala, and P. Merilinen.

Description of

the entropy algorithm as applied in the datex-ohmeda s/5 entropy module.

Anaesthesiologica Scandinavica, 48(2):154161, 2004.

Acta

(Cited on page 8.)

[20] E. R. J. Seitsonen, I. K. J. Korhonen, M. J. van Gils, M. Huiku, J. M. P. Ltjnen, K. T. Kottila, and A. M. Yli-Hankala.

Eeg spectral entropy, heart rate,

photoplethysmography and motor responses to skin incision during sevourane

anaesthesia.

Acta Anaesthesiologica Scandinavica,

49:284292, 2005.

(Cited on

pages 9, 33, 34, 111 and 112.)

[21] D. M. Mathews, P. M. Cirullo, M. M. R. F. Struys, T. De Smet, R. J. Malik,

C. L. Chang, and G. G. Neuman.

Feasibility study for the administration of

remifentanil based on the dierence between response entropy and state entropy.

British Journal of Anaesthesia,

98(6):785791, 2007.

(Cited on pages 9, 37, 42,

90, 194, 198 and 200.)

[22] H. Merskey and N. Bogduk, editors.

Part III: Pain Terms, A Current List with

259

REFERENCES
Denitions and Notes on Usage. Classication of Chronic Pain. IASP Task Force
on Taxonomy. IASP Press, 2nd edition, 1994.

[23] W.

F.

Gannong,

editor.

(Cited on page 9.)

Review of Medical Physiology.

Books/McGraw-Hill, 22nd edition, 2005.

Lange

Medical

(Cited on pages 9, 10, 11, 13, 14,

15, 16, 32 and 188.)

Medical Physiology Principles for Clinical

[24] R. A. Rhoades and D. R. Bell, editors.

Medicine. Lippincott Williams and Wilkins, 3rd edition, 2009.

(Cited on pages 9,

12, 14, 15, 16 and 32.)

[25] A. Guyton and J. E. Hall, editors.

Saunders, 11th edition, 2006.

[26] L. Kruger, editor.

Textbook of Medical Physiology.

Elsevier

(Cited on page 15.)

Methods in Pain Research.

CRC Press, 2001.

(Cited on

pages 16, 17 and 18.)

[27] G. A. Lambert, G. Mallos, and A. S. Zagami.

Von frey's hairs - a review of

their technology and use - a novel automated von frey device improved testing for
hyperalgesia.

Journal of Neuroscience Methods, 177(2):420426, 2009.

(Cited on

page 16.)

[28] M. E. Wewers and N. K. Lowe. A critical review of visual analogue scales in the
measurement of clinical phenomena.

Research in nursing and Health, 13(4), 1990.

(Cited on page 17.)

[29] R. Melzack and W. S. Torgerson. On the language of pain.

5059, 1971.

Anesthesiology, 34(1):

(Cited on page 18.)

[30] L. Schwartz, M. P. Jensen, and J. M Romano. The development and psychometric

evaluation of an instrument to assess spouse responses to pain and well behavior

260

REFERENCES
in patients with chronic pain: The spouse response inventory.

Pain, 6(4):243252, 2005.

The Journal of

(Cited on page 19.)

[31] L. Tesio, C. V. Granger, and R. C. Fiedler.

measure for low-back pain syndromes.

Pain,

A unidimensional pain/disability
69(3):269278, 1997.

(Cited on

pages 19, 73 and 77.)

[32] D. A. Revicki, W. Chen, N. Harnam, K. F. Cook, D. Amtmannb, L. F. Callahan,

M. P. Jensen, and F. J. Keefe.
promis pain behavior item bank.

Development and psychometric analysis of the

Pain, 146(1-2):158169, 2009.

(Cited on pages 19

and 77.)

[33] F. Varoli and V. Pedrazzi. Adapted version of the mcgill pain questionnaire to
brazilian portuguese.

Brazilian Dental Journal,

17(4):328335, 2006.

(Cited on

pages 19 and 20.)

[34] L. Brunton, K. Parker, D. Blumenthal, and I. Buxton, editors.

Gilman's Manual of Pharmacology and Therapeutics.

2008.

Goodman and

McGraw Hill, 12th edition,

(Cited on pages 20, 21, 22, 23, 24, 188, 198 and 229.)

[35] INFARMED.

Folheto informativo. remifentanil ultiva, 2006.

FARMED at 30-08-2006.

(Cited on pages 21, 22, 229, 230, 231 and 232.)

[36] Drug Bank. Drug bank: Remifentanil, August 2011. URL

ca/drugs/DB00899/structure.
[37] INFARMED.

Approved by IN-

http://www.drugbank.

(Cited on page 21.)

Resumo das caractersticas do medicamento. remifentanil ultiva,

2006. Approved by INFARMED at 30-08-2006.

(Cited on pages 22, 229, 230,

231 and 232.)

[38] C. F. Minto, T. W. Schnider, and S. L. Shafer. Pharmacokinetics and pharma-

261

REFERENCES
codynamics of remifentanil. i. model development.

Anesthesiology, 86:1023, 1997.

(Cited on pages 22, 25, 29, 30, 43, 51, 215 and 249.)

[39] C. F. Minto, T. W. Schnider, and S. L. Shafer. Pharmacokinetics and pharmacodynamics of remifentanil. ii. model application.

Anesthesiology, 86:2433, 1997.

(Cited on pages 22, 25, 29, 41, 43, 49, 50, 51, 165, 167, 203, 204, 215 and 249.)

[40] T. G. Monk, V. Saini, B. C. Weldon, and J. C. Sigl. Anesthetic management and

one-year mortality after noncardiac surgery.
410, 2005.

Anesthesia and Analgesia,

100(1):

(Cited on pages 22, 42 and 113.)

[41] S. Karri, A. Maheshwari, C. Koch, M. Xu M, and A. Duncan.

Deeper level of

anesthesia is associated with adverse neurological outcomes in cardiac surgical

patients.

Anesthesia and Analgesia,

108(SCA Supp):1104, 2009.

(Cited on

pages 22 and 113.)

[42] E. Farag, G. J. Chelune, A. Schubert A, and E. J. Mascha.

Deeper level of

anesthesia is associated with adverse neurological outcomes in cardiac surgical

patients.

Anesthesia and Analgesia,

103(3):633640, 2006.

(Cited on pages 22

and 113.)

[43] Drug Bank.

Drug bank: Propofol, August 2011. URL

ca/drugs/DB00818/structure.

http://www.drugbank.

(Cited on page 23.)

Modern Anesthetics, volume 182 of Hand-

[44] J. Schttler and H. Schwilden, editors.

book of Experimental Pharmacology.

Springer, 22nd edition, 2008.

(Cited on

pages 23 and 24.)

[45] M. Kreuzer, G. Schneider, B. Drexler, H. Hentschke, B. Antkowiak, and E. F.

Kochs.

Sevourane and propofol have dierent eects on down states of local

eld potential activity in vitro.

British Journal of Anaesthesia - Proceedings of

262

REFERENCES
the 7th International Symposium Memory and Awareness in Anaesthesia, 100(6):
868P869P, 2008.

(Cited on pages 24 and 188.)

[46] A. Borgeat, O. H. Wilder-Smith, and P. M. Suter. The nonhypnotic therapeutic

applications of propofol.

Anesthesiology, 80(3):642656, 1994.

(Cited on pages 24,

188 and 198.)

[47] B. K. Kataria, S. A. Ved, H. F. Nicodemus, G. R. Hoy, D. Lea, M. Y. Dubois,

J. W. Mandema, and S. L. Shafer. The pharmacokinetics of propofol in children
using three dierent data analysis approaches.
1994.

Anesthesiology,

80(1):104122,

(Cited on pages 24 and 30.)

[48] A. Absalom, D. Amutike, A. Lal, M. White, and G. N. Kenny.

The pharma-

cokinetics of propofol in children using three dierent data analysis approaches.

British Journal of Anaesthesia, 72(4):302306, 2003.

(Cited on pages 24 and 30.)

[49] B. Marsh, M. White, N. Morton, and G. N. C. Kenny. Pharmacokinetic model

driven infusion of propofol in children.
1991.

British Journal of Anaesthesia, 67:4148,

(Cited on pages 24, 25, 30 and 204.)

[50] T. W. Schnider, C. F. Minto, P. L. Gambus, C. Andresen, D. B. Goodale, S. L.

Shafer, and E. J. Youngs. The inuence of method of administration and covariates
on the pharmacokinetics of propofol in adult volunteers.
82, 1998.

Anesthesiology, 88:1170

(Cited on pages 24, 25, 29, 30, 31, 41, 51, 165, 167, 203, 204, 249

and 255.)

[51] A. Absalom and M. M. R. F. Struys, editors.

An Overview of TCI and TIVA.

Academia Press, Belgium, 1st edition, 1991. (Cited on pages 24, 25, 203 and 204.)

[52] J. G. Webster, editor.

Encyclopedia of Medical Devices and Instrumentation,

volume 4 and 5. Wiley-Interscience, 2nd edition, 2006.

263

(Cited on page 25.)

REFERENCES
[53] C. F. Minto, T. W. Schnider, T. G. Short, K. M. Gregg, A. Gentilini, and S. L.
Shafer. Response surface model for anesthetic drug interactions.
92(6):9801002, 2000.

Anesthesiology,

(Cited on page 27.)

[54] C. S. Nunes, H. Alonso, A. Castro, P. Amorim, and T. Mendona. Towards the

control of depth of anaesthesia:

Identication of patient variability.

Control Conference, ECC 2007, pages 310931153, 2007.

European

(Cited on pages 27, 28

and 41.)

[55] A. Castro, C. S. Nunes, P. Amorim, and F. G. Almeida. Hypnotic administration

for anesthesia using sliding-mode control. In

Engineering in Medicine and Biology

Society, 2008. EMBS 2008. 30th Annual International Conference of the IEEE,
pages 5810 5813, aug. 2008.

doi:

10.1109/IEMBS.2008.4650535.

(Cited on

pages 27, 42, 193, 194 and 202.)

[56] A. Castro, E. Amadeu, F. Martins, P. S Couto, P. Amorim, and C. S. Nunes.

Propofol and remifentanil drug interaction on the state entropy of the eeg: Surface
model.

Total Intravenous Anesthesia - TCI World Congress, page 98, 2007.

Presentation.

Oral

(Cited on pages 27 and 28.)

[57] A. Castro, P. Amorim, and C.S. Nunes. Modeling state entropy of the eeg and
auditory evoked potentials - hypnotic and analgesic interactions. In

Engineering

in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International
Conference of the IEEE, pages 1949 1952, aug. 2007.
4352699.

doi: 10.1109/IEMBS.2007.

(Cited on pages 27 and 28.)

[58] P. M. Schumacher, T. W. Bouillon, D. Leibundgut, V. Hartwich, and M. Luginbhl. Time-based online display of a noxious stimulus response index on pharamacological data.

Anesthesiology,

page 17, 2007.

page 28.)

264

ASA Meeting.

(Cited on

REFERENCES
[59] T. W. Bouillon, P. M. Schumacher, D. Leibundgut, S. L. Shafer, and A. M. Zbinden. A novel mechanistic model based on the mac reduction paradigm describes
hypnotic-opioid interaction for suppression of response to stimulation.

ology, page 503, 2007.

ASA Meeting.

Anesthesi-

(Cited on page 28.)

[60] M. Luginbhl, P. M. Schumacher, P. Vuilleumier, H. Vereecke, B. Heyse, T. W.

Bouillon, and M. R. F. Struys.

Noxious stimulation response index:

anesthetic state index based on hypnotic-opioid interaction.

(4):872880, 2010.

A novel

Anesthesiology,

112

(Cited on page 28.)

[61] A. Castro, E. Amadeu, P. S Couto, P. Amorim, and C. S. Nunes. Bispectral index

and state entropy simulation with propofol and remifentanil eect-site steering.

Journal of Neurosurgical Anesthesiology, 20(4):340, 2008.

(Cited on page 28.)

[62] A. Castro, E. Amadeu, P. S Couto, P. Amorim, and C. S. Nunes. Educational

software for bispectral index and state entropy simulation.
A686, 2008.

Anesthesiology,

page

(Cited on page 28.)

[63] S. Bibian, G. A. Dumont, M. Huzmezan, and C. R. Ries.

tainty bounds in anesthetic pkpd models.

Quantifying uncer-

Annual International Conference of

the IEEE Engineering in Medicine and Biology Society (EMBS), pages 786789,
2004.

(Cited on pages 28 and 41.)

[64] T. Bouillon and S. L. Shafer. Does size matter?

1998.

Anesthesiology,

89(3):557559,

(Cited on page 30.)

[65] A. R. Absalom, V. Mani, T. De Smet, and M. M. R. F. Struys. Pharmacokinetic

models for propofoldening and illuminating the devil in the detail.

Journal of Anaesthesia, 103(1):2637, 2009.

(Cited on page 30.)

[66] W. Miltner, R. Johnson Jr., C. Braun, and W. Larbig.

265

British

Somatosensory event-

REFERENCES
related potentials to painful and non-painful stimuli: eects of attention.
38(3):303312, 1989.

(Cited on pages 32 and 186.)

[67] A. McPherson and S. Renfrew.

Assymetry of perception of size between the

right and left hands in normal subjects.

Psychology, 5(2):6674, 1953.

[68] D. R. Haslam.

Pain,

The Quarterly Journal of Experimental

(Cited on pages 32 and 156.)

Lateral dominance in the perception of size and of pain.

Quarterly Journal of Experimental Psychology,

22(3):503507, 1970.

The

(Cited on

pages 32 and 156.)

[69] R. G. Hallin and H. E. Torebjrk. Electrically induced a and c bre responses

in intact human skin nerves.

Experimental Brain Research, 16(3):309320, 1973.

(Cited on page 32.)

[70] H. Adraensen, J. Gybels, H. O. Handwerker, and J. Van Hees. Response properties

of thin myelinated (a ) bers in human skin nerves.
49(1):111122, 1983.

Journal of Neurophysiology,

(Cited on page 32.)

[71] S. V. Revenko, L. V. Baidakova, V. V. Ermishkin, N. A. Mangushevs, and L. Ya.

Selektor. Study of cutaneous nociception by means of chemical agents.

ologiya, 24(5):517529, 1992.

Neirozi-

(Cited on pages 32 and 155.)

[72] W. Ellermeier and W. Westphal.

Gender dierences in pain ratings and pupil

reactions to painful pressure stimuli.

Pain,

61(3):435439, 1995.

(Cited on

pages 32, 39 and 90.)

[73] J. S. Walker and J. J. Carmody. Experimental pain in healthy human subjects:

gender dierences in nociception and in response to ibuprofen.

Analgesia, 86(6):12571262, 1998.

Anesthesia and

(Cited on pages 32, 155 and 156.)

[74] A. Dahan, B. Kest, A. R. Waxman, and E. Sarton.

266

Sex-specic responses to

REFERENCES
opiates:
2008.

animal and human studies.

Anesthesia and Analgesia,

107(1):8395,

(Cited on pages 32, 155 and 156.)

[75] M. Pereira, B. Rehberg-Klug, A. Castro, and P. Amorim. Inuence of diuse noxious inhibitory control, temporal summation and expectation on pain perception
in healthy volunteers. In

Proceeding of the 2011 Annual Meeting of the American

Society of Anesthesiologists, page Accepted, 2011.

[76] H. A. Haga and N. I. Dolvik.

(Cited on page 32.)

Electroencephalographic and cardiovascular va-

riables as nociceptive indicators in isourane-anaesthetized horses.

Anaesthesia and Analgesia, 32(3):128135, 2005.

Veterinary

(Cited on page 33.)

[77] M. Rantanen, A. Yli-Hankala, M. Van Gils, H. Yppril-Wolters, P. Takala,

M. Huiku, M. Kymlinen, E. Seitsonen, and I. Korhonen.

Novel multipara-

meter approach for measurement of nociception at skin incision during general

anaesthesia.

British Journal of Anaesthesia,

96(3):367376, 2006.

(Cited on

pages 33, 34, 72, 90, 92, 111 and 112.)

[78] M. Huiku, K. Uutela, M. van Gils, I. Korhonen, M. Kymlinen, P. Merilinen, M. Paloheimo, M. Rantanen, P. Takala, H. Vierti-Oja, and A. Yli-Hankala.
Assessment of surgical stress during general anaesthesia.

esthesia, 98(4):447455, 2007.

British Journal of Ana-

(Cited on pages 33, 34, 35, 68, 72, 77, 79, 88, 92,

112, 130, 131, 133, 135, 140 and 145.)

[79] M. M. R. F. Struys, C. Vanpeteghem, M. Huiku, K. Uutela, N. B. K. Blyaert, and

E. P. Mortier. Changes in a surgical stress index in response to standardized pain
stimuli during propofol-remifentanil infusion.
(3):359367, 2007.

British Journal of Anaesthesia, 99

(Cited on pages 33, 72 and 90.)

[80] J. Ahonen, R. Jokela, K. Utela, and M. Huiku.

267

Surgical stress index reects

REFERENCES
surgical stress in gynaecological laparoscopic day-case surgery.

Anaesthesia, 98(4):456461, 2007.

British Journal of

(Cited on page 35.)

[81] M. Sorbello, C. Micali, G. Morello, A. Paratore, and S. Mangieameli. Accordance

of surgical stress index and clinical judgement in laparoscopic cholecistectomy.

Anesthesiology, page 818, 2007.

ASA Meeting.

(Cited on page 35.)

[82] H. Kallio, L. I. Lindberg, A. S. Majander, M. L. Niskanen, K. H. Uutela, and

M. P. J. Paloheimo. Surgical stress index (ssi) as a novel device for measuring
stress from intubation in children.

Anesthesiology,

page 80, 2007.

(Cited on

page 35.)

[83] M. M. R. F. Struys, C. Vanpeteghem, M. Huiku, K. Uutela, N. B. K. Blyaert, and

E. P. Mortier. Changes in surgical stress index in response to standardized pain
stimuli during propofol-remifentanil infusion.
(3):359367, 2007.

British Journal of Anaesthesia, 99

(Cited on page 35.)

[84] H. Kallio, L. I. Lindberg, A. S. Majander, K. H. Uutela, M. L. Niskanen, and

M. P. J. Paloheimo.

Measurement of surgical stress in anaesthetized children.

British Journal of Anaesthesiology, 101(3):383389, 2008.

(Cited on page 35.)

[85] M. Gruenewald, P. Meybohm, C. Ilies, J. Hcker, R. Hanss, J. Scholz, and B. Bein.

Inuence of dierent remifentanil concentrations on the performance of the surgical stress index to detect a standardized painful stimulus during sevourane
anaesthesia.

British Journal of Anaesthesia,

103(4):586593, 2009.

(Cited on

pages 35, 72 and 90.)

[86] M. P. J. Paloheimo, S. Sahanne, and K. H. Uutela. Autonomic nervous system

state: the eect of general anaesthesia and bilateral tonsillectomy after unilateral
inltration of lidocaine.

British Journal of Anaesthesia,

(Cited on pages 35, 68, 72, 90, 92 and 111.)

268

104(5):587895, 2010.

REFERENCES
[87] M. Bloom, A. Jurmann, G. Cu, and A. Bekker. Bis variability reects analgesia.

Journal of Neurosurgical Anesthesiology, 17(4):254255, 2005.

(Cited on pages 35,

72 and 92.)

[88] B. Rehberg, C. Ryll, D. Hadzidiakos, F. V. Dincklage, and J. H. Baars.

Va-

riability comparison of the composite auditory evoked potential index and the
bispectral index during propofol-fentanyl anesthesia.

Simulation, 107(1):117124, 2008.

Technology, Computing and

(Cited on page 35.)

[89] I. Takamatsu, M. Ozaki, and T. Kazama. Entropy indices vs the bispectral indextm for estimating nociception during sevourane anaesthesia.

of Anaesthesia, 96(5):620626, 2006.

British Journal

(Cited on pages 35 and 36.)

[90] A. J. Aho, A. Yli-Hankala, L.-P. Lyytikinen, and V. Jntti.

Facial muscle

anctivity, response entropy, and state entropy indices during noxious stimuli in
propofol-nitrous oxide or propofol-nitrous oxide-remifentanil anaesthesia without
neuromuscular block.

British Journal of Anaesthesia, 102(2):227233, 2009.

(Ci-

ted on page 36.)

[91] M. Kawaguchi, I. Takamatsu, K. Masui, and T. Kazama. Eect of landiolol on

bispectral index and spectral entropy responses to tracheal intubation during propofol anaesthesia.

British Journal of Anaesthesia, 101(2):273278, 2008.

(Cited

on page 37.)

[92] M. Kawaguchi, I. Takamutsu, and T. Kazama.

Rocuronium dose-dependently

supresses the spectral entropy response to tracheal intubation during propofol

anaesthesia.

British Journal of Anaesthesia,

102(5):667672, 2009.

(Cited on

page 37.)

[93] M. Valjus, A. R. Jokela, and K. Kortila. Response entropytm is not more sensitive
than state entropytm in distinguishing the use of esmolol instead of remifentanil

269

REFERENCES
in patients undergoing gynaechological laparoscopy.

dinavica, 50(1):3239, 2005.

Acta Anaesthesiologica Scan-

(Cited on page 37.)

[94] G. Weil, S. Passot, F. Servin, and V. Billard. Does spectral entropy reect the response to intubation or incision during propofol-remifentanil anesthesia?

sia and Analgesia, 106(1):152159, 2008.

Anesthe-

(Cited on page 37.)

[95] V. Bonhomme, V. Llabres, P. Y. Dewandre, J. F. Brichant, and P. Hans. Combined use of bispectral indextm and a-linetm autorregressive indextm to assess antinociceptive component of balanced anaesthesia during lumbar arthodesis.

Journal of Anaesthesia, 96(3):353360, 2006.

British

(Cited on pages 37 and 157.)

[96] E. Weber, P. Gambus, M. Jospin, X. Borrat, and S. Cortes. Prediction of nociceptive response by auditory evoked potentials during ultrasonographic exploration.

Anesthesiology, page 813, 2007.

ASA Meeting.

(Cited on page 37.)

[97] A. Cividjian, J.Y. Martinez, L. Bourdon, J. Escarment, and L. Quintin. A new

cardiovascular index predicts better than eeg unexpected movement during surgery.

Anesthesiology, 105:A1549, 2006.

(Cited on pages 37 and 38.)

[98] A. Cividjian, J.Y. Martinez, E. Combourieu P. Precloux, A. M. Beraud, Y. Rochette, M. Cler, L. Bourdon, J. Escarment, and L. Quintin. Beat-by-beat cardiovascular index to predict unexpected intraoperative movement in anaesthetized
unparalyzed patients:

a retrospective analysis.

and Computing, 21:91101, 2007.

Journal of Clinical Monitoring

(Cited on page 38.)

[99] P. F. Wey, C. Lions, J. Y. Martinez, A. Cividjian, and J. Escarment. Reduction of

unexpected movements during endoscopy using a cardiovascular index (cardean).

Anesthesiology, 109:A1286, 2008.

(Cited on page 38.)

[100] J. Y. Martinez, P. F. Wey, C. Lions, A. Cividjian, M. Rabilloud, A. Bissery,

L. Bourdon, M. Puidupin, J. Escarment, and L. Quintin. A beat-by-beat cardi-

270

REFERENCES
ovascular index, cardean: A prospective randomized assessment of its utility for
the reduction of movement during colonoscopy.
765772, 2010.

Anesthesia and Analgesia, 110(3):

(Cited on page 38.)

[101] A. Cividjian, M. Rosssi, M.C. Fevre, L. Quintin, and J.F. Payen.

Cardean, a

beat-by-beat on-line index assesses nociception: a randomized trial upon surgery.

In

Proceeding of the 2010 Annual Meeting of the American Society of Anesthesi-

ologists, page A1358, 2010.

[102] F. H. Netter, editor.
edition, 1997.

(Cited on page 38.)

Atlas of Human Anatomy.

ICON Learning Systems, 2nd

(Cited on page 38.)

[103] H. L. Bennet, L. Patel, N. Farida, S. Beddell, and M. Bobbin. Separation of the

hypnotic component of anesthesia and facial emg responses to surgical stimulation.

Anesthesiology, page 730, 2007.

[104] C. J. De Luca.

(Cited on page 38.)

Fundamental concepts in emg acquisition, 2001.

DelSys Inc.

(Cited on page 39.)

[105] C. J. De Luca. Surface electromiography detection and recording, 2002. DelSys

Inc.

(Cited on page 39.)

[106] H. Storm. Skin conductance and the stress response from heel stick in preterm
infants.

Archives Disease Childhood Fetal Neonatal, 83:143147, 2000.

(Cited on

page 39.)

[107] H. Storm. The development of a software program for analyzing skin conductance
changes in preterm infants.

Clinical Neurophysiology, 112:15621568, 2001.

(Cited

on page 39.)

[108] Med-Storm. Med-storm stress detector - user manual, 2008. Med-Storm Innovation AS Vers. 1.0.

(Cited on page 39.)

271

REFERENCES
[109] H. Storm. Changes in skin conductance as a tool to monitor nociceptive simulation
and pain.

Current Opinion in Anaesthesiology,

21(6):796804, 2008.

(Cited on

pages 39 and 90.)

[110] T. Ledowski, J. Bromilow, M. J. Paech, H. Storm, R. Hacking, and S. A. Schug.

Monitoring of skin conductance to assess postoperative pain intensity.
(6):862865, 2006.

Pain,

97

(Cited on page 39.)

[111] A. C. Gjerstad, H. Storm, R. Hagen, M. Huiku, E. Qvigstad, and J. Raeder. Comparison of skin conductance with entropy during intubation, tetanic stimulation
and emergence from general anaesthesia.
(1):815, 2007.

Acta Anaesthesiologica Scandinavica, 51

(Cited on page 39.)

[112] D. Afonso, A. Castro, E. Amadeu, and P. Amorim. Avaliao objectiva da dor:

conductncia elctrica cutnea.

Dor, 17:58, 2010.

(Cited on pages 39 and 54.)

[113] E. K. Choo, W. Magruder, C. J. Montgomery, J. Lim, R. Brant, and J. M.

Ansermino.

Skin conductance uctuations correlate poorly with postoperative

self-report pain measures in school-aged children.

2010.

Anesthesiology, 113(1):175182,

(Cited on page 39.)

[114] D. Afonso, A. Castro, E. Amadeu, and P. Amorim. Avaliao objectiva da dor:

Conductncia elctrica cutnea. In

8 Convnio da ASTOR on, 2010.

(Cited on

pages 39 and 54.)

[115] L. Barvais, E. Engelman, J. M. Eba, E. Coussaert, F. Cantraine, and G. N.

Kenny. Eect site concentrations of remifentanil and pupil response to noxious
stimulation.

British Journal of Anaesthesia,

91(3):347352, 2003.

(Cited on

[116] M. Larson, P. D. Berry, J. May, A. Bjorksten, and D. I. Sessler.

Latency of

pages 39 and 40.)

272

REFERENCES
pupillary reex dilation during general anesthesia.
97(2):725730, 2004.

Journal of Applied Physiology,

(Cited on pages 39 and 40.)

[117] P. Richeb, R. Pfei, S. Black, C. Joachin, and G. Janvier. Pupillary diameter

monitoring to identify development of peroperative acute remifentanil tolerance.

Anesthesiology, page 809, 2007.

ASA Meeting.

(Cited on page 39.)

[118] F.-X. Donnette, E. Bourgeois, I. Murat, and I. Constant. The mac of sevourane
allowing inhibition of pupillary response to nociception in children.
page 1633, 2007. ASA Meeting.

Anesthesiology,

(Cited on page 39.)

[119] P. S. Micalos, E. L. Drinkwater, J. Cannon, L. Arendt-Nielsen, and F. E. Marino.

Reliability of the nociceptive exor reex (riii) threshold and association with pain
threshold.

European Journal of Applied Physiology,

105(1):5562, 2009.

(Cited

on pages 40, 90 and 163.)

[120] F. von Dincklage, M. Hackbarth, R. Mager, B. Rehberg, and J. H. Baars. Monitoring of the responsiveness to noxious stimuli during anaesthesia with propofol and
remifentanil by using riii reex threshold and bispectral index.

Anaesthesia, 104(2):201208, 2010.

British Journal of

(Cited on pages 40 and 163.)

[121] T. D. Smet, M. M. R. F. Struys, S. Greenwald, E. P. Mortier, and S. L. Shafer.

Estimation of optimal modeling weights for a bayesian-based closed-loop system
for propofol administration using the bispectral index as a controlled variable: a
simulation study.

Anesthesia and Analgesia, 105(6):16291638, 2007.

(Cited on

pages 42, 194 and 198.)

[122] N. Liu, T. Chazot, B. Trillat, G. A. Dumont, and M. Fischler. Closed-loop titration of propofol guided by the bispectral index.

et de Ranimation, 26(10):850854, 2007.

273

Annales Franaises d'Anesthsie

(Cited on pages 42 and 194.)

REFERENCES
[123] K. Volyanskyy, W. M. Haddad, and J. M. Bailey.

Adaptive disturbance rejec-

tion control for compartmental systems with application to introperative anesthesia inuenced by hemorrhage and hemodilution eects.

Adaptive Control and Signal Processing,

International Journal of

23(1):129, 2009.

(Cited on pages 42,

43, 194 and 195.)

[124] T. J. Gilhuly, S. R. Hutchings, G. A. Dumont, and B. A. M. Struys. Development and pilot testing of the neuromuscular blockade advisory system.

Methods and Programs in Biomedicine, 89(2):179188, 2008.

Computer

(Cited on pages 42

and 194.)

[125] T. Mendona, H. Magalh aes, J. M. Lemos, P. Rocha, and S. Esteves. Multiple

model adaptive control of neuromuscular blockade: Design guidelines and clinical cases.

European Control Conference, ECC 2007,

2007.

(Cited on pages 42

and 194.)

[126] D. G. Mason, J. J. Ross, N. D. Edwards, D. A. Linkens, and C. S. Reilly. Selflearning fuzzy control of atracurium-induced neuromuscular block during surgery.

Medical and Biological Engineering and Computing, 35(5):498503, 1997.

(Cited

on pages 42 and 194.)

[127] N. Liu, T. Chazot, S. Hamada, A. Landais, N. Boichut, C. Dussaussoy, B. Trillat,

L. Beydon, E. Samain, D. I. Sessler, and M. Fischler. Closed-loop coadministration
of propofol and remifentanil guided by bispectral index: A randomized multicenter
study.

Anesthesia and Analgesia, 112(3):516518, 2011.

(Cited on pages 42, 194,

198 and 200.)

[128] J. E. Slotine and W. Li, editors.

Jersey, 1991.

Applied Nonlinear Control.

Prentice Hall, New

(Cited on pages 42, 203, 205 and 208.)

[129] P. J. Manberg, C. M. Vozella, and S. D. Kelley.

274

Regulatory challenges facing

REFERENCES
closed-loop control anesthetic drug infusion devices.

Therapeutics, 84(1):166169, 2008.

Clinical Pharmacology and

(Cited on pages 43, 79 and 194.)

[130] I. Korhonen and A. Yli-Hankala. Photoplethysmography and nociception.

Anaesthesiologica Scandinavica,

53(8):975985, 2009.

Acta

(Cited on pages 62, 68

and 72.)

[131] L. Srnmo and P. Laguna.

Bioelectrical signal processing in cardiac and neurolo-

gical applications. Elsevier Academic Press, 1st edition, 2005.

(Cited on pages 62,

63, 64 and 190.)

[132] M. Malik. Heart rate variability - standards of measurement, physiological interpretation, and clinical use - task force of the european society of cardiology the
north american society of pacing electrophysiology.
1996.

Circulation, 93(5):10431065,

(Cited on page 62.)

[133] J. M. Hyman.

Accurate monotonicity preserving cubic interpolation.

Journal on Scientic and Statistical Computing,

4(4):645654, 1983.

SIAM

(Cited on

page 66.)

[134] M. Sarfraz. Shape preserving rational cubic interpolation.

8(2-3):106111, 1993.

Extracta Mathematicae,

(Cited on page 66.)

[135] W. S. Cleveland. Robust locally weighted regression and smoothing scatterplots.

Journal of the American Statistical Association,

74(368):829836, 1979.

(Cited

on page 66.)

[136] A. Castro, P. Amorim, C. S. Nunes, and F. G. Almeida. Development of a scale to

assess noxious stimuli intensity during anesthesia and surgery using rasch analysis.

Journal of Neurosurgical Anesthesiology, 2011.

(Cited on page 71.)

[137] T. M. Hemmerling and N. Le. Brief review: Neuromuscular monitoring: ab update

275

REFERENCES
for the clinician.

Canadian Journal of Anaesthesia, 54(1):5872, 2007.

(Cited on

page 72.)

[138] H. Storm, K. Myre, M. Rostrup, O. Stokland, M. D. Lien, and J. C. Raeder.

Skin conductance correlates with perioperative stress.

Scandinavica, 46(7):887895, 2002.

Acta Anaesthesiologica

(Cited on pages 72 and 92.)

[139] K. Ho, J. Spence, and M. F. Murphy. Review of pain-measurement tools.

of Emergency Medicine, 27(4):427432, 1996.

[140] G. Rasch.

(Cited on page 72.)

Probabilistic Models for Some Intelligence and Attainment Tests.

nish Institute for Educational Research, Copenhagen, 1960.

Chicago: Mesa Press, 1992.

[141] J. M. Linacre.
1994.

Annals

Da-

Expanded edition,

(Cited on pages 73 and 76.)

Many-facet Rasch measurement. Mesa Press, Chicago, 2nd edition,

(Cited on pages 73 and 76.)

[142] C. Decruynaere, J. L. Thonnard, and L. Plaghki. Measure of experimental pain

using rasch analysis.

European Journal of Pain, 11(4):469474, 2007.

(Cited on

pages 73, 77 and 187.)

[143] K. Pesudovs and B. A. Noble. Improving subjective scaling of pain using rasch
analysis.

Journal of Pain, 6(9):630636, 2005.

(Cited on pages 73 and 77.)

[144] B. D. Wright and J. M. Linacre. Observations are always ordinal; measurements,

however, must be interval.
(12):857860, 1989.

Archives of Physical Medicine and Rehabilitation,

70

(Cited on pages 74 and 187.)

[145] G. Rasch. On general laws and the meaning of measurement in psychology.

Proc.

Fourth Berkeley Symp. On Math. Statist. And Prob., University of California

Press, 4:321333, 1961.

(Cited on page 76.)

276

REFERENCES
[146] G. Rasch. A mathematical theory of objectivity and its consequences for model
construction.

Invited paper at the Institute of Mathematical Statistics, European

Branch, 1968.

(Cited on page 76.)

[147] E. B. Andersen. Sucient statistics and latent trait models.

(1):6981, 1977.

Psychometrika,

(Cited on page 76.)

[148] D. Andrich. A rating formulation for ordered response categories.

43(4):561573, 1978.

[149] D. Andrich.

42

Psychometrika,

(Cited on page 76.)

An extension of the rasch model for ratings providing both loca-

tion and dispersion parameters.

Psychometrika, 47(1):105113, 1982.

(Cited on

page 76.)

[150] G. N. Masters. A rasch model for partial credit scoring.

149174, 1982.

[151] D. Andrich.
dit scoring.

Psychometrika,

47(2):

(Cited on page 76.)

A general form of rasch's extended logistic model for partial cre-

Applied Measurement in Education,

1(4):363378, 1988.

(Cited on

page 76.)

[152] B. D. Wright. Estimating rasch measures for extreme scores.

Transactions, 12(2):632633, 1998.

(Cited on page 76.)

[153] J. M. Linacre. Estimating rasch measures for extreme scores.

Transactions, 12(2):638, 1998.

Rasch Measurement

Rasch Meascurement

(Cited on page 76.)

[154] D. P. O'Connor. Comparison of two psychometric scaling methods for ratings of

acute musculoskeletal pain.

Pain, 110(1-2):488494, 2004.

(Cited on page 77.)

[155] J. Lai, K. Dineen, B. B. Reeve, J. V. Roenn, D. Shervin, M. McGuire, R. K.

Bode, J. Paice, and D. Cella. An item response theory-based pain item bank can

277

REFERENCES
enhance measurement precision.
(3):278288, 2005.

Journal of Pain and Symptom Management, 30

(Cited on page 77.)

[156] L. E. Caudle, K. A. Williams, and K. Pesudovs.

ophthalmic pain severity measure.
2007.

The eye sensation scale:

an

Optometry and Vision Science, 84(8):752762,

(Cited on page 77.)

[157] C. Decruynaere and L. Plaghki. Mesurer la doleur: les apports de l'analyse de

rasch.

Douleur et Analgsie, 20(2):8590, 2007.

(Cited on page 77.)

[158] C. Decruynaere, J. L. Thonnard, and L. Plaghki. How many response levels do

children distinguish on faces scales for pain assessment?

Pain, 13(6):641648, 2009.

European Journal of

(Cited on page 77.)

[159] P. S. Hinds, J. Yang, J. S. Gattuso, M. Hockenberry, H. Jones, S. Zupanec, C. Li,

V. M. Crabtree, B. N. Mandrell, R. A. Schoumacher, K. Vallance, S. Sanford, and
D. K. Srivastava.

Psychometric and clinical assessment of the 10-item reduced

version of the fatigue scale - child instrument.

Management, 39(3):572578, 2010.

[160] L. Tesio.

Journal of Pain and Symptom

(Cited on page 77.)

Measuring behaviours and perceptions:

rehabilitation research.

Rasch analysis as a tool for

Journal of Rehabilitation Medicine, 35(3):105115, 2003.

(Cited on page 77.)

[161] A. M. Davis, E. M. Badley, D. E. Beaton, J. Kopec, J. G. Wright, N. L. Young,

and J. I. Williams.

Rasch analysis of the western ontario mcmaster (womac)

osteoarthritis index: results from community and arthroplasty samples.

of Clinical Epidemiology, 56(11):10761083, 2003.

Journal

(Cited on page 77.)

[162] R. D. Hayes, H. Liu, K. Spritzer, and D. Cella. Item response theory analyses of
physical functioning items in the medical outcomes study.
3238, 2007.

(Cited on page 77.)

278

Medical Care,

45(5):

REFERENCES
[163] J. A. Teresi, K. Ocepek-Welikson, M. Kleinman, K. F. Cook, P. K. Crane, L. E.
Gibbons, L. S. Morales, M. Orlando-Edelen, and D. Cella. Evaluating measurement equivalence using the item response theory log-likelihood ratio (irtlr) method
to assess dierential item functioning (dif ): applications (with illustrations) to
measures of physical functioning ability and general distress.

search, 16:4368, 2007.

Quality of Life Re-

(Cited on page 77.)

[164] L. S. Walker, J. E. Beck, J. Garber, and W. Lambert. Children's somatization inventory: psychometric properties of the revised form (csi-24).

Psychology, 34(4):430440, 2009.

Journal of Pediatric

(Cited on page 77.)

[165] G. Velde, D. Beaton, S. Hogg-Johnston, E. Hurwitz, and A. Tennant. Rasch analysis provides new insights into the measurement properties of the neck disability
index.

Arthritis and Rheumatism, 61(4):544551, 2009.

(Cited on page 77.)

[166] K. Y. Chang, M. Y. Tsou, K. H. Chan, S. H. Chang, J. J. Tai, and H. H. Chen.

Item analysis for the written test of taiwanese board certication examination
in anaesthesiology using rasch model.
717722, 2010.

British Journal of Anaesthesia,

104(6):

(Cited on page 77.)

[167] T. W. Bouillon, J. Bruhn, L. Radulescu, C. Andresen, T. J. Shafer, C. Cohane,

and S. L. Shafer. Pharmacodynamic interaction between propofol and remifentanil regarding hypnosis, tolerance of laryngoscopy, bispectral index, and electroencephalographic approximate entropy.

Anesthesiology,

100(6):13531372, 2004.

(Cited on pages 77, 78, 79, 88, 113, 128, 130, 131, 133 and 145.)

[168] A. M. Zbinden, S. Petersen-Felix, D. A. Thomson, and F. R. C. Anaes. Anesthetic

depth dened using multiple noxious stimuli during isourane/oxygen anesthesia.
ii hemodynamic responses.

Anesthesiology,

ges 77, 90 and 111.)

279

80(2):261267, 1994.

(Cited on pa-

REFERENCES
[169] T. Kazama, K. Ikeda, and K. Morita. Reduction by fentanyl of the cp50 values of
propofol and hemodynamic responses to various noxious stimuli.
87(2):213227, 1997.

Anesthesiology,

(Cited on pages 77, 90 and 111.)

[170] A. Albertin, A. Casati, L. Federica, V. Roberto, V. Travaglini, P. Bergonzi, and

G. Torri.

The eect-site concentration of remifentanil blunting cardiovascular

responses to tracheal intubation and skin incision during bispectral index-guided

propofol anesthesia.

Anesthesia and Analgesia, 101(1):125130, 2005.

(Cited on

pages 77, 79, 90, 92 and 111.)

[171] M. E. Ausems, C. C. Hug, D. R. Stanski, and A. G. L. Burm. Plasma concentrations of alfentanil required to supplement nitrous oxide anesthesia for general
surgery.

Anesthesiology, 65(4):362373, 1986.

(Cited on page 79.)

[172] P. M. Sanderson, M. O. Watson, and W. J. Russell. Advanced patient monitoring

displays: Tools or continuous informing.
168, 2005.

Anesthesia and Analgesia,

101(1):161

(Cited on page 90.)

[173] M. B. Weinger and C. E. Englund.

Ergonomic and human factors aecting

anesthetic vigilance and monitoring performance in the operating room environment.

Anesthesiology, 73(5):9951021, 1990.

(Cited on page 90.)

[174] A. Castro, F.G. Almeida, P. Amorim, and C.S. Nunes. A wavelet based method
for steady-state detection in anesthesia. In

Engineering in Medicine and Biology

Society, 2009. EMBC 2009. Annual International Conference of the IEEE, pages
954 957, sept. 2009. doi: 10.1109/IEMBS.2009.5334037.

(Cited on pages 112

and 113.)

[175] A. Castro, F. G. Almeida, E. Amadeu, P. Amorim, and C. S. Nunes. Steady-state

detection applied to bis guided anesthesia - a wavelet based method.

Neurosurgical Anesthesiology, 21(4), 2009.

280

Journal of

(Cited on pages 112 and 113.)

REFERENCES
[176] A. Castro, F. G. Almeida, E. Amadeu, P. Amorim, and C. S. Nunes. A wavelet
based algorithm for steady-state detection: application in bis guided anesthesia.

Anesthesiology, page A1296, 2009.

(Cited on pages 112 and 113.)

[177] A. Castro, C. S. Nunes, P. Amorim, and F. G. Almeida.

Anaesthesia steady-

state index: a combined measure of bis, blood pressure and heart rate.

Journal of Anaesthesiology, 27(47):64, 2010.

European

(Cited on pages 112 and 117.)

[178] T. Jiang, B. Chen, X. He, and P. Stuart. Application of steady-state detection

method based on wavelet transform.
(4):569578, 2003.

[179] Z. Szab.

Computers and Chemical Engineering,

27

(Cited on page 114.)

Contour image data compression using spline wavelets.

Electrical Engineering, 55(11-12):290295, 2004.

Journal of

(Cited on page 114.)

[180] M. Bahoura, M. Hassani, and M. Hubin. Dsp implementation of wavelet transform

for real time ecg wave forms detection and heart rate analysis.

and Programs in Biomedicine, 52(1):35  44, 1997.

Computer Methods

(Cited on page 114.)

[181] S. Mallat and S. Zhong. Characterization of signals from multiscale edges.

Analysis and Machine Intelligence, IEEE Transactions on,

1992.

Pattern

14(7):710 732, jul

(Cited on pages 114, 117 and 119.)

[182] L. Ljung, editor.

System Identication: theory for the User.

Prentice Hall, 1987.

(Cited on page 136.)

[183] A. Castro, P. Amorim, C. S. Nunes, and F. G. Almeida. Seps may objectively

translate nociception/anti-nociception balance with anesthetics.

rosurgical Anesthesiology, 2011.

Journal of Neu-

(Cited on page 149.)

[184] K. Bornhvd, M. Quante, V. Glauche, B. Bromm, C. Weiller, and C. Bchel.

Painful stimuli evoke dierent stimulus-response functions in the amygdala, pre-

281

REFERENCES
frontal, insula and somatosensory cortex: a single-trial fmri study.
13261336, 2002.

Brain, 125(6):

(Cited on pages 150 and 161.)

[185] K. Roberts, A. Papadaki, C. Gonalves, M. Tighe, D. Atherton, R. Shenoy,

D. McRobbie, and P. Anand.

Contact heat evoked potentials using simultane-

ous eeg and fmri and their correlation with evoked pain.
(8):8, 2008.

BMC Anesthesiology, 17

(Cited on pages 150 and 162.)

[186] C. Christmann, C. Koeppe, D. F. Braus, M. Ruf, and H. Flor. A simultaneous

eeg-fmri study of painful electric stimulation.

Neuroimage, 34(4):14281437, 2007.

(Cited on pages 150 and 162.)

[187] I. Tracey. Can neuroimaging studies identify pain phenotypes in humans?

Reviews - Neurology, 7(3):173181, 2011.

Nature

(Cited on pages 150 and 161.)

[188] I. Crabb, C. Thornton, K. M. Konieczko, A. Chan, R. Aquilina, N. Frazer, C. J.

Dor, and D. E. Newton. Remifentanil reduces auditory and somatosensory evoked responses during isourane anaesthesia in a dose-dependent manner.

Journal of Anaesthesia, 76(6):795801, 1996.

British

(Cited on pages 150, 175, 183, 185

and 188.)

[189] A. Kumar, A. Bhattacharya, and N. Makhija.

anaesthesia and analgesia.

Evoked potential monitoring in

Anaesthesia, 55(3):225241, 2000.

(Cited on pages 151,

152, 157, 158, 175, 185 and 186.)

[190] C. Thornton and R. M. Sharpe. Evoked responses in anaesthesia.

of Anaesthesia, 81(5):771781, 1998.

British Journal

(Cited on pages 151, 152, 154, 156, 157, 163

and 186.)

[191] L. Arendt-Nielsen and P. Bjerring. Sensory ans pain threshold characteristics to

laser stimuli.
1988.

Journal of Neurology, Neurosurgery, and Psychiatry,

(Cited on pages 154 and 186.)

282

51(1):3542,

REFERENCES
[192] L. Arendt-Nielsen and P. Bjerring. Reaction times to painless and painful
argon laser stimulation.
1988.

co2

and

European Journal of Applied Physiology, 58(3):266273,

(Cited on pages 154 and 186.)

[193] E. H. Chudler and W. K. Dong.

potentials.

The assessment of pain by cerebral evoked

Pain, 16(3):221244, 1983.

(Cited on page 155.)

[194] S. V. Revenko, V. V. Ermishkin, and L. Ya. Selektor.

Excitation of feline cu-

taneous mechanosensitive c-ber sensory units by intraarterial administration of

potassium ions.

Neurophysiology, 20(2):105110, 1988.

(Cited on page 155.)

[195] X. Navarro, J. J. Lazaro, M. Buti, E. Verdu, P. J. Fabregas, and B. Castellano.

Electrophysiological evaluation of spinal reexes during epidural anesthesia in an
experimental model.

Muscle and Nerve, 19(1):2936, 1996.

(Cited on page 155.)

[196] M. A. Smith and J. D. Gray. Age-related dierences in sensitivity to the antinociceptive eects of opioids in male rats.
445453, 2001.

Psychopharmacology (Berl),

156(4):

(Cited on page 155.)

[197] A. C. Chen and C. R. Chapman.

Aspirin analgesia evaluated by event-related

potentials in man: Possible central action in brain.

39(4):359364, 1980.

Experimental Brain Research,

(Cited on pages 157 and 159.)

[198] A. Carmon, J. Mor, and J. Goldberg. Evoked cerebral responses to noxious thermal stimuli in humans.

Experimental Brain Research, 25(1):103107, 1976.

(Cited

on page 157.)

[199] A. Carmon, Y. Dotan, and Y. Sarne. Correlation of subjective pain experience

with cerebral evoked responses to noxious thermal stimulations.

Brain Research, 33(3-4):445453, 1978.

Experimental

(Cited on page 159.)

[200] A. Carmon, Y. Friedman, R. Coger, and B. Kenton. Single trial analysis of evoked

283

REFERENCES
potentials to noxious thermal stimulation in man.

Pain, 8(1):2132, 1980.

(Cited

on page 159.)

[201] B. L. Loughnan, P. S. Sebel, D. Thomas, C. F. Rutherfoord, and H. Rogers.

Evoked potentials following diazepam or fentanyl.
1987.

Anaesthesia,

42(2):195198,

(Cited on page 159.)

[202] L. Arendt-Nielsen and P. Bjerring.

Laser-induced pain for evaluation of local

anaesthesia: A comparison of topical application (emla) and local injection (lidocaine).

Anesthesia and Analgesia, 67(2):115123, 1988.

(Cited on page 159.)

[203] L. Arendt-Nielsen, B. Oberg, and P. Bjerring. Analgesic ecacy of i.m. alfentanil.

British Journal of Anaesthesia, 65(2):164168, 1990.

(Cited on page 159.)

[204] C. Thornton, P. Creagh-Barry, C. Jordan, N. P. Lu, C. J. Dor, M. Henley, and

D. E. Newton. Somatosensory and auditory evoked responses recorded simultaneously:

dierential eects of nitrous oxide and isourane.

Anaesthesia, 68(5):508514, 1992.

British Journal of

(Cited on pages 160 and 186.)

[205] Y. Granovsky, M. Granot, RR Nir, and D. Yarnitsky.

Objective correlate of

subjective pain perception by contact heat-evoked potentials.

Pain, 9(1):5363, 2008.

(Cited on page 160.)

[206] Encyclopaedia brittannica, inc. on-line, 2007.

brain.

The Journal of

Functional areas of the human

(Cited on page 161.)

[207] S. Ohara, N. E. Crone, N. Weiss, R. D. Treede, and F. A. Lenz.

Amplitudes

of laser evoked potential recorded from primary somatosensory, parasylvian and

medial frontal cortex are graded with stimulus intensity.
2004.

(Cited on page 160.)

284

Pain, 110(1-2):318328,

REFERENCES
[208] E. Kochs, E. Scharein, O. Mllenberg, B. Bromm, and J. am Schulte Esch. Analgesic ecacy of low-dose ketamine: Somatosensory-evoked responses in relation to
subjective pain ratings.

Anesthesiology, 85(2):304314, 1996.

(Cited on page 162.)

[209] F. von Dincklage, M. Hackbarth, M. Schneider, J. H. Baars, and B. Rehberg.

Introduction of a continual riii reex threshold tracking algorithm.
1260:2429, 2009.

Brain Research,

(Cited on page 163.)

[210] C. R. France, J. L. Rhudy, and S. McGlone. Using normalized emg to dene the
nociceptive exion reex (nfr) threshold: further evaluation of standardized nfr
scoring criteria.

Pain, 145(1-2):211218, 2009.

[211] E. Kochs and G. Schneider.

monitors.
16, 2001.

Algesimetry:

(Cited on page 163.)

Concepts for intelligent anesthesia

The Internet Journal of Thoracic and Cardiovascular Surgery,

2(1):

(Cited on page 163.)

[212] E. Baran, B. M. Abrams, A. Cassvan, R. Q. Cracco, A. A. Eisen, J. Kimura, W. T.

Liberson, W. C. Stolov, G. H. Kraft, W. J. Litchy, and M. J. Amino. Guidelines
for somatosensory evoked potentials.

Electrodiagnostic Medicine, 1998.

American Association of Neuromuscular and

(Cited on page 164.)

[213] J. R. Toleikis. Intraoperative monitoring using somatosensory evoked potentials.

American Society of Neurophysiological Monitoring, 2002.

(Cited on page 164.)

[214] ACNS. Guidelines on short-latency somatosensory evoked potentials.

Clinical Neurophysiology Society, 2008.

American

(Cited on page 164.)

[215] E. S. Fu, Y. Zhang, Q. Li, R. W. Morris, and R. C. Levitt. Baseline dierences

in mechanical nociception: Genetic regulation.

Anesthesiology, page A505, 2010.

(Cited on page 186.)

[216] E. Anker-Mller, N. Spangsberg, L. Arendt-Nielsen, P. Schultz, M. S. Kristensen,

285

REFERENCES
and P. Bjerring. Subhypnotic doses of thiopentone and propofol cause analgesia to
experimentally induced acute pain.
1991.

British Journal of Anaesthesia, 66(2):185188,

(Cited on pages 188 and 198.)

[217] T. C. O'Connor and S. E. Abram. Inhibition of nociception-induced spinal sensitization by anesthetic agents.

Anesthesiology,

82(1):259266, 1995.

(Cited on

pages 188 and 198.)

[218] A. Castro, C. S. Nunes, P. Amorim, and F. G. Almeida.

administration - sliding-mode control.

Congress, 2009.

Oral Presentation.

Propofol automatic

Total Intravenous Anesthesia - TCI World

(Cited on pages 193 and 202.)

[219] A. Castro, C. S. Nunes, P. Amorim, and F. G. Almeida. Automatic control of

propofol's eect on the state entropy using a sliding-mode technique (a simulation study).

European Journal of Anaesthesiology,

26(45):34, 2009.

(Cited on

pages 193 and 202.)

[220] N. Ty Smith, M. L. Quinn, J. Flick, Y. Fukui, R. Fleming, and J. R. Coles. Automatic control in anesthesia: a comparison in performance between the anesthetist
and the machine.

Anesthesia and Analgesia,

63(8):715722, 1984.

(Cited on

pages 193 and 200.)

[221] G. Schils, F. Sasse, and V. Rideout. Automatic control of anesthesia using two
feedback variables.

Annals of Biomedical Engineering.

(Cited on pages 193, 198

and 200.)

[222] O. Simanski, A. Schubert, R. Kaehler, M. Jandat, J. Bajoratt, R. Hofmockel, and

B. Lampe. Automatic drug delivery in anesthesia: From the beginning until now.
In

Control Automation, 2007. MED '07. Mediterranean Conference on,

6, june 2007.

(Cited on pages 194, 198 and 200.)

286

pages 1

REFERENCES
[223] A. R. Absalom and G. N. C. Kenny. Closed-loop control of propofol anaesthesia
using bispectral index

T M : performance assessment in patients receiving computer-

controlled propofol and manually controlled remifentanil infusions for minor surgery.

British Journal of Anaesthesia, 90(6):737741, 2003.

(Cited on page 198.)

[224] C. Schaniel, A. Gentilini, M. Morari, C. Bieniok, R. Wymann, and T.W. Schnider. Perioperative closed-loop control of analgesia in humans. In

Engineering in

Medicine and Biology Society, 2001. Proceedings of the 23rd Annual International
Conference of the IEEE,

volume 1, page 449 vol.1, 2001.

(Cited on pages 198

and 200.)

[225] S. Yelneedi, S. Lakshminarayanan, and G.P. Rangaiah. Advanced regulatory controller for automatic control of anesthesia. In

IFAC, Proceedings of the 17th World

Congress The International Federation of Automatic Control, pages 1163611641,

2008.

(Cited on page 198.)

[226] P. Dua, V. Dua, and E.N. Pistikopoulos.

anesthesia.

In

Model based parametric control in

European Symposium on Computer-Aided Process Engineering,

pages 10151020, 2005.

(Cited on page 198.)

[227] E. Mortier, M. Struys, T. De Smet, L. Versichelen, and G. Rolly.

controlled administration of propofol using bispectral analysis.
(8):749754, 1998.

Closed-loop

Anaesthesiaa, 53

(Cited on page 198.)

[228] C.W. Frei, M. Derighetti, M. Morari, A.H. Glattfelder, and A.M. Zbinden. Improving regulation of mean arterial blood pressure during anesthesia through estimates of surgery eects.
1456 1464, nov. 2000.

Biomedical Engineering, IEEE Transactions on, 47(11):

(Cited on page 200.)

[229] R. K. Ellerkmann, M. Soehle, T. M. Alves, V. M. Liermann, I. Wenningmann,

H. Roepcke, S. Kreuer, A. Hoeft, and J. Bruhn. Spectral entropy and bispectral

287

REFERENCES
index as measures of the electroencephalographic eects of propofol.

and Analgesia, 102(5):14561462, 2006.

(Cited on pages 203, 204 and 205.)

[230] C. Edwards and S. K. Spurgeon, editors.

Applications.

Sliding Mode Control - Theory and

Taylor and Francis, London, 1998.

288

Anesthesia

(Cited on pages 203 and 208.)