Book of Pharmacovigilance

Quantitative Signal Detection
in Pharmacovigilance
Eugène P. van Puijenbroek

QUANTITATIVE SIGNAL DETECTION IN PHARMACOVIGILANCE
Eugène Paul van Puijenbroek
Title: Quantitative Signal Detection in Pharmacovigilance

Thesis Utrecht– With ref. – With summary in Dutch
ISBN 90-393-2796-3
©
2001 E.P. van Puijenbroek
Printed by: Print Partners Ipskamp, Enschede

QUANTITATIVE SIGNAL DETECTION IN PHARMACOVIGILANCE
Kwantitatieve Signaaldetectie in de Geneesmiddelenbewaking
(met een samenvatting in het Nederlands)
PROEFSCHRIFT
Ter verkrijging van de graad van doctor aan de Universiteit Utrecht

op gezag van de Rector Magnificus, Prof. dr. WH Gispen,
ingevolge het besluit van het College voor Promoties
in het openbaar te verdedigen
op maandag 1 oktober 2001
des middags te 16:15 uur
door
Eugène Paul van Puijenbroek

geboren op 28 september 1960
te Tilburg
Promotores
Prof. dr. HGM Leufkens

Department of Pharmacoepidemiology and Pharmacotherapy, Utrecht Institute for
Pharmaceutical Sciences
Prof. dr. ACG Egberts

Pharmaceutical Sciences
The work presented in this thesis was performed at the Netherlands

Pharmacovigilance Foundation Lareb and the Utrecht Institute for
Pharmaceutical Sciences.
In memory of my parents
To Emmy, Linde and Sterre

CONTENTS
Introduction 9
PART 1 METHODOLOGICAL ASPECTS OF QUANTITATIVE SIGNAL DETECTION 21
1.1 A comparison of measures of disproportionality for signal detection

in spontaneous reporting systems for adverse drug reactions 23
1.2 On the assessment of adverse drug reactions from spontaneous

reporting systems. The influence of underreporting on odds ratios 41
PART 2 PRACTICAL APPROACHES OF QUANTITATIVE SIGNAL DETECTION 65
2.1 Different risks for NSAID-induced anaphylaxis 67
2.2 Signalling possible drug-drug interactions in a spontaneous reporting

system: delay of withdrawal bleeding during concomitant use of oral
contraceptives and itraconazole 81
2.3 Detecting drug-drug interactions using a database for spontaneous

adverse drug reactions: an example with diuretics and non-steroidal
anti-inflammatory drugs 93
2.4 Association between terbinafine and arthralgia, fever and urticaria:

symptoms or syndrome? 107
PART 3 IMPLEMENTATION OF THE QUANTITATIVE APPROACH 123
3.1 Determinants of signal selection in a spontaneous reporting

system for adverse drug reactions 125
3.2 Practical application of quantitative signal detection in the

spontaneous reporting system for adverse drug reactions in
the Netherlands 141
General discussion 155
Summary 171
Samenvatting 177
Addendum to Chapter 1.2 183
Addendum to Chapter 2.4 187
Dankwoord 191
List of publications 193
Curriculum vitae 197

Introduction
Introduction
The need for pharmacovigilance
The development of a new drug is a challenging but also costly process since it
needs to focus on quality, efficacy as well as safety. Concerning the latter aspect,
a reliable overview of the possible adverse drug reactions (ADRs) associated with
the new drug is indispensable. Some of the ADRs can be predicted on the basis of
experiences with pharmacologically related drugs. Others will be detected in the
clinical trials conducted before marketing authorisation. Clinical trials, however,
are primarily designed for and more specifically aimed at the efficacy assessments
of new drugs. For the detection of ADRs, clinical trials have certain limitations.1
As some ADRs are rare or have a long latency period and others only occur after
prolonged use of the drug or are restricted to specific patient groups, they are
generally difficult to detect or predict at this stage.2 Consequently, many ADRs
only become manifest after the drug has been marketed.
The 1960s brought an awareness of the importance of a systematic surveillance of
drugs for ADRs, also after marketing, with the detection of the association
between thalidomide, a hypnotic drug, and severe congenital malformations of the
limbs of children whose mothers had used this drug in the early stages of
pregnancy. Thalidomide was introduced in 1956. Although the congenital
malformation involved, phocomelia, is a rare condition in absence of the use of
this drug, it was not until 1961 that the first report of the possible association
between thalidomide and phocomelia was published by McBride.3 In the
following years various other serious ADRs in suspected relation to a drug were
detected after marketing. Examples are diethylstilbestrol (DES), which has been
associated with an increased incidence of vaginal clear-cell adenocarcinoma in
women and their daughters, the latter having been exposed to DES in utero,4,5 and
the association between the use of practolol and the oculomucocutaneous
syndrome.6 Also, more recently, additional serious ADRs were detected, some of
them for drugs that had been on the market for quite some time. Examples are the
visual field defects occurring during the use of vigabatrin,7,8 introduced on the
international market in 1989, and the risk of valve disorders associated with the
use of fenfluramine-phentermine therapy for the treatment of obesitas.9
The dramatic experiences with thalidomide in the beginning of the 1960s gave
rise to the establishment of national pharmacovigilance centres in many western
countries, enabling a more systematic surveillance of drugs for possible ADRs.
11
Introduction
These centres usually rely on the method of ‘spontaneous reporting’ to monitor

for unexpected ADRs. Health professionals are urged (in most countries) or
obliged by law to report their suspicions of possible associations between drugs
and ADRs to these centres. In essence, the centres managing such ‘spontaneous
reporting systems’ (SRSs) function as a diagnostic tool. They are instruments that
help health authorities, health professionals and industry gain more insight into
the safety aspects of the drugs approved for marketing. Signals derived from these
systems may either confirm or disprove the validity of the efficacy-safety trade-
off of the drugs involved. The detection of any (additional) ADRs may lead to
changes in the official product information or, more rarely, to the drug’s
suspension or withdrawal from the market. Between 1972 and 1994, 59
pharmaceutical products were recalled from the British market from a total of 583
newly approved substances. In addition to commercial reasons, safety issues (4%
of the marketed products) were the primary cause of market withdrawal.10 During
the period of 1981 to 2000, the Food and Drug Administration (FDA) in the
United States approved 543 new molecular entities. Fourteen of these approvals
(2.6%) were rescinded for safety reasons. The number of withdrawals increased
slightly during this period. Between 1986 and 1990 two drugs (1.9%) were
withdrawn and five (2.7%) between 1996 and 2000.11 Unfortunately, even though
pre-marketing research has been improved over the years, still, it is not until a
drug has been used in clinical practice for some length of time that unpredicted
safety-related problems manifest themselves, as is reflected by the above-
mentioned withdrawal rates of approved and marketed drugs.
Also the European Agency for the Evaluation of Medicinal Products (EMEA)
observed a clear increase in the number of withdrawn applications in the
centralised procedure in the period between 1995 and 1998. The average
withdrawal rate in this period was 22.5% of which 50% concerned approvals
rescinded on the grounds of safety.12 Since these applications concern drugs that
are not yet available on the market, this may reflect a growing concern for safety
aspects.
Despite all the controversies that have arisen around the spontaneous reporting
systems (SRS) over the past decades and the many debates held on their
limitations, the approach has remained an important tool for pharmacovigilance.
One of the aims of the SRSs is the so-called signal detection as a first step in the
analysis of a possible association between a suspected drug and an ADR.
12
Introduction
Meyboom defined a signal as a set of data constituting a hypothesis that is

relevant to the rational and safe use of a drug in humans.13 A signal was defined
by the WHO Collaborating Centre for International Drug Monitoring as reported
information on a possible causal relationship between an adverse event and a
drug, of which the relationship is unknown or incompletely documented
previously.14 Both definitions share the characteristic that new knowledge about
the safety of a drug is generated. They explicitly cover a broad area that is not
necessarily restricted to the detection of adverse events and a drug, but may also
imply the detection of other aspects, for instance risk factors in patients, drug-
drug interactions, drug-food interactions and drug-related syndromes. Signal
detection is a complex process in which the identification of unknown
associations, causality assessment and quantification of the associations are
closely related. As with other diagnostic systems, an optimal balance between
sensitivity and specificity, false positive and false negative signals must be found.
In other words, an optimal signal-to-noise ratio must be established. After a signal
has been identified often other methods are required to confirm or repudiate the
signal, to determine its incidence and to identify patients at high or low risk. In
practice, however, many signals will never be checked mainly as a result of
limited resources. Evidence supplied by an SRS often remains the only
information for regulatory bodies to act on.
In the Netherlands, and on behalf of the Medicines Evaluation Board,15 the
Pharmacovigilance Foundation Lareb is responsible for the management and
maintenance of the spontaneous reporting system concerning drugs that have been
approved for the Dutch market. Both physicians and pharmacists are urged to
report suspected ADRs to Lareb. In the Dutch healthcare system, patients usually
obtain most of their drugs from the same pharmacist. Consequently, in the
majority of cases a complete medication history of the patient is available and not
only comprises the drugs prescribed by the patient’s general practitioner, but also
those drugs that have been prescribed by specialists. Co-operation between
pharmacist and physician therefore enables an optimal documentation of the SRS
reports. This not only enhances the quality of the clinical documentation, but it
also provides a database containing reliable and high-quality pharmaceutical
information. Currently, a total of 3,000 to 3,500 reports are received and
evaluated annually and the Lareb database now contains over 30,000 reports. The
reports are sent to the MEB on a weekly basis. The board subsequently forwards
13
Introduction
the signals derived from them, to the EMEA. In addition, and on a quarterly basis,
the reports are sent to the WHO Centre for International Drug Monitoring (the
Uppsala Monitoring Centre).
Methodology of spontaneous reporting
First suspicions of possible ADRs are often generated and published by health
professionals in the form of case reports.16,17 Pharmacovigilance or ‘post
marketing surveillance’ is mainly based on spontaneous reports of such
observations, which obviously play a vital role in the detection of new
associations.18 Physicians and pharmacists report to an SRS on a voluntary, i.e.
non-interventional, basis. Due to the ‘spontaneous’ character of the reporting, the
method has some limitations. The most noticeable problem is (selective)
underreporting.19,20 Stephens, among others, has provided an overview of various
factors responsible for underreporting.21 Unfortunately, too many factors are
involved in underreporting to make a reliable estimate of its extent. Since not all
ADRs are reported, the data set of an SRS does not necessarily constitute a valid
representation of the ADRs occurring in daily practice. Although other techniques
are available for the detection of ADRs for marketed drugs, for instance
‘prescription event monitoring’ and ‘intensive monitoring’,22 spontaneous
reporting still is an important method of post marketing surveillance.23 Data sets
from spontaneous reporting systems are mainly used for the analysis of
associations between a suspected ADR and a particular drug, but they may also be
applied for the analysis of more complex relations, for instance drug-drug
interactions, clustering of symptoms and the identification of risk factors in
patients.
Quantitative signal detection
Methods applied in analytical epidemiology, such as the case control design,

which are primarily used for the evaluation of signals, may also serve as
techniques for signal detection.13 Statistical quantitative approaches can be
efficient tools to analyse the data set of an SRS and help minimise the risk that
possible signals are missed. The reports sent to an SRS are coded with different
kinds of information, like patient characteristics and data concerning
14
Introduction
(concomitant) medication and the suspected ADR(s). Among these covariates

different relationships relevant for signal detection may exist. Evaluation of such
complex relations can be more efficiently carried out by means of a computer-
assisted quantitative analysis.
Quantitative signal detection refers to the application of numerical procedures in
pharmacovigilance, which can not only be used for signal detection and analysis
of associations between ADR(s) and drug(s), but also for drug-drug interactions,
drug-related syndromes, risk factors in patients and time trends after marketing of
the drug under investigation. Finney already described the basic principles of the
majority of these methods in 1974,24 although actual, practical application of the
concepts was only made possible in recent years due to the introduction of and
advances in information technology. In the past decades, various statistical
measures have been proposed for the application of computer-assisted
quantitative signal detection procedures. Examples of such measures are the
proportional ADR reporting ratio,25,26 the Reporting Odds Ratio,27-29 and the
Bayesian Propagation Neural Network analysis (BCPNN) of the Uppsala
Monitoring Centre.30-32 Although the rationale and the methodology of the various
approaches differ, they all share the characteristic that they search the databases
for disproportionality. In other words, they provide an answer to the question:
Does the number of observed cases exceed the number of expected cases? The
statistical measures of disproportionality all express the extent to which the
reported ADR is associated with the suspected drug compared with the other
drugs in the database. The occurrence of ADRs related to other drugs in the
database is used as a proxy for the background incidence of ADRs. Given the
possibility to generate signals based on a quantitative approach, computerisation
of this process may provide a preliminary sifting of the large amount of
information in the database. A survey by Olsson conducted in 1999 showed that
16 out of 43 national pharmacovigilance centres applied a combination of some
kind of automatic screening method and a manual screening in the detection of
ADRs. For 7 centres no specific method was specified. Unfortunately,
methodological details about the automated approaches used in these centres were
not available. 33
Signal assessment is often described as a sequential process. After a suspicion has
been raised concerning a possible relation between a drug and an ADR it needs to
be corroborated by the accumulation of additional data. The final step, hypothesis
15
Introduction
testing, involves the confirmation and quantification of the relation between drug
and ADR by epidemiological methods.34 Signals generated by quantitative
approaches should be carefully evaluated in combination with additional
information, like other information in the reports, information from the literature,
etc. Signal strengthening does not always require the use of (pharmaco-)
epidemiological techniques, and can also be carried out by accumulating and
weighing information from different sources. Although in every step of the
process both individual case reports and analytical techniques are involved, the
relative importance of these approaches differs per step (Figure 1).
Case Analytical
reports techniques
Signal Signal Signal

detection strengthening testing
Figure 1. The process of signal detection, strengthening and testing, and the relative
importance of information derived from case reports and from analytical techniques
Hypotheses may be formulated in the literature based on observations in medical

practice, or they may follow from the analyses of datasets of spontaneous
reporting systems. With respect to reports sent to SRSs, signal detection is
traditionally based on a ‘case-by-case’ approach. Trained assessors review each
report sent to an SRS and evaluate if the reports show a pattern representing a
possible signal. In this more or less subjective process quantitative, clinical and
pharmacological aspects are taken into account. The results yielded by the
quantitative approaches, presented by a point estimate and confidence interval,
may suggest that a clear and objective judgement can be made with respect to the
existence of a true relationship. However, they should always be regarded as
having a warning function and not as actual evidence of a possible relationship.
The presence of a statistically significant result does not necessarily imply an
actual causal relationship between the ADR and the drug, nor does the absence of
a statistically significant result necessarily disprove such a possible relationship.
16
Introduction
Objectives and outline of this thesis
Before 1960, virtually no systematic surveillance of drugs for unexpected adverse

drug reactions was performed after marketing. The detection of ADRs depended
to a large extent on the observation by physicians or pharmacists and subsequent
communications and publications in the literature. Despite the methodological
limitations inherent to the method of spontaneous reporting, it offers a relatively
fast and efficient way of generating signals that can be used for balancing efficacy
and safety.35 The primary goal of pharmacovigilance is to generate signals that
indicate the presence of an association between a drug and an ADR. Actual
testing of the association should be carried out by other methods. Thanks to the
availability of (new) statistical techniques spontaneous reporting systems can be
used more efficiently. Datamining techniques can be applied and more complex
relationships can be analysed.
Reports sent to Lareb may concern one or more drugs and one or more ADRs
(Figure 2). Generally, in signal detection we are interested in the relationship
between the patient (characteristics), suspected drug(s) and clinical events. This
most often concerns one drug and one event. Nevertheless, also other
combinations may be of interest. For example, if a certain ADR occurs more often
than expected in the event two drugs are used concomitantly, this may indicate
the existence of a drug-drug interaction. Similarly, a clustering of two ADRs may
indicate the existence of a drug-related syndrome, i.e. the clustering of two or
more symptoms.24,36 All three approaches are used in the analyses of the reports of
an SRS such as the Lareb data set.
Patient data Association between drug and ADR

Drug-drug interactions
Syndromes
ADR Drug
ADR Drug
Figure 2. The relationship between patient, ADR and drug
17
Introduction
The objective of this thesis is to evaluate the value and possibilities of quantitative
signal detection in spontaneous reporting systems and to add further conceptual
insight both into the analysis of the relationship between adverse drug reactions
and a suspected drug as well as into complex relationships like drug-drug
interactions and drug-related syndromes. The first part of this thesis focuses on
the methodology of quantitative signal detection. In Chapter 1.1 the performance
of various measures of disproportionality used by SRS centres for the analyses of
data of spontaneous reports of ADRs is compared by applying these measures to
the dataset of Lareb. Chapter 1.2 focuses on the influence of underreporting when
using Reporting Odds Ratios on the assessment of ADRs, drug-drug interactions
and drug-related syndromes. In Part 2 of this thesis examples of quantitative
signal detection are given and examples of new approaches are introduced.
Chapter 2.1 deals with the analysis of a possible association between a single drug
and a suspected ADR. As illustrated in Chapter 2.2 and 2.3 data sent to an SRS
also allow analyses of possible drug-drug interactions. Likewise, they can be used
to study the possibility of clustering of symptoms, as is illustrated in Chapter 2.4.
In the third part of the thesis the implementation of quantitative approaches in
signal detection is discussed. Chapter 3.1 deals with the determinants responsible
for the selection of signals that are disseminated to the Medicines Evaluation
Board and the value of the results of the quantitative signal selection in this
process. Although in various centres quantitative techniques are being developed,
they are generally not routinely implemented. Therefore, practical aspects of the
implementation of quantitative signal detection are described in Chapter 3.2.
Finally, in the General discussion we will elaborate on the methodological
considerations and practical implications of a quantitative approach. Additionally,
suggestions for future research are made.
References
1. Ioannidis JP, Contopoulos-Ioannidis DG. Reporting of safety data from randomised trials.
Lancet 1998;352:1752-3.
2. Stephens MDB. The pre-marketing establishment of the side-effect profile of a new drug. In:
Stephens MDB, Talbot JCC, Routledge PA (eds). Detection of new adverse drug reactions.
New York: Grove's dictionaries inc, 1998:197-253.
3. McBride WG. Thalidomide and congenital malformations. Lancet 1961:1358.
4. Robboy SJ, Noller KL, O'Brien P, Kaufman RH, Townsend D, Barnes AB, Gundersen J,
Lawrence WD, Bergstrahl E, McGorray S. Increased incidence of cervical and vaginal
18
Introduction
dysplasia in 3,980 diethylstilbestrol-exposed young women. Experience of the National

Collaborative Diethylstilbestrol Adenosis Project. JAMA 1984;252:2979-83.
5. Robboy SJ, Young RH, Welch WR, Truslow GY , Prat J, Herbst AL, Scully RE. Atypical
vaginal adenosis and cervical ectropion. Association with clear cell adenocarcinoma in
diethylstilbestrol-exposed offspring. Cancer 1984;54:869-75.
6. Wright P. Untoward effects associated with practolol administration: oculomucocutaneous
syndrome. Br Med J 1975;1:595-8.
7. Wong IC, Mawer GE, Sander JW. Severe persistent visual field constriction associated with
vigabatrin. Reaction might be dose dependent. BMJ 1997;314:1693-4.
8. Blackwell N, Hayllar J, Kelly G. Severe persistent visual field constriction associated with
vigabatrin. Patients taking vigabatrin should have regular visual field testing. BMJ
1997;314:1694.
9. Connolly HM, Crary JL, McGoon MD, Hensrud DD, Edwards BS, Edwards WD, Schaff
HV. Valvular heart disease associated with fenfluramine-phentermine. N Engl J Med
1997;337:581-8.
10. Jefferys DB, Leakey D, Lewis JA, Payne S , Rawlins MD. New active substances authorized
in the United Kingdom between 1972 and 1994. Br J Clin Pharmacol 1998;45:151-6.
11. Food and Drug Administration. CDER 2000 Report to the nation. improving public health
through human drugs. Rockville, Maryland. 2001.
12. Report on withdrawn centralised applications 1995-1998. European Agency for the
Evaluation of Medicinal Products. 1999; EMEA/H/14994/99.
13. Meyboom RH, Egberts AC, Edwards IR, Hekster YA, de Koning FH, Gribnau FW.
Principles of signal detection in pharmacovigilance. Drug Saf 1997;16:355-65.
14. Edwards IR, Aronson JK. Adverse drug reactions: definitions, diagnosis, and management.
Lancet 2000;356:1255-9.
15. Broekmans AW, Lekkerkerker JFF, De Koning GHP, Vree PH. Nieuwe regels voor het
melden van bijwerkingen in Nederland na 1995. Ned Tijdschr Geneeskd 1996;140:1166-7.
16. Venning GR. Identification of adverse reactions to new drugs. II (continued): How were 18
important adverse reactions discovered and with what delays? Br Med J 1983;286:365-8.
17. Venning GR. Identification of adverse reactions to new drugs. II-How were 18 important
adverse reactions discovered and with what delays? Br Med J 1983;286:289-92.
18. Edwards IR, Lindquist M, Wiholm BE, Napke E. Quality criteria for early signals of
possible adverse drug reactions. Lancet 1990; 336:156-8.
19. Alvarez-Requejo A, Carvajal A, Begaud B, Moride Y, Vega T, Arias LH. Under-reporting
of adverse drug reactions. Estimate based on a spontaneous reporting scheme and a sentinel
system. Eur J Clin Pharmacol 1998;54:483-8.
20. Begaud B, Miremont F, Tubert-Bitter P. Expected number of adverse reactions in a sample.
Post Marketing Surveillance 1993;799-118.
21. Stephens MDB. Introduction. In: Stephens MDB, Talbot JCC, Routledge PA (eds).
Detection of new adverse drug reactions. New York: Grove's dictionaries inc, 1998:1-59.
22. Inman WHW, Gill EP (eds). Monitoring for Drug Safety. Lancaster: MTP press, 1986.
23. Bortnichak EA, Wanju SD. Epidemiologists and adverse event data - A challenge to the
field. Pharmacoepidemiol Drug Saf 1999;8:457-61.
19
Introduction
24. Finney DJ. Systemic signalling of adverse reactions to drugs. Methods Inf Med 1974;13:1-
10.
25. Evans S. Pharmacovigilance: a science or fielding emergencies? Stat Med 2000;19:3199-
209.
26. Evans SJW, Waller P, Davis S. Proportional Reporting Ratios: The uses of epidemiological
methods for signal generation. Pharmacoepidemiol Drug Saf 1998;7(Suppl 2): S102-S102.
27. Moore N, Kreft-Jais C, Haramburu F, Noblet C, Andrejak M, Ollagnier M, Begaud B.
Reports of hypoglycaemia associated with the use of ACE inhibitors and other drugs: a
case/non-case study in the French pharmacovigilance system database. Br J Clin Pharmacol
1997;44:513-8.
28. Egberts ACG, Meyboom RHB, de Koning GHP, Bakker A, Leufkens HGM. Non-puerperal
lactation associated with antidepressant drug use. Br J Clin Pharmacol 1997;44:277-81.
29. Stricker BHCh, Tijssen JGP. Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;45:1177-84.
30. Bate A, Lindquist M, Edwards IR, Olsson S, Orre R, Lansner A, De Freitas RM. A Bayesian
neural network method for adverse drug reaction signal generation. Eur J Clin Pharmacol
1998;54:315-21.
31. Lindquist M, Ståhl M, Bate A, Edwards IR, Meyboom RH. A retrospective evaluation of a
data mining approach to aid finding new adverse drug reaction signals in the WHO
international database. Drug Saf 2000;23:533-42.
32. Orre R, Lansner A, Bate A, Lindquist M. Bayesian neural networks with confidence
estimations applied to data mining. Computational Statistics and Data Analysis 2000;34:473-
93.
33. Olsson S (ed). National Pharmacovigilance systems, country profiles and overview. the
Uppsala Monitoring Centre, Uppsala, 1999.
34. Meyboom RHB, Hekster YA, Egberts ACG, Edwards IR. Causal or casual? The role of
causality assessment in pharmacovigilance. Drug Saf 1997;16:374-89.
35. Rossi AC, Knapp DE, Anello C, O'Neill RT, Graham CF, Mendelis PS, Stanley GR.
Discovery of adverse drug reactions. A comparison of selected phase IV studies with
spontaneous reporting methods. JAMA 1983;249:2226-8.
36. Amery W. Analysis of the information in a central ADE database. International Journal of
risk & safety in Medicine 1993;5:123.
20
Part 1
METHODOLOGICAL ASPECTS OF QUANTITATIVE SIGNAL

DETECTION
The first part of this thesis consists of two studies concerning methodological
aspects of quantitative signal detection. To focus attention of reviewers,
quantitative procedures can be used to sift data in different ways. In various
centres, however, different measures are used to quantify the extent to which an
association between clinical event and drug is reported disproportionally to the
generality of the database. The objective of the study in Chapter 1.1 is to examine
the level of concordance of the various estimates when applied to the dataset of
the Netherlands Pharmacovigilance Foundation Lareb.
In addition to the classical approach, where for each case it is assessed if the
reported association represents a new ADR, quantitative techniques are being
developed to facilitate the signal detection process. Like the classical approach,
these new techniques are also subject to some limitations. One of the most
noticeable problems is (selective) underreporting. For quantitative signal
detection on the dataset of Lareb, a Reporting Odds Ratio is used as measure of
disproportionality. Chapter 1.2 focuses on the influence of selective and non-
selective underreporting when using Reporting Odds Ratios in the detection of
associations between drugs and ADRs, drug-drug interactions and drug-related
syndromes.
1.1
A COMPARISON OF MEASURES OF DISPROPORTIONALITY

FOR SIGNAL DETECTION IN SPONTANEOUS REPORTING
SYSTEMS FOR ADVERSE DRUG REACTIONS
EP van Puijenbroek1, A Bate2,3, HGM Leufkens4, M Lindquist2, R Orre5 and

ACG Egberts4,6
1
Netherlands Pharmacovigilance Foundation Lareb, 's-Hertogenbosch, the Netherlands
2
The Uppsala Monitoring Centre, Uppsala, Sweden
3
Department of Clinical Pharmacology, Umeå University, Sweden
4
Pharmaceutical Sciences, Utrecht, the Netherlands
5
Department of Mathematics, Stockholm University, Sweden
6
Hospital Pharmacy Midden-Brabant, TweeSteden Hospital and St. Elisabeth Hospital,
Tilburg, the Netherlands
SUBMITTED FOR PUBLICATION

Chapter 1.1
Summary
Objective A continuous systematic review of all reported combinations of drugs

and suspected adverse reactions (ADRs) reported to a spontaneous reporting
system, is necessary to optimise signal detection. To focus attention of human
reviewers, quantitative procedures can be used to sift data in different ways. In
various centres, different measures are used to quantify the extent to which an
ADR is reported disproportionally to a certain drug compared to the generality of
the database. The objective of this study is to examine the level of concordance of
the various estimates to the measure used by the WHO Collaborating Centre for
International ADR monitoring, the information component (IC), when applied to
the dataset of the Netherlands Pharmacovigilance Foundation Lareb.
Methods The Reporting Odds Ratio-1.96 standard errors (se), Proportional
Reporting Ratio-1.96se, Yule’s Q-1.96se, the Poisson probability and Chi-square
test of all 17,330 combinations were compared with the IC minus 2 standard
deviations. Additionally, the concordance of the various tests, stratified according
to the number of reports per combination, was examined.
Results In general, sensitivity was high with respect to the reference measure
when a combination of point- and precision estimate was used. The concordance
increased dramatically when the number of reports per combination increased.
Conclusion This study shows that the different measures used are broadly
comparable when 4 or more cases per combination have been collected.
24
Comparison of measures of disproportionality
Introduction
When medicinal products are marketed, case reports of suspected adverse drug
reactions (ADRs) are reported to spontaneous reporting systems on a national
level. One task is to detect and investigate possible new side effects of these
drugs. All case reports are filed in databases at the National Centres as well as
sent to the WHO Collaborating Centre for International Drug Monitoring (the
Uppsala Monitoring Centre).1 Usually, trained assessors regularly examine every
incoming reported combination between a drug and a suspected ADR for possible
signals in a case-by-case analysis. A systematic continuous review of the
combinations present in the database is necessary to optimise the primary goal of
spontaneous reporting systems, i.e. monitoring for unexpected or unknown ADRs
or signal detection.2 The WHO defines a signal as: ‘Reported information on a
possible causal relationship between an adverse event and a drug, of which the
relationship is unknown or incompletely documented previously’. Often, a limited
number of reports represent a signal.3 Because of increasingly large numbers of
case reports being stored in databases, adequate signal detection without
automated quantitative screening is becoming time consuming and inefficient,
because of the sheer load of information to be assessed. In quantitative signal
detection, combinations of a drug and a clinical event that are disproportionately
high represented in the database, may represent an important signal based upon a
difference from the background frequency.4 Subsequently these combinations
must still be analysed and interpreted by the critical human mind. In contrast to
hypothesis testing where quantitative estimates are used to express the frequency
of a signal, in spontaneous reporting systems they are used to determine the
probability of a combination being a signal or not, based on disproportionate
reporting.
The use of a measure of disproportionality is currently applied in various national
spontaneous reporting centres as well as the Uppsala Monitoring Centre. Several
point estimates like the Reporting Odds Ratio (ROR), Proportional ADR
Reporting Ratio (PRR) or Yule’s Q, have been used, in combination with
additional estimators of the precision of point estimates such as the Chi square
test, or the lower limits of the 95% confidence intervals of the point estimates.5-9
Furthermore, the chance of the number of reports being reported on a certain
combination, under the assumption that no relationship exists between the
25
Chapter 1.1
reported suspected ADR and the suspected medication, can be calculated by

means of the Poisson probability.10,11 Another approach is the use of Bayesian
logic, specifying the relation between the prior and posterior probability before
and after linking data fields, and of adding new data to the database, currently
being used for example by the Uppsala Monitoring Centre in the Bayesian
Confidence Propagation Neural Network analysis (BCPNN).12-14 This relationship
is expressed as the ‘information component’ (IC). Also the FDA is using a
Bayesian approach.15
The aim of this study is to examine the concordance of the various estimates, and
to clarify the way they are related to each other, when applied to the dataset of
suspected ADRs reported to the Netherlands Pharmacovigilance Foundation
Lareb. Calculations of measures of disproportionality are primarily based upon a
two by two contingency table (Figure 1).
Reports with the event Reports with other events
Reports with the suspected drug a b

All other reports c d
Figure 1. Two by two contingency table (for corresponding formulas see the overview at
the end of this chapter)
Since all the measures of disproportionality are based on the same principles of
calculation using the 2x2 table, results should be closely concordant. On the other
hand it is important to know how the different methods perform in practice,
particularly at low numbers of reports of a particular combination.
There is no true ‘gold’ standard to compare methods, but the BCPNN has been
tested for performance for signal determination against standard literature sources
on a retrospective basis.13 In order to obtain maximum information out of the
comparison, it was decided to express the level of concordance of the other
methods with the IC-2sd in terms of sensitivity, specificity, positive predictive
value and negative predictive values, instead of using a measure of concordance
such as the kappa statistic. It is appreciated that the use of ‘sensitivity’ and
‘specificity’ under these circumstances can be misleading, and these terms should
be seen in a relative sense.
26
Methods
The dataset of the Netherlands Pharmacovigilance Foundation Lareb was used for
the analysis. Lareb maintains the spontaneous adverse drug reaction reporting
system in the Netherlands on behalf of the Dutch Medicines Evaluation Board.16
All reports received by January 1st 2000 were included in the analysis.
Based on the 2x2 table and with respect to the background frequency of
associations of drugs and suspected ADRs in the database, the following point
and precision estimates were calculated for all combinations: IC-2sd, ROR minus
1.96 standard error (se), PRR-1.96se, Yule’s Q-1.96se, Chi square (with Yates'
correction) and the Poisson probability. In the event the measures could not be
calculated for mathematical reasons, the missing combinations were excluded
from the analysis. For the corresponding formulas see the overview at the end of
this chapter. The concordance of the different measures of disproportionality were
compared with the results of the BCPNN analysis as reference measure.
The BCPNN analysis calculates IC and IC-2sd values for all drug-ADR
combinations in its routine use. Those combinations that have IC minus 2
standard deviation (sd) greater than zero are highlighted for review.12-14,17 In
spontaneous reporting, a fairly small number of reports may be sufficient to
generate a signal.18,19 Furthermore, for the use of the various approaches
described, a certain statistical distribution is assumed. Since it is not clear whether
these assumptions are always fulfilled, calculations might be less appropriate in
the event of small numbers. For both reasons, the concordance of the various tests
with the results of the BCPNN as the number of reports per combination varied,
was also studied. In this respect, separate calculations were made for those
situations in which number of reports per combination was greater or equal to
2,3,4 or 6. Only combinations with a minimum amount of 2 reports were selected,
since the use of a measure of disproportionality was not considered useful in the
event that only one report has been received, even though measures of
disproportionality can be calculated in the latter situation.
For ROR-1.96se, PRR-1.96se, Yule’s Q-1.96se, Chi square (with Yates'
correction) and the use of the Poisson probability, sensitivity, specificity, positive
predictive value and negative predictive value as well as the percentage of cases
in which the various measures could be calculated were determined with IC-2sd
as the reference measure. A positive association was defined as IC-2sd>0 and
27
Chapter 1.1
negative as IC-2sd<0. For calculating point and precision estimates, including the
IC and IC-2sd and the indicators of the concordance, Microsoft Excel 97 was
used.
Results
On January 1st 2000, 26,555 reports were filed in the database of the Netherlands
Pharmacovigilance Foundation Lareb. These reports involved a total number of
39,790 reported suspected adverse drug reactions (ADRs) which concerned
17,330 different combinations between a drug and a suspected ADR. The number
of combinations with one, two, three and four or more reports was 11,856, 2,455,
1,072 and 1,947 respectively. The mean number of reports per combination was
2.3. The results of the comparisons for different numbers of reports per
combination are presented in Table 1.
In the majority of cases a combination of point and precision estimate could be
calculated. In the event 2 or 3 cases per combination were reported, calculating
the ROR, PRR and ROR-1.96se was not possible in about 1-5% of the cases. In
general sensitivity was high but the specificity was rather low for all measures
applied. In all situations the concordance of combination of point estimate and
precision estimate increased dramatically in case the number of reports per
combination increases. In the event that 4 or more reports were received on a
combination, all approaches gave comparable results.
Discussion
In comparison with IC-2sd, the various methods were highly sensitive but had a
rather low specificity. Only when the Poisson probability and Chi square were
used, all combinations could be analysed. Nevertheless, in the dataset used in this
study, for instance the ROR-1.96se could be calculated for 99.95% of all the
drug-ADR combinations reported 3 or more times.
When 4 or more reports per combination were present, no clear differences were
found between the use of the various measures, and the use of the IC-2sd.
Although the percentage of combinations in the Lareb dataset with 4 or more
reports were received is rather low (11.2%), this subset is of particular interest for
signal detection.
28
Table 1. Sensitivity, specificity, positive predictive value and negative predictive value
concerning the use of various point estimates and tests in comparison with IC-2sd,
regarding different numbers of reports per combination. Also the percentage of
combinations for which a point estimate could be calculated, is provided.
Test Number Sensitivity Specificity Positive Negative Percentage
of reports predictive predictive calculated
value value
ROR-1.96se>1 a³2 1.00 0.66 0.41 1.00 94.8

a³3 1.00 0.77 0.70 1.00 99.9
a³4 1.00 0.84 0.83 1.00 100.0
a³6 1.00 0.89 0.91 1.00 100.0
PRR-1.96se>1 a³2 1.00 0.70 0.44 1.00 94.8

a³3 1.00 0.81 0.73 1.00 99.9
a³4 1.00 0.86 0.85 1.00 99.9
a³6 1.00 0.88 0.90 1.00 100.0
Yule’s Q - a³2 1.00 0.60 0.38 1.00 99.5

1.96se>1
a³3 1.00 0.73 0.67 1.00 99.9
a³4 1.00 0.81 0.80 1.00 99.9
a³6 1.00 0.86 0.89 1.00 100.0
Chi square a³2 1.00 0.71 0.46 1.00 100.0

(Yates' corr) a³3 1.00 0.80 0.73 1.00 100.0
(p<0.05) a³4 1.00 0.83 0.82 1.00 100.0
a³6 1.00 0.85 0.88 1.00 100.0
Poisson a³2 1.00 0.53 0.34 1.00 100.0

(p<0.05) a³3 1.00 0.66 0.61 1.00 100.0
a³4 1.00 0.74 0.75 1.00 100.0
a³6 1.00 0.79 0.84 1.00 100.0
For drug-ADR combinations in the dataset which are listed on less than four
reports, results of the various analyses show some differences. For all measures
sensitivity is still high with respect to IC-2sd, but specificity rapidly declines for
such a low number of reports. This implies that all combinations highlighted as
potential signals by use of IC-2sd are also highlighted by the other measures
under investigation, whereas not all combinations highlighted by the other
29
Chapter 1.1
measures have a positive IC-2sd. So, either the number of false positive signals
increases for combinations of less than four reports, for each of the measures
compared to the IC-2sd; or the potential positives highlighted by the other
measures are in fact true positives which the BCPNN might go on to highlight
later, as more information accumulates. Further detailed investigation is needed to
attempt to determine which of these scenarios is most likely. This evaluation is
made harder due to the lack of a true gold standard for discrimination of true and
false signals. Thus careful ongoing evaluation of how these potential signals
develop over time may be the most appropriate method of investigation. When the
number of combinations for which a measure could be calculated was not 100%,
the missing combinations were excluded from the analysis. These combinations
should also be checked by other techniques to determine if they might represent
true signals.
For all tests, results are more comparable with IC-2sd when the number of reports
per combination increases. Although not presented here, other measures such as
the Poisson probability were also studied as the reference measure, but whichever
method was used for comparison, poor concordance was found at low counter
values. This lack of concordance may be explained by the fact that for a small
number of reports the assumed type of distribution of the various classical
methods (e.g. Gaussian or Poisson distribution) will have a strong influence. For
the tests used, some basic assumptions should be applied. For instance,
concerning the use of Chi square, on tables with more than a single degree of
freedom, a minimum expected frequency of 5 can be regarded as adequate. If
there is only one degree of freedom (which is the case in our two by two
contingency table), a minimum expected frequency of 10 is much safer.20 When
the expected numbers are small, but greater than 5, another option is to apply
continuity correction (Yates' correction).21 Even so the results of the test should
be interpreted cautiously. In general only a small number of reports per
combination is necessary to trigger a signal. In this situation, calculation of the
Poisson probability therefore is safer, although when using a Bayesian
implementation, the IC can also be calculated for small numbers. Calculating the
confidence interval of the odds ratio and Yule’s Q is also subject to limitations.
For small numbers of reports the distribution may be skewed, and calculations
based upon a Gaussian distribution cannot be applied without caution.
30
The prior assumption for the Bayesian derivation of the IC is of independence

between the drug-ADR. This causes a dampening effect of the IC at very low
numbers of cases, making positive ICs less positive and negative ICs less negative
than might be envisaged looking purely at the proportion of expected/observed
cases.
These differences between the quantitative measures serve to reiterate the crucial
importance of clinical and pharmaceutical information, as well as other data in
signal detection.3 In the event of small numbers of cases other aspects contribute
increasingly in the selection of signals, such as the clinical information available
and the level of documentation of the reports as the potential signal can be
analysed on a case-by-case basis. Any combination selected in spontaneous
reporting databases using purely statistical methods, should be carefully evaluated
and confirmed by other means before making any kind of regulatory decision.22
Additional analysis of the signals, i.e. by evaluation of the original reports, is
therefore warranted.23-25
Methodological considerations
Although in quantitative signal detection no true gold standard is available, we
have chosen the information component of the WHO as being the reference
measure for the following reasons. Firstly, when used for analysing the WHO
database, this approach yielded a positive predictive value of 44% and a negative
predictive value of 85% in the detection of signals as compared with reference
literature sources.13 In this study the authors discuss the difficulty of defining a
gold standard in signal detection. We considered that the availability of the
performance information in that study, albeit subject to limitations, would allow
the results of the current study to be placed in a useful context. Secondly, in
contrast to other measures, both point estimate (IC) and its probability interval
can be calculated under all circumstances. The confidence intervals for the ROR,
PRR, Yule’s Q were calculated as the standard error, but in the Bayesian
approach, the standard deviation was calculated from the IC distribution.
Furthermore, for calculating the lower limit of the confidence interval IC-2sd
instead of IC-1.96sd has been used since it is routinely implemented in this way,
and was used for the retrospective evaluation study outlined above.
An alternative method of comparison could have been used such as the kappa
statistic, but a drawback is that this does not distinguish between a situation of
31
Chapter 1.1
high sensitivity and low specificity, and one of poor sensitivity and high
specificity.
In the Poisson probability or the Chi-square test only the chance that the observed
frequency differs from the expected frequency is provided. This situation differs
from the other tests like Yule’s Q-1.96se and ROR-1.96se and PRR-1.96se where
the prior assumption is made that combinations we are looking for, occur more
often than the expected frequency. If we wanted to look for combinations that
occur less frequently in combination with a certain drug we should use for
instance ROR+1.96se. In our analysis we did not take these differences into
account.
Choice of a measure of disproportionality

In this study the level of concordance between the different measures used in
quantitative signal detection have been examined. The different implementations
of these measures have not been considered, as these are necessarily dependent on
other factors such as the data set used. For example the MCA use a PRR>3 and
Chi square>4 and 3 or more cases as a filter for signal detection.26 The choice of a
suitable method, therefore, also depends on the dataset available. In Table 2 the
conditions, advantages and disadvantages in which the various tests can be used
are provided.
Apart from sensitivity, specificity, positive and negative predictive value, the
possibility for correcting for covariates can be useful. For this reason, the
Netherlands Pharmacovigilance Foundation Lareb presently uses the ROR-1.96 se
to calculate disproportionality, since in logistic regression analysis these
adjustments can easily be made. Similarly the BCPNN approach can be adapted
to adjust for covariates. A major drawback with the ROR, however, is the fact that
in case an ADR is specifically associated with a certain drug, there is a risk that
the number of reports in cell b or c of the contingency table (Figure 1) contains no
reports and subsequently the odds ratio cannot be calculated. This situation may
occur when rare ADRs are being reported. An example of this type of
combination is for instance the detection of a phocomelia associated with
thalidomide, or practolol associated with the oculomucocutaneous syndrome.
32
Table 2. Conditions, advantages and disadvantages of different measures of

disproportionality
Measure Expected Conditions Advantage Disadvantage
‘null value’
ROR-1.96se 1 Cells a,b,c and -Easy applicable -Odds ratio and

d have to -Different standard error can
contain reports adjustments possible not be calculated if
in logistic regression denominator zero
analysis (specific ADRs)
-In logistic regr. -Interpretation
analysis, interaction difficult
terms can be used for -Results not always
the analysis of drug reliable in the event
interactions and of small numbers in
syndromes cells a,b,c and d of
the contingency table
PRR-1.96se 1 Cells a and c -Easy interpretation Standard error cannot

have to contain always be calculated
reports
Chi square Always applicable Difficult to interpret

(Yates' corr.)
Yule’s Q-1.96se 0 cells a,b,c and -Standard error

d have to cannot always be
contain reports calculated
-Difficult to interpret
Poisson Only for rare Correction for Only p-value

events different covariates provided
can be easily
established in
Poisson regression
IC-2sd 0 none -Always applicable Relatively non

-Large numbers of transparent for
calculations can be people not familiar
made efficiently with Bayesian
-Can be used for statistics
pattern recognition in
higher dimensions
33
Chapter 1.1
The ROR is a transparent measure, easily interpretable which can be easily

programmed in database programmes or spreadsheet programmes. An additional
advantage of using the odds ratio is the fact that non-selective underreporting of a
drug or adverse drug reaction has no influence on the value of the ROR compared
with the population of patients experiencing an ADR.5
Disproportionality is simply one way of selecting drug-ADR combinations that
may be interesting for clinical review. No individual approach to detect signals is
adequate and the concurrent use of other methods is therefore essential.
Conclusion
Statistical analyses have been shown to be useful tools in aiding signal detection
in spontaneous reporting systems. The various measures that are being applied in
quantitative signal detection in various national centres, are comparable when
more than 4 or more reports constitute the drug-ADR combination. The
heterogeneity of the data collected in databases of spontaneous reporting systems
and the variety of biases influencing data (such as underreporting) are likely to
have more influence on the potential for signal detection than the small
behavioural differences between the measures detected in this study. Although no
‘gold standard’ is available, each method has its own advantages and
disadvantages with respect to applicability in different situations and possibilities
for implementation. Since quantitative signal detection cannot take into account
clinical aspects, a case-by-case approach will remain necessary both as an adjunct
and an alternative.
Acknowledgement
The authors gratefully acknowledge Prof IR Edwards for his valuable comments.
34
Overview of measures of disproportionality
Variables used in the different formulas correspond to the 2x2 contingency table
of Figure 1
Reporting Odds Ratio (ROR)
The ROR can be expressed as5
ROR =
(a / c) = ad
(b / d ) bc
The standard error of ln(ROR) and 95 % confidence interval can be calculated by
se(lnROR) = (a1 + b1 + 1c + d1 ) 95%CI = eln(ROR)±1.96

æ1 1 1 1ö
ç + + + ÷
èa b c dø
Proportional ADR Reporting Ratio (PRR)
The PRR can be expressed as
a / (a + b )
PRR =
c / (c + d )
The standard error of ln(PRR) and 95 % confidence interval can be calculated

by27
se(lnPRR) = (a1 - a+1b + 1c - c+1d ) 95 % CI = e ln( PRR ) ±1 .96

æ1
ç -
1 1
+ -
1 ö
÷
è a a +b c c+ d ø
Chi square (Yates' correction)
Chi square tests for a 2 by 2 table, with Yates' correction can be expressed as
2
c = å
(O - E ) 2
- 1
2
35
Chapter 1.1
The summation applies over all four cells of the contingency table. O is the
observed frequency and E is the expected frequency of the reports. For example,
in case the contingency table is used for the first cell O and E should be calculated
as:
O=a E=
(a + b )(a + c )
(a + b + c + d )
For the other cells it takes the value contained in the cell, i.e. b, c, d in turn
Yule’s Q
Yule’s Q can be expressed as28
ad - bc
Q=
ad + bc
The standard error of Yule’s Q and the 95% CI is calculated by
æ1 1 1 1ö
= 12 æç1 - Q ö÷ ç + + + ÷
2
se Q
è ø èa b c d ø
95%CI = Q ± 1.96 SE Q
Poisson probability
The Poisson probability is calculated by10
-m k
a -1 e ´m
p = 1- å
k =0 k!
where m is the expected number of reports:
m=
(a + b )(a + c )
(a + b + c + d )
36
Information component, resulting from the BCPNN and its variance can be
calculated as 17
(cij + g ij )(C + a )(C + b )

E ( ICij ) = log 2
(C + g )( ci + a i )( c j + b j )
C - cij + g - g ij C - ci + a - a i C - c j + b - bi
+ +
(cij + g ij )(1 + C + g ) (ci + a i )(1 + C + a ) (ci + b j )(1 + C + b )
V ( ICij ) =
(log 2) 2
where
(C + a ) (C + b )
g = g ij ×
(ci + a i ) (c j + b j )
and γij=1, αi=1, α=2, βj=1, β=2, C is the total number of reports in the database,
Cij the number of combinations between a specific drug [i] and the suspected
adverse drug reaction [j], Ci the total number of reports on drugs [i] in the
database and Cj the total number of reports on the suspected ADR [j] in the
database.
References
1. Olsson S. The role of the WHO programme on International Drug Monitoring in
coordinating worldwide drug safety efforts. Drug Saf 1998;19:1-10.
2. Finney DJ. The design and logic of a monitor of drug use. J Chron Dis 1965;18:77-98.
3. Edwards IR, Lindquist M, Wiholm B-E, Napke E. Quality criteria for early signals of
possible adverse drug reactions. Lancet 1990;336:156-8.
4. Finney DJ. Statisical logic in the monitoring of reactions to therapeutic drugs. Methods Inf
Med 1971;10:237-45.
1992;45:1177-84.
6. Evans SJW, Waller P, Davis S. Proportional Reporting Ratios: The uses of epidemiological
methods for signal generation. Pharmacoepidemiol Drug Saf 1998;7(Suppl 2):S102.
7. Egberts ACG, Van der Hofstede JW, Meyboom RHB, de Koning GHP, Bakker A, Leufkens
HGM. Transformation of a database of spontaneously reported suspected adverse drug
37
Chapter 1.1
reactions and its use as a tool in signal detection. In: Egberts ACG. Pharmacoepidemiologic
approaches to the evaluation of antidepressant drugs. (Thesis) Utrecht University, Utrecht,
the Netherlands, 1997:111-24.
8. Evans SJW. Pharmacovigilance: a science or fielding emergencies? Stat Med 2000;
19:3199-209.
1997;44:513-8.
10. Tubert P, Begaud B, Pere JC, Haramburu F, Lellouch J. Power and weakness of spontaneous
reporting: a probabilistic approach. J Clin Epidemiol 1992;45:283-6.
11. The Poisson distribution. In: Hayes WL. Statistics. Fort Worth: Holt, Rinehart and Winston,
inc, 1988:144-6.
1998;54:315-21.
13. Lindquist M, Ståhl M, Bate A, Edwards IR, Fucik H, and Nunes AM. From association to
alert -a revised approach to International signal analysis. Pharmacoepidemiol Drug Saf
1999;8:S15-S25.
15. DuMouchel W. Bayesian data mining in large frequency tables, with an application to the
FDA spontaneous reporting system. The American Statistician 1999;53:177-89.
17. Orre R, Lansner A, Bate A, Lindquist M. Bayesian Neural Networks with confidence
93.
18. Waller P and Lee EH. Responding to drug safety issues. Pharmacoepidemiol Drug Saf
1999;8:535-52.
19. Meyboom RHB, Egberts ACG, Edwards IR, Hekster YA, de Koning GHP, and Gribnau FW.
20. The assumptions in chi-square tests of association In: Hayes WL. Statistics. Fort Worth:
Holt, Rinehart and Winston inc, 1988:780-1.
21. Tubert-Bitter P, Bégaud B. Comparing safety of drugs. Post Marketing Surveillance
1993;7:119-37.
22. Begaud B, Moride Y, Tubert-Bitter P, Chaslerie A, Haramburu F. False-positives in
spontaneous reporting: should we worry about them? Br J Clin Pharmacol 1994;38:401-4.
23. Amery WK. Assessment of signals generated form spontaneously reported adverse events.
Pharmacoepidemiol Drug Saf 1999;8:301-4.
24. Meyboom RH, Egberts AC, Gribnau FW, Hekster YA. Pharmacovigilance in perspective.
Drug Saf 1999;21:429-47.
38
25. Amery WK. Signal generation from spontaneous event reports. Pharmacoepidemiol Drug
Saf 1999;8:147-50.
26. Evans SJW. Recognition of signals from spontaneous reporting systems - Comparison of
different approaches: Poisson model, neural networks, Proportional Reporting Ratios.
Proceedings of 20th Annual Conference of the International Society for Clinical Biostatistics
1999;119(abstract).
27. Introduction to Categorial Statistics In: Greenland S, Rothman KJ. Greenland S, Rothman
KJ (eds). Modern Epidemiology. Philadelphia: Lippincott-Raven, 2001:231-52.
28. Griffin C. The Advanced Theory of Statistics. London: 1995.
39
1.2
ON THE ASSESSMENT OF ADVERSE DRUG REACTIONS

FROM SPONTANEOUS REPORTING SYSTEMS. THE
INFLUENCE OF UNDERREPORTING ON ODDS RATIOS
PGM van der Heijden1,2, EP van Puijenbroek3, S van Buuren1

and JW van der Hofstede1
1
TNO Prevention and Health, Leiden, the Netherlands
2
Utrecht University, Department of Methodology and Statistics, Utrecht, the Netherlands
3
Netherlands Pharmacovigilance Foundation Lareb, ′s-Hertogenbosch, the Netherlands
SUBMITTED FOR PUBLICATION

Chapter 1.2
Summary
Objective A well-known problem in spontaneous reporting systems (SRSs) for

Adverse Drug Reactions (ADRs) is underreporting, i.e. the problem that not all
occurrences of ADRs are reported to the SRS. We look at the question of how to
draw statistical conclusions from analyses of SRS data using Reporting Odds
Ratios.
Methods We will show that certain underreporting problems play no role in
assessing ADRs from SRSs: the results from the analyses turn out to be biased by
some specific underreporting problems, but not by others. The influence of
underreporting on the Reporting Odds Ratio is illustrated by two examples.
Results If the assumption holds that there is an underreporting problem for a first
drug, and an underreporting problem for a second drug, but that these two
underreporting problems do not influence each other, then reporting odds ratios
estimated from SRSs are useful for signalling drug-drug interactions in the ADR
experiencing population. Similar results hold for covariate-drug interactions.
Conclusion In contrast to general underreporting effects, paired underreporting
effects result in biases in the odds ratios estimated from the SRS. If there is no so-
called joint underreporting effect of two drugs on the ADR, then SRS data can be
useful for assessing the presence of drug-drug interactions.
42
Influence of underreporting on odds ratios
Introduction
Before marketing a new drug, many adverse drug reactions (ADRs) may either be
suspected from chemical similarity to known drugs or detected in clinical trials. In
these trials, drugs are used in a selected, rather small population.1 Detection of
ADRs in clinical trials is hampered by the fact that rare ADRs and ADRs with a
long time to onset are difficult to detect. Since trials are carried out under
controlled circumstances, the detection of ADRs in specific populations, like the
elderly, patients with chronic diseases or patients with multiple drug use are even
more difficult to detect. Spontaneous reporting systems (SRSs) are commonly
used to detect new or unexpected ADRs after the marketing of drugs. Because of
methodological reasons, such as selective underreporting, SRSs can only be used
to signal the possible existence of new or unexpected ADRs. Further (pharmaco-
epidemiological) studies are needed to evaluate these ADRs in more detail.2-4
A special case is the detection of drug-drug interactions. Since in clinical trials
drugs are used in a specific population, and multiple drug use is often a criterion
for exclusion, the detection of drug-drug interactions is more difficult. Until now,
a drug-drug interaction is usually detected when it is suspected by a physician or
pharmacist, and subsequently reported to an SRS. In daily practice, an interaction
between two different drugs is often detected by the occurrence of an ADR. When
a patient who already uses one or more drugs is administered another drug,
however, it is not always clear if the ADR is caused by the new drug, the drug-
drug interaction, or by some other cause. Based on the idea that in the event of
drug-drug interactions the chance for an ADR to occur is increased, SRSs can be
used to detect such drug-drug interactions.5-7 It is essential in this respect to
compare four different situations: reports of patients who, among all other
medication, use both drugs suspected of causing a possible drug-drug interaction,
the two situations where only one of the suspected drugs is used, and finally the
control situation where neither drug is used.
The amount of absolute underreporting can be quite large. By comparing the
’true’ ADR rate and the reported ADR rate, sometimes only 1 in 70 ADRs were
reported.8-10 Much recent statistical work on the analysis of SRS data.11-13
acknowledges the additional complexities created by underreporting, but at the
same time tends to ignore these complexities in the quantitative analysis. The
present paper gives conditions under which this strategy is appropriate.
43
Chapter 1.2
In this paper we discuss aspects of the statistical study of SRS data, with special
attention to the underreporting problem. We will show that there are different
types of underreporting problems, and that not all underreporting problems are
bothersome in the analysis of SRS data. For the underreporting problems that can
be problematic, we discuss what can be concluded under what type of
assumptions. In section 2 we further motivate the problem by giving three typical
examples and do a typical analysis of the SRS data. The question that remains to
be answered is what can be concluded from the analysis in the light of
underreporting problems. For each of the typical examples the answer will then be
given in a separate section, namely sections 3 to 5. We summarize and discuss our
results in section 6.
2. Motivating examples
Suppose we are interested in the question of whether some specific drug leads to a
specific adverse drug reaction (ADR). In the population that takes the drug there
is a group of patients who experience this ADR, and there is a group of patients
who do not experience this ADR. We are interested in whether the drug is
associated with the specific ADR. How can spontaneous reporting systems (SRS)
help us to answer this question?
It is clear that usually only some of the patients who take the specific drug and
experience the specific ADR report to the SRS. This is called the underreporting
problem. This underreporting problem can be more or less severe according to the
seriousness of the ADR in relation to the indication for use, the clarity of the
causal relation between the drug and the ADR, or due to other possible
reasons.8,10,14,15
SRSs are known for their signalling function. Because of various confounders and
underreporting, a causal relationship cannot be determined. Further studies under
controlled circumstances are necessary to demonstrate any possible association.
Apart from qualitative aspects of the reports, the number of reports concerning a
possible association between a drug and an ADR may also indicate the presence
of a true relationship. The number of reports necessary to generate a signal,
however, depends on the total number of reports in the database, the total number
of ADRs in the database and the number of reports concerning the association.
44
In SRSs this problem is often tackled by comparing the different profiles of ADRs
reported. Usually a cross-classification of all reports available from the SRS is
constructed, i.e. not only reports involving both the specific ADR and the specific
drug, but also reports involving other ADRs and other drugs. Based on this cross-
classification for instance a Reporting Odds Ratio can be calculated.16,17 Another
option would be applying a ’Proportional Reporting Ratio’, based on the same
2x2 contingency table. This Proportional Reporting Ratio can be considered as a
representation of the safety profile of a drug.18 In other words, other reports in the
database provide a proxy of the ’background incidence’ of the ADRs. Indeed the
underreporting factors cannot account properly for the size of the exposed
population, but regarding the population of patients who actually experience an
ADR, the use of the Reporting Odds Ratio may provide a valid estimate. The
Reporting Odds Ratio offers advantages in the sense that in a logistic model
adjustments for various confounders can be made and statistical interactions
between various covariances can be analyzed in more detail.15-17,19
The Netherlands Pharmacovigilance Foundation Lareb collects and analyzes
reports of suspected adverse drug reactions from health professionals in the
Netherlands. Every report is assessed on a regular basis. However, due to the
increasing amount of reports, analysis by the human mind alone becomes more
difficult and statistical analysis of the data may be helpful. Moreover, a statistical
approach enables the identification of more complex relationships like drug-drug
interactions and the analysis of syndromes.6,7 These reports have a ’spontaneous’
character, which implies that reporting of the suspected ADR by a physician or
pharmacist is not compulsory. For this reason, underreporting is inherent in this
approach and signals of possible ADRs from SRSs should be considered in this
perspective.14,20 For every possible association present in the database, the
influence of underreporting on various factors should be carefully weighted. We
discuss three examples using data provided by Lareb.
2.1 Diuretics and possible ADRs

Assume that we would like to investigate whether there is an association between
the presence of diuretic drugs on the report forms and ADRs possibly
representing the presence of congestive heart failure (see Table 1).
45
Chapter 1.2
Table 1. Presence of a diuretic drug, and signs of congestive heart failure as an ADR.
Observed frequencies
Diuretic drug Congestive heart failure
absent present
absent 7,820 227

present 1,697 78
The suspected ADRs of reports to Lareb are coded by means of the WHO adverse
drug reaction terminology,21 drugs are coded according the ATC terminology. A
report may be used to report one or more suspected drugs and one or more
suspected adverse drug reactions. Between January 1st 1990 and January 1st
1999, a total number of 9,822 reports concerning patients older than 50 year were
received by Lareb, of which sex of the patient involved and the type of reporting
health professional (physician of pharmacist) were known.7 A selection was made
of WHO-preferred terms that might indicate the presence of signs of congestive
heart failure. Cases were defined as reports in which one of the following WHO
preferred terms were present: oedema, oedema dependent, oedema generalized,
oedema peripheral, cardiac failure, cardiac failure left, cardiac failure right, or
oedema legs. Non-cases were defined as all other reports. Exposure categories
were the presence of diuretics among the medication used (ATC code beginning
with C03) versus no diuretics. A typical way to analyze the data is to fit a model
in which the probability of signs of congestive heart failure in the reports where
diuretic drugs are present, is identical to this probability in the reports where
diuretics are absent. This is equivalent to fitting the independence model. This
model is rejected (likelihood ratio chi-square is 10.9, df is 1, p<0.001). The
observed odds ratio is 1.58 (95 percent confidence interval is 1.21-2.05), showing
that the use of diuretics and signs of possible congestive heart failure are related
in the SRS data, which is not surprising, given the fact that diuretics are
commonly used in the treatment of congestive heart failure. The reports
concerned therefore might either represent the background incidence of oedema
or congestive heart failure, or a lack of efficacy of the drugs concerned. Since we
are interested in the population of patients that actually uses the drugs in question,
we subsequently want to evaluate whether this relation in the SRS data is a close
representation of the relation in the ADR experiencing population? For this we
46
need a better understanding of the underreporting problem, and we will elaborate

this problem in section 3.
2.2 Drug-drug interaction of diuretics and NSAIDs
Besides generating signals concerning ADRs, databases of SRSs may also be
used to generate signals of possible drug-drug interactions or identifying risk
factors in patients. Consider the following example. Several case-reports and
studies suggest that concomitant use of diuretics and Non-Steroidal Anti-
Inflammatory Drugs (NSAID, ATC code beginning with M01A) can lead to an
increased risk of developing signs of congestive heart failure,22,23 due to a
decreased efficacy of the diuretics involved. For the data in Table 1 we now
would like to investigate whether there is an indication of a drug-drug interaction
of diuretics and NSAIDs with signs associated with the presence of congestive
heart failure.7 Exposure categories were the use of NSAIDs or diuretics versus the
use of neither of these drugs (see Table 2).
Table 2. Two drugs: Presence of a diuretic drug and presence of an NSAID. Congestive
heart failure as ADR. Observed frequencies
NSAID Diuretic drug Congestive heart failure
absent present
absent absent 6,527 185

present 1,444 53
present absent 1,293 42
present 253 25
A typical analysis would again involve the use of odds ratios. The odds ratio for
diuretics and congestive heart failure without NSAID prescription is 1.29,
whereas this odds ratio with NSAID prescription is 3.04. In a loglinear model
with no three-factor interaction these two odds ratios would be restricted to be
equal, but such a model fits poorly (LR chi-square is 7.53, df is 1, p<0.01),
showing that these two observed odds ratios differ significantly. We conclude
that, in the SRS data, the concomitant use of NSAIDs and diuretics leads to an
increase in signs of congestive heart failure. But what can be said about the ADR
experiencing population? This problem will be addressed in section 4.
47
Chapter 1.2
2.3 Sex differences for ADRs of diclofenac

Controlling for covariates is illustrated by a last example. Acute allergic reactions
are associated with various NSAIDs including diclofenac.24 Although these ADRs
are mentioned in the Dutch Summary of Product characteristics, Lareb received a
substantial number of reports on anaphylactoid reactions or anaphylactic shock
associated with diclofenac. We analysed this association using a case-control
design, controlling for sex. All 17,399 reports received by Lareb between 1990
and 1999 of patients older than 10 years were included.
All reports were coded using the WHO adverse drug terminology.21 Cases were
defined as all reports coded with ‘anaphylactic shock’ or ‘anaphylactoid reaction’.
All other reports were considered as non-cases. See Table 3.
Table 3. Diclofenac as drug, anaphylactic reactions as ADR, and sex as covariate.

Observed frequencies
Sex Diclofenac Anaphylactic reactions
absent present
male absent 6,108 30

present 181 11
female absent 10,658 61
present 331 19
The observed odds ratio for males is 12.4 (95% CI 8.7-17.8) and for females 10.0
(95% CI 7.6-13.1). These results show that, for males as well as females,
anaphylactic reactions are disproportionally reported on diclofenac as compared
to other NSAIDs, which would suggest an increased risk of these reactions during
the use of diclofenac.
By fitting the loglinear model without interaction term for sex by drug by ADR,
we investigate whether the two observed odds ratios differ significantly. The
likelihood ratio chi-square is 0.22 (df is 1, n.s.) showing that the data do not
provide evidence for a difference. The estimated odds ratio estimated in this way
is 10.8. But in what way might underreporting influence these results? This will
be discussed in section 5.
48
3 The underreporting problem for one drug
3.1 Theory
We will now discuss the question raised at the end of section 2.1: in what way
does underreporting distort the relation between estimates from the SRS data and
the ADR experiencing population? Table 4 will help to answer this problem.
Table 4. The underreporting problem in data from a spontaneous reporting system

Specific drug Specific ADR
absent present
absent n11 n12

present n21 n22
Assume a variable ’specific ADR’ (for example, congestive heart failure) having
levels ’present’ (in report) and ’absent’; assume further a variable ’specific drug’
(for example, diuretics) having levels ’present’ (in report) and ’absent’. Table 4
shows the observed counts in the SRS population, denoted by nik, where i=1,2
indexes the levels of the specific drug, and k=1,2 indexes the levels of the specific
ADR. Naturally, the number n11 is in most cases much greater than the other three
numbers, since this embraces all the reports where neither the specific drug nor
the specific ADR play a role (compare Table 1). If the specific ADR is more
common with the specific drug than with other drugs, one would expect n22/(
n21+n22) to be larger than n12/( n11+ n12). When looking for ADRs in the SRS
population, a relative high count of an ADR reported for a certain drug is used as
a signal for a more detailed study. However, different forms of underreporting
have effects on these numbers. Therefore we want to know in what way the
elements in Table 4, showing the information in the SRS population, provide us
with information of the ADRs in the ADR experiencing population. It is clear
that, since we have four frequencies in Table 4, each of the frequencies is plagued
by underreporting. Thus we could say that there are four underreporting problems,
one for each frequency. However, it is more insightful to approach these four
underreporting problems differently, namely by distinguishing between four
separate processes that cause these four frequencies to be underreported. We
distinguish four types of problems of underreporting:
49
Chapter 1.2
(i) there is an overall underreporting problem, which pertains to all cells of Table
4.
(ii) there is an underreporting problem for the specific drug compared to the other
drugs, in the sense that the overall factor in (i) does not show that for some drugs
the underreporting problem is more severe than for others. For instance: recently
introduced drugs are more likely to be reported. Media attention might also
increase the reporting rate of a specific drug.
(iii) there is an underreporting problem for the specific ADR compared to the
other ADRs, in the sense that the overall factor in (i) does not show that some
ADRs are more often reported than others. For instance: death due to a serious
allergic reaction is more likely to be reported than mild gastro-intestinal side
effects.
(iv) there is an underreporting problem for each combination of levels of the
specific ADR and the specific drug that describes a deviation from the overall
effect (i), the specific drug effect (ii) and the specific ADR effect (iii). For
instance, hair loss caused by chemotherapeutic drugs is a commonly occurring
ADR, that is rarely reported. The ADR is more easily accepted by physicians and
patients because of the necessity for the use of these drugs. Hair loss is a
relatively rare ADR of terbinafine.18 Since this antifungal drug is used frequently
for onychomycosis, which is usually a cosmetic indication, the occurrence of hair
loss is reported relatively frequently to Lareb. Under (iv) therefore it is the
severity of the ADR compared to the severity of the indication that seems to play
a role.
We will now explore this further using appropriate notation. We distinguish the
ADR experiencing population, for which we will denote the true frequencies by
tik, from the SRS population, for which we will denote the expected frequencies
by mik. Thus the observed frequencies nik are realisations of the expected
frequencies mik from the SRS population, but not from the ADR experiencing
population. The problem that we study in this paper is that we would like to make
statements about the ADR experiencing population on the basis of the SRS
population. The question is: under what assumptions are we allowed doing this?
We assume that the true frequencies tik from the ADR experiencing population are
related to the expected frequencies mik from the SRS population by the four
underreporting problems (i) to (iv). There are different ways to work out this
50
relation in mathematical terms. We use deviation coding, which leads to the

following notation:
- the general effect (i) will be denoted by an overall underreporting factor c that is
identical for each of the four frequencies. That is, when we go from the ADR
experiencing population tik to the SRS population with elements mik, each element
tik has to be multiplied with c;
- the drug effect (ii) will be denoted by cd if the specific drug is present and 1/cd if
not (note that the product of cd and 1/cd equals one);
- the ADR effect (iii) will be denoted by ca if the specific ADR is present and 1/ca
if not;
- the effect for the combination (iv) will be denoted by cda if both the specific drug
and ADR are present. In order to let the product of combination effects over the
rows and over the columns equal one, the effect for ADR present and drug absent,
and ADR absent and drug present, will both be 1/cda, and the effect for both ADR
and drug absent by cda.
In Table 5 the expected frequencies mik of the SRS population are expressed in
terms of the true frequencies tik of the ADR experiencing population taking the
four underreporting problems into account. Table 5 shows that the expected
frequency m22 in the SRS population is related to the expected frequency t22 by
m22=ccdcacdat22.
Table 5. One drug. Expected frequencies mik for the SRS population, expressed in terms
of true frequencies tik used for the ADR-experiencing population
Specific drug Specific ADR
absent present
absent ccda cca

t11 t12
cd ca cd cda
present
ccd
t 21 ccd ca cda t 22
ca cda
Using the data from an SRS we are unable to estimate the constants c, cd, ca and
cda to derive the true frequencies of the ADR experiencing population. We will
now look at the way these unknown constants bother us when we want to
51
Chapter 1.2
determine from SRS data whether the specific drug causes the specific ADR in
the ADR experiencing population.
Let us assume that we want to study a possible relation between the specific drug
and the specific ADR by estimating the odds ratio from observed counts derived
from an SRS. In what way would this estimate provide a biased account of the
odds ratio in the ADR experiencing population? The odds ratio θ t for the ADR
experiencing population is
t t11 t 22
q = (1)
t12 t 21
and when we derive the odds ratio θ for the SRS population by using the relations
provided in Table 5, many of the constants vanish:
m11 m22 = ( )4 t11t 22 = ( )4 t (2)

q = cda c da q
m12 m21 t12 t 21
This shows that estimate θ for the SRS population gives a biased account of the
odds ratio of interest θ t due to cda, i.e. combined effect (iv). The underreporting
problems (i), (ii) and (iii) are not important if odds ratios are used. This is not
surprising given the well-known property of the odds ratio that it is insensitive to
marginal changes in a two-way contingency.25 In the context of underreporting in
SRSs, some others already noticed this.16,26 Without detailed knowledge of the
subject matter, it is difficult to say for particular applications whether cda = 1, cda
>1 or 0< cda <1. It should be born in mind that the factor cda should be interpreted
as a factor which corrects the more general correction effects c, cd and ca.
Basically, therefore when a particular contingency table like Table 4 is studied,
the conclusion as to whether the estimate of θ is an over- or an underestimate of
θ t can be drawn only from substantive knowledge of the processes that play a role
in reporting. But equation (2) shows that, in deciding this, we do not need to
bother about the overall underreporting discussed in (i), about general
underreporting of the drug discussed in (ii), and general underreporting of the
ADR discussed in (iii). We only need concern ourselves with the specific
combination underreporting (iv). We now make a remark on the estimation of the
odds ratio by the loglinear model and by logistic regression as this is helpful in
understanding the generalisation of our findings to drug-drug interactions, and it
enables us later in this paper to control for possible covariates such as gender and
age. The loglinear model for the 2 x 2 table with elements mik is
52
A
log mik = u + u i + u k + u ik ,
D AD
(3)
where, u refers to the overall mean effect, u iA refers to the ADR effect, u kD refers
to the drug effect, and u ikAD refers to their interaction effect. These parameters add
up to zero over each index in order to identify the model. Filling in (3) into (2)
gives us the relation between the parameters uij and the odds ratio θ
q = m11 m22 = exp(4u11AD ) q (4)

m12 m21
It follows that, if there is no interaction term u ikAD needed in the loglinear model for
Table 4 (i.e. in the SRS data the specific drug and ADR are unrelated), the
estimated odds ratio will be equal to one. We can also rewrite the loglinear model
into a logistic regression model:
log
mi1 D AD D AD D AD A
= u1 + ui1 - u 2 - ui 2 = 2 u1 + 2 ui1 = b + bi ,
(5)
mi 2 ik
( )
This shows that q = exp 2b1A , and it follows that both the loglinear model as well
as the logistic regression model can be used to test whether the odds ratio in the
SRS departs significantly from 1.
3.2 Diuretics and possible ADRs revisited

In section 2.1 we studied Table 1. The odds ratio estimated in the SRS data was
1.58, and the question is in what way this odds ratio is biased by underreporting.
Section 3.1 shows that three underreporting effects do not concern us here,
namely the general effect (i) that holds for all four frequencies in Table 1, the
drug effect (ii) that the underreporting of ADRs the group for which diuretics are
prescribed differs from ADRs in those for which they are not prescribed, and the
ADR effect (iii) that the underreporting of congestive heart failure may be
different from underreporting other ADRs. Only the combination effect (iv)
53
Chapter 1.2
should concern us here, i.e. the possible effect that physicians and pharmacists
submit in more or fewer reports specifically for the occurrence of signs indicating
congestive heart failure when diuretics are prescribed. Such a possible effect will
bias the observed odds ratio of 1.58 positively or negatively. If the bias is
negative, the odds ratio in the ADR experiencing population will be larger. Since
the aim of an SRS is to generate signals for an existing drug-ADR relation,
negative bias should not worry us in this example. However, in general the bias
may be such that it reduces the odds ratio to a small value that then is not a large
enough signal to stand out from the noise. If the bias is positive, the odds ratio in
the ADR experiencing population will be smaller, and the SRS is possibly
generating a ’false’ signal. In this particular case we may be looking at a situation
in which a lack of efficacy of the diuretics was perceived, possibly due to a
confounding by indication. Nevertheless, since this situation is not to be expected,
the odds ratio may be overestimated in the SRS population.
4 The underreporting problem for two drugs
First we will work out the general case of two drugs (section 4.1), then proceed to
make some further assumptions and study the consequences of these (section 4.2).
In section 4.3 we reconsider the example of drug-drug-ADR interaction of
diuretics and NSAIDs, and discuss the consequences of sections 4.1 and 4.2 for
the interpretation.
4.1 Theory, the general case

In this section we will study the underreporting problem in case of two drugs.
This pertains to a table of 2 x 2 x 2 with elements mijk, where i (i = 1,2) refers to
the presence or absence of the first specific drug in a report, j (j = 1,2) refers to
the presence or absence of the second specific drug, and k (k = 1,2) refers to the
presence or absence of the specific ADR. There are thus eight cells. We adopt the
same approach as in section 2, i.e. for these eights cells we distinguish eight types
of underreporting problems. First we write the expected frequencies of the SRS in
terms of the true frequencies for the ADR experiencing population using
constants c to cdea for these eight types of underreporting problems. Then we will
show in what way the odds ratios calculated on the SRS data are biased as a result
of these underreporting problems.
54
We distinguish the following eight types of underreporting problems:

(i) the general underreporting effect that pertains to each of the eight cells,
denoted by c;
(ii) the general underreporting effect for the first drug, denoted by cd;
(iii) the general underreporting effect for the second drug, denoted by ce;
(iv) the general underreporting effect for ADR, denoted by ca;
(v) the paired underreporting effect between both drugs denoted by cde; this
statistical interaction is unrelated to the column variable ADR, but only shows
whether the first drug is found relatively more or less often together with the
second drug.
(vi) the paired underreporting effect for the first drug and the specific ADR,
denoted by cda (this effect is comparable to combination effect (iv) in section 2);
(vii) the paired underreporting effect for the second drug and the specific ADR,
denoted by cea; and lastly,
(viii) the joint underreporting effect for both drugs with the specific ADR,
denoted by cdea. This underreporting effect shows the effect that a combination of
two drugs has on reporting an ADR, above the general and paired underreporting
effects. For example, cdea can be smaller than 1 when a drug-drug-ADR
interaction is well known so that health professionals do not take the trouble to
file a report; it can be larger than 1 when a drug-drug-ADR interaction is
suspected but not proven yet - for example, in the situation that a new drug is
marketed with properties similar to existing drugs for which a drug-drug-ADR
interaction is proven. See section 4.2 for a thorough discussion of the joint
underreporting effect.
In Table 6 we relate the expected frequencies mijk in the SRS population to the
true frequencies tijk in the ADR experiencing population using the constants c to
cdea. Thus we can answer the question of how the SRS estimates provide a biased
account of the odds ratios in the ADR experiencing population.
In three-way tables, the possible relations between the variables can be
summarised by so-called conditional odds ratios, i.e. two odds ratios between i
and j for k = 1,2, two odds ratios between i and k for k = 1,2, and two odds ratios
between j and k for k = 1,2. We will denote the odds ratios derived from the SRS
between i and j for k = 1,2 as q ij k =1 and q ij k = 2 , and the corresponding odds
ratios derived from the ADR experiencing population by q tij k =1 and q tij k = 2 .
55
Chapter 1.2
Table 6. Two drugs. Expected frequencies mijk for the SRS population, expressed
in terms of true frequencies tijk used for the ADR experiencing population
Specific drug 1 Specific drug 2 Specific ADR
absent present
absent absent ccde cda cea cca cde cdea

t111 t112
cd ce ca cdea cd ce cda cea
present cce cda cdea cce ca cea
t121 t122
cd ca cde cea cd cde cda cdea
present absent ccd cea cdea ccd ca cda
t 211 t 212
ce ca cde cda ce cde cea cdea
present ccd ce cde ccd ce ca cde cda cea cdea t 222
t 221
ca cda cea cdea
The odds ratios for the other variables are denoted in similar ways. If we want to
relate the odds ratios derived from the SRS population to the odds ratios derived
from the ADR experiencing population we get the following results (compare
equations (1) and (2)):
m111 m221 = (c de ) t111 t 221 = (c de ) t

4 4
q ij k =1 = 4 q ij k =1×
(6)
m121 m211 (c dea )
4
t121 t 211 (c dea )
q ij k =2 = (c de ) (c dea ) q ij k =2×
4 t4
(7)
(c da ) t
4
q ik j =1 = q (8)
(c dea )4 ik j =1×
= (c da ) (c dea ) q tik
4 4
q ik j =2 j = 2×
(9)
(c ea )4 t
q jk i =1 = q (10)
(c dea )4 jk i =1×
56
q jk i =2 = (c ea ) (c dea ) q jk i =2×
4 4 t
(11)
We conclude from these equations that

1. general underreporting effects (i), (ii) (for the first drug), (iii) (for the second
drug) and (iv) (for the specific ADR) do not lead to bias in the odds ratios derived
from the SRS population. This follows from the fact that the constants c, cd, ce and
ca do not appear in (6) to (11).
2. the odds ratios estimated from SRS data are biased by paired underreporting
effects, denoted by the constants cde, cda and cea .
3. the odds ratios estimated from SRS data are biased by the joint underreporting
effect for both drugs, and the specific ADR, by the constant cdea.
We will indicate shortly how these odds ratios can be obtained by means of
parameters of a loglinear model. Similar to equation (3), the loglinear model for
the 2 x 2 x 2 table with elements mijk is
log mijk = u + uiD + u Ej + u kA + u ijDE + u ikDA + u EA DEA
jk + u ijk , (12)
where the parameters add up to zero over each index. Therefore the relation
between the loglinear parameters and, for example, the conditional odds ratio
q ik j =1 is
q ik j =1 (
= m111 m221 = exp 4u11
DA DEA
+ 4u111 × ) (13)
m121 m211
In a similar way, the odds ratios q ij k and q jk i can be obtained from the
parameters of a logistic regression model where the logit of the specific ADR is
predicted by the first and the second drug, and their interaction.
4.2 Theory, further assumptions

Regarding cdea, three different situations can be distinguished. Firstly, in the event
that the drug-drug interaction is well known in the literature and is common
knowledge of physicians and pharmacists, the chance that this drug-drug
interaction is reported to an SRS is probably less likely, so in this situation we
might expect that cdea <1. Secondly a drug-drug interaction may not yet be
associated with a specific drug but may be expected, for example in the event that
a related drug is known to cause a similar drug-drug interaction. In this situation,
57
Chapter 1.2
the interaction is probably more easily reported, so we might expect cdea >1.
Finally, in the event that an interaction is not known in the literature and is not to
be suspected, one may assume that cdea =1. After all, generally there is no reason
why in the event the specific combination of both drugs is used, an ADR should
be more or less frequently reported. Regarding the main purpose of SRSs,
however, i.e. generating signals of previously unknown ADRs or drug-drug
interactions, this situation need not necessarily be unfavourable.
Assume now that the joint underreporting effect for both specific drugs with the
specific ADR can be ignored, i.e. in equations (6) to (11) cdea = 1. These equations
then simplify considerably. Under this assumption we can make the following
observations:
1. the odds ratios are only biased because of the paired underreporting effects, i.e.
due to cde, cda and cea.
2. the bias problem is simple in the sense that the pairs of odds-ratios in (6) and
(7), in (8) and (9), and in (10) and (11), are biased by the same factor. For
example, for q ik j =1 and q ik j=2 this factor is c4da .
3. Because of 2., it holds for each pair of odds ratios that the ratio of the elements
of the pair in the SRS population is unbiased for the corresponding ratio of odds
ratios in the ADR experiencing population. For example, the ratio of odds ratios
t t
q ik j =1 / q ik j =2 = q ik j =1 / q ik j =2 . This is a useful result. For example, if we find that
qˆik j =1 = 1.0 and qˆik j = 2 = 4.0, this implies that, in the SRS population, the odds
ratio between the first drug and the specific ADR is estimated as 1 if the second
drug is absent (so then there is no relation) but as 4 when the second drug is
present. This shows that the combination of both drugs coincides with an ADR in
the SRS population. The estimates of both odds ratios are biased by an unknown
factor cda, and therefore we do not know the estimate of these odds ratios in the
ADR experiencing population. However, since q ik j =1 / q ik j =2 = q tik j =1
/ q tik j =2
both
in the SRS population as well as in the ADR experiencing population the odds
ratio between the first drug and the ADR is four times greater when the second
drug is present than when it is absent.
4. Assume now, in addition to the assumption cdea = 1, that in the SRS population
DEA
there is no drug-drug-ADR interaction, i.e. u ijk = 0 . For clarity of exposition, we
58
focus our discussion now on q ik j =1 and q ik j=2 (compare (8), (9) and (13)), but
these results also hold for the other odds ratios. By working out (13) for both
DA
q ik j =1 and q ik j=2 , we find q ik j =1 = q ik j=2 = exp 4 u11 . Second, by substituting
4 t 4 t
this in equations (8) and (9), we find q ik j =1 = q ik j=2 = cda q ik j =1
= cda q ik j =2
. Since
t t
q ik j =1 = q ik j =2 , this implies the next result: if cdea = 1 and there is no drug-drug-
ADR interaction in the SRS population, then there will be no drug-drug-ADR
interaction in the SRS population. We now show that the reverse also holds. If in
the ADR experiencing population the interaction (u ) = 0 ,
DEA t
ijk then
q ik j =1 = q ik j =2 = exp 4 (u11 )
t t t
DA . Underreporting leads to multiplication with a factor
-4
cda (the reverse of what happens in (8) and (9)), and therefore
q ik j =1 = q ik j =2 = c da q ik j =1 = c da q ik j =2 = c da exp 4 (u11 )
t t -4 -4 -4 DA t
This shows that
q ik j =1 = q ik j=2 , and therefore, if the drug-drug-ADR interaction is zero in the

ADR experiencing population, it will also be zero in the SRS population.
Therefore we can carry out a loglinear analysis in the SRS population (or,
equivalently, a logistic regression) to assess whether there is a drug-drug-ADR
interaction in the ADR experiencing population. Under the assumption that cdea =
1, if there is no drug-drug-ADR interaction in the ADR experiencing population,
we can use data from an SRS to assess this. If there is such an interaction, then the
ratio of the odds ratios estimated from data in the SRS is equal to the ratio of the
odds ratios in the ADR experiencing population (see result (iii) and the
Addendum).
4.3 Drug-drug interaction of diuretics and NSAIDs revisited

In the notation of Section 4.2, the odds ratios estimated for the SRS population in
section 2.2, having values 1.29 and 3.04, are q ik j =1 and q ik j=2 . Equations (8) and
(9) show us that these give biased accounts of the odds ratios q tik j =1
and q tik j =2
in
the ADR experiencing population by the underreporting factors cda and cdea. Other
underreporting effects play no role. The factor cda reflects a possible paired
underreporting of the diuretics – congestive heart failure combination. The factor
59
Chapter 1.2
cdea reflects joint underreporting of diuretics and NSAID on congestive heart

failure. These two factors may cause the estimates 1.29 and 3.04 to be biased, but
without further information it is unclear in what direction and in what magnitude.
Let us now assume that we can ignore cdea, i.e. cdea = 1, as is discussed in section
3.2. This assumption generally is true when underreporting of congestive heart
failure due to a diuretic drug is not influenced by the presence or absence of a
NSAID. Then q ik j =1 and q ik j=2 are only biased by the same factor cda. It follows
that we can say that, whatever size of the bias factor cda, in the ADR experiencing
population the odds ratio q tik j =2
is 3.04 / 1.29 = 2.35 times as large as the odds
ratio q tik j =1
. In other words, the relation between diuretics and congestive heart
failure is 2.35 times as strong when a NSAID is used compared with that NSAID
is not used. Some manipulation of equations from Section 3.2 shows that this
DEA DEA
value 2.35 = exp 4 uˆ ijk , and the Appendix shows that if cdea = 1 this uˆ ijk
estimated in the SRS population is unbiased for the ADR experiencing
population.
5 Covariates
5.1 Theory
Here we discuss the assessment of interaction between a specific drug and a
specific ADR in the presence of covariates one would like to control for. We do
this for a simple case only, viz. for the case of one drug and one categorical
covariate. To make the exposition simple, we take as covariate sex, indexed by s,
with levels male and female. The approach of this simple case can also be used
for more complicated cases, such as when there are more drugs and more
covariates. The situation is equivalent to the approach for section 4, with the
second drug replaced by the covariate. Therefore we keep this section brief.
Just as in section 4 there are eight underreporting problems, leading to the eight
constants c, cs, cd, ca, csd, csa, cda and csda. The odds ratios estimated in the SRS
population are biased with respect to the ADR experiencing population using
these constants, and it is possible to find equations similar to equations (6) to (11).
As in section 4, in the general case this gives odds ratios that are difficult to
interpret due to biases which cannot be estimated from the SRS. However, it is
60
clear that the odds ratios are only biased because of paired and joint
underreporting effects csd, csa, cda and csda, and not because of general
underreporting effects c, cs, cd, ca.
If we make the assumption that there is no joint underreporting effect, then csda =
1, and all equations simplify considerably (compare section 4.2). Examples of csda
¹ 1 are difficult to imagine. We give a few examples to see this. Consider the
appearance of facial hair or the disappearance of scalp hair. Both are more likely
to be reported if they occur in women than in men. However, if this reporting is
irrespective of the drug, then csa ¹ 1 but csda = 1. As a second example, there is
wide recognition that women consult their doctor more frequently than men, thus
leading to increased opportunities for ADRs to be reported in women. However,
if this underreporting for males is irrespective of drugs and ADRs, then this
would lead to cs ¹ 1, but csa = cda = csda = 1. As a last example, ADRs affecting
someone’s ability to work may be more likely to be reported if they occur in men
(assuming men make up a larger proportion of the work-force). Again, if this
underreporting for men is irrespective of drug and ADR, then cs ¹ 1, but csa = cda
= csda = 1. So we conclude that, although the specific combination of both drug,
sex and ADR may give rise to a different level of reporting, generally however it
is unlikely that the specific underreporting factor concerning the combination of
both drug, gender and ADR is not equal to one.
As in section 4.2, if the assumption that csda = 1 is realistic it is useful to fit a
loglinear or logistic regression model to assess whether there is an interaction
between drug, sex and ADR in SRS data. If this drug-sex-ADR interaction can be
deleted from the model fitted on the SRS data, this means that there is no
evidence for this drug-sex-ADR interaction in the ADR experiencing population
either. If the drug-sex-ADR interaction is included in the model, however, the
conditional odds ratios are biased, but ratios of these conditional odds ratios are
not (see section 4.2 for details).
5.2 Sex differences for ADRs of diclofenac revisited

We now revisit the analysis of sex differences for ADRs of diclofenac discussed
in section 2.3. We found an odds ratio of 12.4 for males and 10.0 for females. The
drug-sex-ADR interaction in the loglinear model was not significant, and the
estimated odds ratio in this model is 10.8. We now discuss the possible effects of
underreporting on these results.
61
Chapter 1.2
These odds ratios estimated from the SRS data give possibly biased accounts of
the corresponding odds ratios in the ADR experiencing population because of the
paired underreporting cda of drug by ADR and the joint underreporting effect csda
for sex, drug and ADR. Notice that general underreporting effects cs for sex, cd
and ca, as well as the paired underreporting effect for sex and drug csd and for sex
and ADR csa, do not bias these odds ratios.
If we assume that the joint underreporting effect csda can be neglected (i.e. csda =
1), since there is no reason for anaphylactic reactions due to diclofenac to be
reported more frequently in either men or women, then the loglinear model
without interaction term for sex by drug by ADR also holds in the ADR
experiencing population. The only problem to worry about is the underreporting
effect cda, but in this example it is unlikely that this underreporting effect is so
large that it will bring down the estimated odds ratio of 10.8 in the SRS
population to 1 in the ADR experiencing population.
6 Conclusion
Spontaneous Reporting Systems (SRS) are plagued by underreporting problems

of ADRs. The aim of this paper was to clarify the different ways in which
underreporting can play a role. In particular, we have focused on the question of
which types of underreporting problems plague SRS data and which types are
harmless when we want to make statements about the ADR experiencing
population using data from the SRS population.
It turns out that general underreporting effects for specific drugs and for specific
ADRs are harmless, but paired underreporting effects result in biases in the odds
ratios estimated from the SRS. If there is no so-called joint underreporting effect
of two drugs on the ADR, then SRS data can be very useful for assessing the
presence of drug interactions, and their relative size.
We approach the analysis of SRS data by firstly conducting loglinear analyses of
data, secondly by discussing the existence of possible underreporting effects, and
thirdly by discussing whether the odds ratios found in the SRS data are likely to
be affected by these underreporting effects to such an extent that conclusions
about the existence (i.e. not the size) drug - ADR - relations would have to be
modified.
62
We think this paper provides a better understanding of the possible pitfalls, but
also of the validity of drawing conclusions from the analysis of SRS data.
Acknowledgements
We gratefully acknowledge the support from the Dutch Fund for Prevention
(number 28.2632).
References
1. Finney DJ. The detection of adverse reactions to therapeutic drugs. Stat Med 1982;1:153-
161.
2. Finney, DJ. Statistical logic in the monitoring of reactions to therapeutic drugs. Methods Inf
Med 1971;10:237-45.
3. Finney DJ. Systemic signalling of adverse reactions to drugs. Methods Inf Med. 1974;13:1-
10.
4. Waller PC, Lee EN. Responding to drug safety issues. Pharmacoepidemiol Drug Saf
1999:535-552.
Risk and Safety in Medicine 1993;5:123.
6. Van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens HGM. Signalling possible
drug-drug interactions in a spontaneous reporting system: Delay of withdrawal bleeding
during concomitant use of oral contraceptives and itraconazole. Br J Clin Pharmacol
1999;47:689-693.
7. Van Puijenbroek EP, Egberts ACG, Heerdink ER, Leufkens HGM. Detecting drug-drug
interactions using a database for spontaneous adverse drug reactions: An example with
diuretics and non-steroidal anti-inflammatory drugs. Eur J Clin Pharmacol. 2000;56:733-8.
8. de Koning, GHP. A Regionalized Spontaneous Surveillance Program for Adverse Drug
Reactions as a Tool to Improve Pharmacotherapy. (Thesis) Utrecht University, Utrecht, the
Netherlands, 1994:76-96.
9. Stephens MDB. Detection of new adverse drug reactions. London: MacMillan Publishers
Ltd, 1992.
10. Haramburu F. Estimation of underreporting. Post Marketing Surveillance 1993;7:39-49.
FDA Spontaneous Reporting System (with discussion). American Statistician 1999;53:177-
202.
12. Praus M, Schindel F, Fescharek R, Schwarz S. Alert systems for post-marketing surveillance
of adverse drug reactions. Stat Med 1993;12:2383-2393.
13. Tsong Y. Comparing reporting rates of adverse events between drugs with adjustment for
year of marketing and secular trend in total reporting. J Biopharm Stat 1995;5:95-114.
63
Chapter 1.2
14. Martin RM, Kapoor KV, Wilton LV, Mann RD. Underreporting of suspected adverse drug
reactions to newly marketed (”black triangle”) drugs in general practice: observational study.
BMJ 1998;317:119-20.
15. Eland IA, Belton KJ, van Grootheest AC, Meiners AP, Rawlins MD, Stricker BH.
Attitudinal survey of voluntary reporting of adverse drug reactions. Br J Clin Pharmacol
1999;48:623-627.
16. Stricker BHC, Tijssen JPG. Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;10:1177-1184.
17. Egberts AC, Meyboom, RH, De Koning FH, Bakker A, Leufkens HG. Non-puerperal
lactation associated with antidepressant drug use. Br J Clinical Pharmacology 1997;44:277-
281.
18. Evans SJW. Pharmacovigilance: a science or fielding emergencies? Stat Med 2000;19:3199-
3209.
19. Egberts ACG, Van der Hofstede JW, Meyboom RHB, de Koning GHP, Bakker A, Leufkens
HGM. Transformation of a database of spontaneously reported suspected adverse drug
reactions and its use as a tool in signal detection. In: Egberts ACG. Pharmaco-
epidemiological approached to the evaluation of antidepressant drugs. (Thesis) Utrecht
University, Utrecht, the Netherlands, 1997:111-124.
20. Pierfitte C, Begaud B, Lagnaoui R, Moore ND. Is reporting rate a good predictor of risks
associated with drugs? Br J Clin Pharmacol 1999;47:329-31.
21. Anonymous. WHO Adverse Drug Reaction Dictionary. WHO Centre for International Drug
Monitoring, Uppsala, Sweden, 1995.
22. Feenstra J, Grobbee DE, Mosterd A, Stricker BHC. Adverse cardiovascular effects of
NSAIDs in patients with congestive heart failure. Drug Saf 1997;17:166-180.
23. Heerdink ER, Leufkens HG, Herings RM, Ottervanger JP, Stricker BH, Bakker A. NSAIDs
associated with increased risk of congestive heart failure in elderly patients taking diuretics.
Arch Intern Med 1998;158:1108-1112.
24. O’Brien WM. Adverse reactions to nonsteroidal anti-inflammatory drugs. Diclofenac
compared with other nonsteroidal anti-inflammatory drugs. Am J Med 1986;80:70-80.
25. Agresti. Categorical data analysis. John Wiley and Sons, 1990.
26. Baudraz-Rosselet F, Rakosi T, Wili PB, Kenzelmann R. Treatment of onychomycosis with
terbinafine. Br J Dermatol 1992:126:40-46.
64
Part 2
PRACTICAL APPROACHES OF QUANTITATIVE

SIGNAL DETECTION
In the second part of this thesis, examples of quantitative signal detection are
given and two new approaches are introduced. Chapter 2.1 deals with the analysis
of a possible association between a single drug and a suspected adverse drug
reaction (ADR). The aim of this study is to investigate whether the risk of
anaphylactic reactions being reported during the use of various non-steroidal anti-
inflammatory drugs (NSAIDs) is greater than with other classes of drugs and
whether differences among NSAIDs exist.
Drug-drug interactions are relatively rarely reported to spontaneous reporting
systems (SRSs) for adverse drug reactions. In Chapter 2.2 a new quantitative
approach is introduced to analyse possible drug-drug interactions in datasets of
SRSs. As an example we analyse the ADR 'delayed withdrawal bleeding'
resulting from a possible interaction between itraconazole and oral contraceptives
in reports received by the Netherlands Pharmacovigilance Foundation Lareb. As a
second example, in Chapter 2.3 the influence of concomitant use of diuretics and
non-steroidal anti-inflammatory drugs (NSAIDs) on symptoms indicating a
decreased efficacy of diuretics is examined. Likewise drug-drug interactions,
datasets of SRS can be used to study the possibility of clustering of symptoms, as
is illustrated in Chapter 2.4. The objective of this study is to analyse the clustering
of the ADRs arthralgia, urticaria and fever during the use of the antimycotic drug
terbinafine, statistically in order to determine whether or not these symptoms are
interrelated.
2.1
DIFFERENT RISKS FOR NSAID-INDUCED ANAPHYLAXIS
EP van Puijenbroek1, ACG Egberts2,3, RHB Meyboom1,4 and HGM Leufkens3
1
2
3
4
The Uppsala Monitoring Centre, Uppsala, Sweden
THE ANNALS OF PHARMACOTHERAPY (IN PRESS)

Chapter 2.1
Summary
Objective After marketing of drugs, Spontaneous Reporting Systems (SRS) can

provide valuable information regarding the occurrence of suspected adverse drug
reactions (ADRs). The Netherlands Pharmacovigilance Foundation received a
substantial number of anaphylactic reaction reports related to the use of non-
steroidal anti-inflammatory drugs (NSAIDs). The aim of our study was to
investigate whether the risk of anaphylactic reactions being reported during the
use of various NSAIDs is greater than with other classes of drugs and if
differences among NSAIDs exist.
Methods In a case/non-case design Reporting Odds Ratios (RORs) were
calculated using logistic regression analysis. Cases were defined as reports in
which anaphylactic or anaphylactoid reactions were reported, all other reports
were considered as non-cases. The index group consisted of reports on which
NSAIDs were mentioned as the suspected medication, the reference group
consisted of all other reports.
Results Between January 1985 and November 2000, the Netherlands
Pharmacovigilance Foundation Lareb received 76 cases concerning anaphylactic
reactions to NSAIDs. These drugs are strongly associated with anaphylactic
reactions. The ROR, adjusted for age, gender and source of the reports was 9.4
(95% Confidence Interval 6.9-12.7). Anaphylactic reactions associated with the
use of naproxen, ibuprofen and diclofenac were reported disproportionally with
respect to other drugs. The corresponding Reporting Odds Ratios from logistic
regression analysis, adjusted for age, gender and reporting source for diclofenac,
naproxen and ibuprofen were 17.2 (95% CI 12.1-24.5), 9.1 (95% CI 5.2-15.9) and
5.5 (95% CI 2.5-11.9), respectively.
Conclusion The results of this study strengthen previous findings concerning the
relative high risk of developing an anaphylactic reaction during the use of
NSAIDs, particularly diclofenac, ibuprofen and naproxen.
68
NSAIDs and anaphylactic reactions
Introduction
Non-steroidal anti-inflammatory drugs (NSAIDs) have a prominent place in the

treatment of pain and arthritis.1,2 NSAIDs, however, are also associated with
serious and non-serious complications. The most frequently occurring adverse
drug reactions (ADRs) related to NSAID use concern the gastrointestinal tract and
kidney. Rarely blood cell or liver function disorders can occur.3 NSAIDs have
also been associated with acute allergic reactions, varying from urticaria and
angioedema to anaphylactic reactions. With respect to the occurrence of
anaphylactic reactions, limited evidence is available concerning the relative safety
of these drugs. Van der Klauw et al., have shown that among patients admitted to
a hospital because of anaphylactic reactions, NSAIDs were a frequent cause.4
Also in spontaneous reporting systems (SRS) for adverse drug reactions NSAIDs
were frequently associated with anaphylactic reactions.5 Although these
potentially serious ADRs are reported with various NSAIDs,4-20 it is not clear
whether there are differences between individual agents.
After new drugs reach the market, SRS can provide valuable information
regarding possible new ADRs. Furthermore, datasets of SRS contain information
about ADRs that are already known, but were reported because of the seriousness
of the reaction involved or concern of the reporting health professional. In this
context a substantial number of reports of anaphylactic reactions related to the use
of NSAIDs were received by the Netherlands Pharmacovigilance Foundation
Lareb. The aim of our study was to investigate whether the risk of anaphylactic
reactions being reported during the use of various NSAIDs is greater than with
other classes of drugs and if differences among NSAIDs exist.
Methods
Setting
The Netherlands Pharmacovigilance Foundation Lareb maintains the national
spontaneous reporting system (SRS) on behalf of the Dutch Medicines Evaluation
Board. ADRs are reported by physicians and pharmacists on a voluntarily basis.
Reports that are received are subject to review by qualified assessors. Special
attention is paid to the description of the ADR. If needed, additional information
concerning the clinical details of the report is retrieved. Data concerning the
69
Chapter 2.1
suspected adverse drug reaction and the drugs involved are coded using the WHO
adverse drug reaction terminology and the Anatomical Therapeutic Chemical
(ATC) classification system respectively, and subsequently filed in a database.21
In the ATC classification system, the drugs are divided into different groups
according to the organ or system on which they act and their chemical,
pharmacological and therapeutic properties.22
The primary goal of an SRS is to give an early warning of a possible causal
relationship between an ADR and a drug of which the relation was previously
unknown or incompletely documented.23 For signal detection every new report is
reviewed on a regular basis in a discussion meeting in which the reported
association between the suspected drug and ADR is assessed. This meeting is
attended by assessors with specific experience on the field of spontaneous
reporting. In addition to this ‘case-by-case’ analysis, the extent to which a
possible ADR is reported for a given suspected drug can be analysed
statistically.24-29 We analysed the association between NSAIDs and anaphylactic
reactions using a case/non-case design and expressing the strength of the
association as the ADR Reporting Odds Ratio (ROR).24,29
Design
The analysis included all reports received by Lareb between January 1st 1985 and
November 1st 2000 in which data concerning age and gender of the patients were
available. Patients younger than 10 years of age were excluded, because NSAIDs
are rarely used in children in the Netherlands. In a case/non-case design, RORs
were calculated by means of logistic regression analysis. The ROR is defined as
the ratio of the exposure odds among reported cases to the exposure odds among
reported non-cases.24,29 Cases were defined as all reports coded with the WHO
preferred term ‘anaphylactic shock’ or ‘anaphylactoid reaction’. These codes
encompass the WHO included terms anaphylactic shock, anaphylactoid reaction,
anaphylactic reaction, anaphylaxis and red neck syndrome. Both preferred terms
are considered to be a manifestation of an anaphylactic reaction, which can be
considered as an acute systemic adverse reaction, simultaneously involving
several organ systems.30 Characteristic symptoms of such a reaction may involve
the skin, respiratory system, cardiovascular system, gastrointestinal system or
neuropsychological symptoms.31 In the event that allergic symptoms concerning
two or more organ systems were reported, the ADRs were classified as an
70
anaphylactic reaction. Additionally, if shock-like symptoms were reported and the

blood pressure was known to have dropped, reports were classified as
anaphylactic shock. If symptoms pertaining to only one system and organ class
were reported, like urticaria, bronchospasm or shock, the ADRs were coded as
such, and subsequently were treated as non-cases. All other reports were
considered as non-cases. The index groups consisted of reports on which an
NSAID (ATC code beginning with M01A) was mentioned as the suspected
medication. The reference group consisted of all reports in which no NSAID was
mentioned as the suspected medication.
Since 1985, the Netherlands Pharmacovigilance Foundation Lareb received
reports on 23 different ATC codes concerning NSAIDs or combinations of
NSAIDs and other drugs. Because of this large volume, results of the analysis of
all separate NSAIDs were restricted to those drugs from which more than 100
reports were received.
NSAIDs can be used in combination with other drugs, such as misoprostol.
Theoretically one of the other drugs in the combination could be responsible for
the anaphylactic reaction. In case combinations of NSAIDs and other drugs have
been registered in the Netherlands, however, reports on these drugs were regarded
as reports on the NSAID present in the combination concerned. In the event
where next to the NSAID another suspected medication is reported, the reports
are also regarded as reports on the NSAID involved.
RORs, adjusted for age and gender of the patients, year of reporting and source of
the reports, either physician or pharmacist, were expressed as point estimates with
corresponding 95% confidence intervals. For all statistical analyses, SPSS 10.0
was used.
Results
Between January 1985 and November 2000, a total number of 28,003 reports of
suspected ADRs were received. Of these, 841 reports were excluded because the
patient was younger than 10 years, while 252 reports were excluded because
gender or age of the patients were not reported. Of the remaining 26,910 reports,
76 concerned a possible relationship between an NSAID and an anaphylactic
reaction. All cases referred to oral dosage form, except for one case concerning
71
Chapter 2.1
the use of diclofenac (injection). Table 1 shows the distribution of age, gender and
source of the reports among cases and non-cases.
Table 1. Distributions of age, gender and source of the reports among cases and non-cases
Cases Non-cases
(n=190) (n=26,720)
n (%) n (%)
Mean age (years) (standard 52.2 (17.6) 47.7 (15.5) p<0.01*

deviation)
Number of females (%) 121 (63.7) 17,294 (64.7) n.s.**
Number of reporting 149 (78.4) 18,690 (69.9) p<0.05**

physicians (%)***
*Student-t test; ** Pearson Chi square; n.s. not statistically significant; *** in comparison
with the number of pharmacists
Among the cases, the age of patients was statistically significant higher and they
were also more frequently reported by physicians as compared with pharmacists.
The distribution between cases and non-cases for different NSAIDs are shown in
table 2.
Table 2. Distribution of cases and non-cases among various NSAIDs and other suspected
drugs
Cases Non cases
(n=190) (n=26,720)
n (%) n (%)
All NSAIDs 76 (40) 2,420 (9.1)
Diclofenac 51 (26.8) 866 (3.2)

Naproxen 15 (7.9) 448 (1.7)
Ibuprofen 7 (3.7) 296 (1.1)
Piroxicam 1 (0.5) 180 (0.7)
Ketoprofen 0 175 (0.7)
Indomethacin 0 105 (0.4)
Other NSAIDs 2 (1.1) 350 (1.3)
Other suspected drugs 114 (60.0) 24,300 (90.9)
72
Among the NSAIDs on which more than 100 reports were received, anaphylactic
reactions were reported on diclofenac, naproxen, ibuprofen and piroxicam. On
ketoprofen and indomethacin no cases were reported, implying no ROR could be
calculated for the latter drugs. Among the NSAIDs on which less than 100 reports
were received, there were 2 additional cases (sulindac and nabumetone). Table 3
shows the results of the univariate and multivariate analysis.
Table 3. Results of univariate and multivariate logistic regression analysis

Univariate analysis Multivariate analysis
ROR (95% CI) ROR (95% CI)
All NSAIDs 6.7 (5.0-9.0) 9.4 (6.9-12.7)
Diclofenac 12.5 (9.0-17.6) 17.2 (12.1-24.5)

Naproxen 7.1 (4.1-12.3) 9.1 (5.2-15.9)
Ibuprofen 5.0 (2.3-10.9) 5.5 (2.5-11.9)
Piroxicam 1.2 (0.2-8.5) 1.2 (0.2-11.7)
Other NSAIDs 1.2 (0.3-4.9) 1.4 (0.3-5.6)
NSAIDs were strongly associated with reports of anaphylactic reactions. The

ROR, adjusted for age, gender and source of the reports was 9.4 (95% confidence
interval 6.9-12.7).
The total number of reported reactions, the RORs with corresponding confidence
intervals, adjusted for age, gender and source of the reports and the number of
cases for different NSAIDs are shown in Figure 1. Anaphylactic reactions were
disproportionally more reported for ibuprofen, naproxen and diclofenac. For
diclofenac, the adjusted ROR was 17.2 (95% CI 12.1-24.5), for naproxen 9.1
(95% CI 5.2-15.9) and for ibuprofen 5.5 (95% CI 2.5-11.9).
73
Chapter 2.1
100 60 reporting odds ratio

number of reported cases
50
reporting odds ratio (95% CI)

10 40
number of reports
30
1 20
10
0,1 0
diclofenac naproxen ibuprofen piroxicam
NSAID
Figure 1. Results of the analysis of anaphylaxis associated with various NSAIDs versus
all other reported cases. The total number of reported cases and the Reporting Odds Ratios
(semi-logarithmic scale) are shown, adjusted for gender and age of the patient, year of
reporting and source of the reports, with corresponding 95% confidence intervals for
diclofenac, ibuprofen, naproxen and piroxicam.
Discussion
The results of our study shows that risk of an anaphylactic reaction being reported
is increased during the use of NSAIDs, notably on diclofenac, naproxen and
ibuprofen. Among the NSAIDs, diclofenac appears to have the highest reporting
rate.
Interpretation of the results originating from spontaneous reporting systems
should be done with great care. The method applied provides quantitative
information about the extent of the reported associations between reported
suspected drugs and ADRs with respect to other reports sent to the SRS. The
rationale of the case/non case design is that the proportion of ADRs is relatively
constant over time and for this reason, the reference group can be considered to be
a measure of the ‘background frequency’ of the suspected ADRs. This implies
that interpretation of the quantitative results based on data sets of SRS requires
specific knowledge of the composition of the database. For this reason
spontaneous reporting systems are primarily used for signal detection purposes
and not for hypothesis testing. An estimation of the actual incidence of the ADR
74
in populations using the suspected drug cannot be made. Prescription data cannot
be used to estimate the actual use of the drugs since some NSAIDs, like ibuprofen
and naproxen, are also available without prescription in the Netherlands.
A case cohort study in the Netherlands showed previously that diclofenac in
particular was among the most frequent causes of anaphylactic reactions leading
to hospital admission, the relative risk of anaphylaxis relative to all other drugs
being 9.5 (95% CI 3.7-24.5).4 The results of our study are in accordance with
these findings, but furthermore demonstrate that anaphylactic reactions are also
reported disproportionally for naproxen and ibuprofen.
Anaphylaxis is an immediate (type I) hypersensitivity reaction to an allergen,
caused by its rapid cross-linking with specific IgE on tissue mast cells and
peripheral blood basophils. It requires previous exposure to the foreign antigen.
An anaphylactoid reaction, however, is not an IgE mediated response but,
similarly, involves inflammatory mediators to be released from mast cells and
basophils. This activation of immune cells may occur both directly and as the
result of disturbances in arachidonic acid metabolism and immune complex-
mediated activation of complement.32 These non-IgE mediated reactions, or
anaphylactoid reactions, do not require previous exposure and may also be caused
by NSAIDs.20,32,33 Although pathophysiology differs to a certain extent,
anaphylactic reactions and anaphylactoid reactions share the same clinical
features and cannot be distinguished on clinical grounds.32,33 It is unclear whether
anaphylactic or anaphylactoid reactions predominate. For this reason, no
distinction could be made between both anaphylaxis and anaphylactoid reactions
in this study.
Among reported cases of an anaphylactic reaction during the use of a NSAID, the
reaction was fatal in one case. This patient, a 62-year-old female, used diclofenac.
There were four fatal cases among the patients in whom an anaphylactic reaction
was reported in association with another drug. Fatal cases associated with an
anaphylactic reaction were not statistically significant between NSAIDs and other
drugs (Fisher’s exact test p>0.05).
NSAIDs can be subclassified with respect to their chemical structure. Diclofenac,
together with tolmetin and ketorolac belong to the heteroaryl acetic acids,
ibuprofen, naproxen, flurbiprofen, ketoprofen, fenoprofen and oxaprozin to the
arylpropionic acids.34 When NSAIDs that have a similar chemical structure are
grouped, heteroaryl acetic acids have an adjusted ROR of 19.7 (95% CI 13.8-
75
Chapter 2.1
28.1) while the adjusted ROR for the arylpropionic acids was 6.7 (95% CI 4.2-
10.6). This suggests that the risk for an anaphylactic reaction is higher with the
use of heteroaryl acetic acids.
In our study, of the 166 reports on the combination of diclofenac and misoprostol
2 reports were characterised as anaphylactic reactions and were labelled as reports
on diclofenac. The distribution of cases and non-cases among diclofenac and the
combination of diclofenac/misoprostol were similar (Fisher’s exact test p>0.05),
showing that misoprostol was not likely to have an additional effect on the chance
of an anaphylactic reaction being reported. In the event that combinations of
NSAIDs and other drugs are considered to be non-cases, the ROR adjusted for
year of reporting, age, gender and source of the reports for diclofenac referred to
all reports on non NSAIDs was 17.0 (95% CI 11.7-24.6).
Next to the suspected NSAID another suspected medication has been reported in
95 cases. In our study, these reports were not excluded, but regarded as reports on
the NSAID involved. In two of these reports an anaphylactic reaction has been
reported. The distribution among cases and non-cases between the reports in
which two or more suspected medications were reported did not differ from
reports were only an NSAID was reported (Fisher’s exact test p>0.05). This
suggests that the anaphylactic reactions where likely to be caused by the NSAIDs
involved. When reports with other suspected drugs beside NSAIDs and were
regarded as non cases, the adjusted ROR for diclofenac was still 20.0 (95% CI
13.7-29.2).
Non-selective reporting of either the suspected drug or the suspected ADR has a
similar effect on numerator and denominator of the ROR. For this reason, non-
selective reporting has no influence on the magnitude of the ROR.24 Selective
reporting on the combination of drug and ADR, reflecting the concern of health
care professionals involved, however, may influence the ROR. We believe that
this non-differential bias, for instance precipitated by specific media attention to
anaphylactic reactions on certain drugs, is not likely to have occurred. Another
confounding factor can be the intermittent use of some types NSAIDs that
enhances the chance of sensitisation.
Elevated concentrations of leukotrienes can be found in tissues or exudates in
several diseases, including asthma, diverse allergic states psoriasis,
spondyloarthritis, and gout.35 It is unclear, if these elevated concentrations also
enhances the chance for anaphylaxis and the indication for use subsequently
76
might have been a confounding factor. Unfortunately it was not possible to take
intermittent use or the indication for use into account in this study, since this
information is only available for a limited number of reports.
In the context of studying ADRs, the presence of contraindications is rather
commonly predictive of the outcome criteria for ADRs, and may act as a
confounder.36 For example in studying the risk for peptic ulcers among patients
using NSAIDs, a previous history of gastric complaints may act as a confounder.
Anaphylactic reactions, however, can be considered ‘type B’ or ‘idiosyncratic’
effects. These ADRs are characterised by their unpredictable nature, occur rarely
and are not primarily related to the main pharmacological action of the drug.23,37
Cross-hypersensitivity among NSAIDs may occur,3 and a previous history of an
anaphylactic reaction on an NSAID may be a reason to restrain from prescribing
another NSAID in the future. Presumably, however, this condition is rare and
therefore the risk of channelling is low. Channelling refers to the phenomenon
that a drug is prescribed preferably for a specific group of patients with certain
recognised risk factors.38
Finally, in the event of concomitant use of beta-blocking drugs, signs of an
anaphylactic reaction may become more severe.39 Theoretically this may lead to a
greater reporting rate. In an additional analysis we looked for the existence of
possible risk factors among the users of NSAIDs. Age, gender and use of beta-
blocking agents among the users of NSAIDs, however, did not differ significantly
from other reports in the database.
Conclusion
Spontaneous ADR reporting data in the Netherlands show that anaphylactic

reactions are more frequently reported in association with NSAIDs than is
expected form the background frequency in the database, and that the connection
is particularly strong for diclofenac, ibuprofen and naproxen. Statistical analyses
of reporting patterns support the view that diclofenac carries a higher risk for
anaphylactic reactions than other NSAIDs. Although an estimation of the actual
incidence of the ADR in populations using the suspected drug cannot be made,
the results of our study lend support to previous findings concerning the relative
high risk of developing an anaphylactic reaction with NSAIDs. Anaphylaxis is a
rare ADR. Even if the risk for developing this ADR is increased by NSAID use,
77
Chapter 2.1
the actual risk is still low. More study is needed to determine the actual incidence
of NSAID-induced anaphylaxis.
References
1. Grond S, Radbruch L, Meuser T, Sabatowski R, Loick G, Lehmann KA. Assessment and
treatment of neuropathic cancer pain following WHO guidelines. Pain 1999;79:15-20.
2. Zech DF, Grond S, Lynch J, Hertel D, Lehmann KA. Validation of World Health
Organization Guidelines for cancer pain relief: a 10-year prospective study. Pain
1995;63:65-76.
3. Biscarini L. Non-steroidal anti-inflammatory drugs. In: Dukes NMG, Aronson JK (eds).
Meyler's Side Effects of Drugs. Amsterdam: Elsevier Science, 2000:246-309.
4. van der Klauw MM, Stricker BH, Herings RM, Cost WS, Valkenburg HA, Wilson JH. A
population based case-cohort study of drug-induced anaphylaxis. Br J Clin Pharmacol
1993;35:400-8.
5. van der Klauw MM, Wilson JH, Stricker BH. Drug-associated anaphylaxis: 20 years of
reporting in The Netherlands (1974-1994) and review of the literature. Clin Exp Allergy
1996;26:1355-63.
6. Settipane GA. Adverse reactions of aspirin and related drugs. Arch Intern Med 1981;141(3
Spec No):328-32.
7. Shelley ED, Shelley WB. Ibuprofen urticaria. J Am Acad Dermatol 1987;17:1057-8.
8. Ahmad S. Anaphylaxis from tolmetin. N Engl J Med 1980;303:1417.
9. Szczeklik A, Gryglewski RJ, Czerniawska-Mysik G. Clinical patterns of hypersensitivity to
nonsteroidal anti-inflammatory drugs and their pathogenesis. J Allergy Clin Immunol.
1977;60:276-84.
10. Stern RS, Bigby M. An expanded profile of cutaneous reactions to nonsteroidal anti-
inflammatory drugs. Reports to a specialty-based system for spontaneous reporting of
adverse reactions to drugs. JAMA 1984;252:1433-7.
11. Cistero A, Urias S, Guindo J, Lleonart R , Garcia-Moll M, Geli A, Bayes de Luna A.
Coronary artery spasm and acute myocardial infarction in naproxen-associated anaphylactic
reaction. Allergy 1992;47:576-8.
12. Inoue K, Motonaga A, Nishimura T, Yokota M, Miki N, Fujisawa H, Ueda F, Shibata Y,
Kimura K. Mechanism of anti-inflammatory action of etodolac. Arzneimittelforschung
1991;41:235-9.
13. Milman U, Hermoni D. Anaphylactic reaction to oral diclofenac sodium sustained-release
tablet. Isr J Med Sci 1994;30:909-10.
14. Moore ME, Goldsmith DP. Nonsteroidal anti-inflammatory intolerance. An anaphylactic
reaction to tolmetin. Arch Intern Med 1980;140:1105-6.
15. Alkhawajah AM, Eifawal M, Mahmoud SF. Fatal anaphylactic reaction to diclofenac.
Forensic Sci Int 1993;60:107-10.
78
16. Brogden RN, Heel RC, Pakes GE, Speight TM, Avery GS. Diclofenac sodium: a review of
its pharmacological properties and therapeutic use in rheumatic diseases and pain of varying
origin. Drugs 1980;20:24-48.
17. McKenna F. Efficacy of diclofenac/misoprostol vs diclofenac in the treatment of ankylosing
spondylitis. Drugs 1993;45(Suppl 1):24-30.
18. Dux S, Groslop I, Garty M, Rosenfeld JB. Anaphylactic shock induced by diclofenac. Br
Med J 1983;286:1861.
19. Levy JH, Shanewise JS. Anaphylaxis and coronary disease. N Engl J Med 1996;335:1925.
20. O'brien WM. Adverse reactions to nonsteroidal anti-inflammatory drugs. Diclofenac
compared with other nonsteroidal anti-inflammatory drugs. Am J Med 1986;80:70-80.
Monitoring, Uppsala, 1995.
22. WHO Collaborating Centre for Drug Statistics Methodology. July 1st 2001.
http://www.whocc.nmd.no/.
Lancet 2000;356:1255-9.
24. Stricker BHC, Tijssen JGP. Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;45:1177-84.
25. Finney DJ. Statistical aspects of monitoring for dangers in drug therapy. Methods Inf Med
1971;10:1-8.
26. Finney DJ. The Design and Logic of a Monitor of Drug Use. J Chron Dis 1965;18:77-98.
1998;54:315-21.
1997;44:513-8.
30. Kay AB. Allergy and allergic diseases: Allergic diseases and their treatment. N Engl J Med
2001;344:109-13.
31. Brankowski Z; Bruppacher R; Crusius Ieal. Reporting adverse drug reactions, Definitions of
Terms and Criteria for their Use. Geneva: Council for International Organizations of
Medical Sciences (CIOMS), 1999.
32. Murrant T, Bihari D. Anaphylaxis and anaphylactoid reactions. Int J Clin Pract 2000;54:322-
8.
33. Yunginger JW. Anaphylaxis. Ann Allergy 1992;69:87-96.
34. Insel PA. Analgesics-antipyretics and antinflammatory agents. In: Hardman JG, Limbird LE,
Molinoff BP, Ruddon RW, Goodman Gilman A (eds). Goodman & Gilman's The
Pharmacological Basis of Therapeutics. New York: McGraw-Hill, 1999:621.
35. Goetzl EJ, Payan DG, Goldman DW. Immunopathogenetic roles of leukotrienes in human
diseases. J Clin Immunol 1984;4:79-84.
79
Chapter 2.1
36. Miettinen OS. The need for randomization in the study of intended effects. Stat.Med
1983;2:267-71.
37. Meyboom RHB, Lindquist M, Egberts ACG. An ABC of drug-related problems. Drug Saf
2000;22:415-23.
38. Leufkens HG, Urquhart J, Stricker BH, Bakker A, Petri H. Channelling of controlled release
formulation of ketoprofen (Oscorel) in patients with history of gastrointestinal problems. J
Epidemiol Community Health 1992;46:428-32.
39. Toogood JH. Risk of anaphylaxis in patients receiving beta-blocker drugs. J Allergy Clin
Immunol 1988;81:1-5.
80
2.2
SIGNALLING POSSIBLE DRUG-DRUG INTERACTIONS IN A

SPONTANEOUS REPORTING SYSTEM: DELAY OF
WITHDRAWAL BLEEDING DURING CONCOMITANT USE OF
ORAL CONTRACEPTIVES AND ITRACONAZOLE
EP van Puijenbroek1, ACG Egberts2,3, RHB Meyboom1, and HGM Leufkens2
1
2
Utrecht Institute for Pharmaceutical Sciences, Department of Pharmacoepidemiology and
Pharmacotherapy, Utrecht, the Netherlands
3
BRITISH JOURNAL OF CLINICAL PHARMACOLOGY 1999;47:689-693

Chapter 2.2
Summary
Objective In spontaneous adverse drug reaction reporting systems, there is a

growing need for methods facilitating the automated detection of signals
concerning possible adverse drug reactions. In addition, special attention is
needed for the detection of adverse drug reactions resulting from possible drug-
drug interactions. We describe a method for detecting possible drug-drug
interactions using logistic regression analysis to calculate ADR Reporting Odds
Ratios.
Methods To illustrate this method, we analysed the adverse drug reaction
‘delayed withdrawal bleeding’ resulting from a possible interaction between
itraconazole and oral contraceptives in reports received by the Netherlands
Pharmacovigilance Foundation Lareb between 1991 and 1998.
Results In total 5,503 reports were included in the study. The ADR Reporting
Odds Ratio, adjusted for year of reporting, age and source of the reports, for a
delayed withdrawal bleeding in women who used both drugs concomitantly
compared with women who used neither oral contraceptives, nor itraconazole,
was 85 (95% Confidence Interval 32-230).
Conclusion Since spontaneous reporting systems can only generate signals
concerning possible relationships, this association needs to be analysed by other
methods in more detail in order to determine the real strength of the relationship.
This approach might be a promising tool for the development of procedures for
automated detection of possible drug-drug interactions in spontaneous reporting
systems.
82
Drug-drug interaction between itraconazole and oral contraceptives
Introduction
The Netherlands Pharmacovigilance Foundation Lareb maintains the spontaneous

adverse drug reaction reporting system in the Netherlands.1 Since 1996 Lareb is
the national centre in support of the Medicines Evaluation Board. One of the
major goals of this system is the timely detection of possible new adverse drug
reactions (ADRs) and interactions. These signals are currently being detected by a
systematic review of all individual reports that are received in a certain time span
by a team of experts in this field. With the increasing number of reports, there is a
growing need for a computerised system that aids in this process.
Special cases are those ADRs resulting from drug-drug interactions. Reports to
Lareb usually concern an event ascribed to only one suspected drug. Rarely the
presence of a suspected interaction is mentioned, which makes the detection of
possible drug-drug interactions difficult. The detection of possible interactions is
based on the following concept: when a suspected ADR is reported more
frequently on the combination of two drugs as compared with the situation where
these drugs are used in absence of one each other, this association might indicate
the existence of a drug-drug interaction.
In 1974 Finney described the concept of ‘reaction proportion signalling’. Based
upon all records in the database, contingency tables can be constructed and
subsequently disproportional drug-ADR combinations are identified, and used for
the detection of signals.2 By means of these contingency tables, ‘ADR Reporting
Odds Ratios’ can be calculated. This concept, which was previously described for
analyzing reported ADRs in a case non-case design,3,4 is applied for the analysis
of possible interactions in this study.
Since the approval for marketing of itraconazole in the Netherlands, 19 reports of

delayed withdrawal bleeding in women using OCs have been received. In 10 of
these reports OCs and itraconazole were used concomitantly. Nine of these reports
concerned OCs containing desogestrel. One other woman used a cyproterone-acetate
containing OC and experienced a delayed withdrawal bleeding. Because of the
unexpected high number of reports concerning a delay in the withdrawal bleeding
where OCs and itraconazole were used concomitantly, our analysis was focussed on
this association. Clinical details of this possible interaction were recently described
as a case series on the basis of reports submitted to Lareb.5,6 The objective of the
83
Chapter 2.2
present article was to illustrate this method by analyzing a possible interaction

between oral contraceptives (OCs) and the antifungal agent itraconazole.
Methods
Source
Health professionals send reports to Lareb on a voluntary basis, by means of a
limited number of questions on a special report form. Information is collected
about the patient and the suspected ADR. Also detailed information is requested
about all drugs that were used at the time the reaction occurred. After being
received by Lareb, information such as ADR(s) and indication for use is coded,
and subsequently reports are filed in the Lareb database. This database can be
used for further evaluation and analysis of the data. On 1 January 1998, it
consisted of a total of approximately 19,500 reports.
Selection of cases and non-cases

The domain was restricted to reports of women between the age of 15 and 50
years. In order to make an appropriate comparison between women who used
itraconazole, which was licensed in the Netherlands in 1991, and women who did
not use this antifungal drug, only reports with a reporting date later than January
1991 were included. In February 1998 we published an article concerning the
possible association between the use of itraconazole and oral contraceptives in
one of the major Dutch Journals.5 To avoid the risk of reporting bias due to this
publication, only reports received before January 1998 were used.
The following data were used in the study:
-Source: A report might originate from either a physician or a pharmacist.
-Age of the patients.
-Reported reactions: All menstruation disorders and disorders of the withdrawal
bleeding in the database were checked and, if necessary, recoded. If the WHO
preferred term ‘menses onset delayed’ or ‘missed withdrawal bleeding’ was
recorded on a report form, the ADR was labelled as ‘delayed bleeding’. After
checking the original description of the ADR, also some cases of ‘amenorrhoea’
or ‘oligomenorrhoea’ were recoded as ‘delayed bleeding’. In all other cases, the
ADR was coded as ‘other ADRs’. One report of a woman reporting a ‘delayed
84
withdrawal bleeding’ in association with an initial transient positive pregnancy

test was coded as ‘other ADR’, because a miscarriage could not be ruled out.
-Year of reporting.
-Presence of itraconazole among the drugs used.
-Presence of OCs among the drugs used.
-Although on the forms received by Lareb, the health professional makes a
distinction between suspected and non-suspected medication, all drugs were taken
into account in the present study.
Cases were defined as patients who reported a ‘delayed bleeding’, while non-
cases consisted of all other reports. The reports were divided into three index
groups:
-Reports of women who used OCs, but no itraconazole at that time.
-Reports of women who used itraconazole, but no OC at that time.
-Reports of women who used itraconazole and OC concomitantly.
The reference group consisted of patients who used neither OCs, nor itraconazole.
Analysis
For the comparison of index and reference groups, ADR Reporting Odds Ratios
were calculated. These ratios are defined as the ratio of the exposure odds of
reported cases of ‘delayed bleeding’ to the exposure odds of other ADRs. The
ADR Reporting Odds Ratios were calculated as (a:c)/(b:d) (Figure 1) (4).
Reported suspected ADR

'Delayed bleeding' No 'delayed bleeding'
reported reported
Medication Index group a b

Reference group c d
Figure 1. Two by two table used for the calculation of ADR Reporting Odds Ratios.
ADR Reporting Odds Ratios were adjusted for source, year of reporting and age,
and calculated by logistic regression analysis. Odds ratios were expressed as point
estimates with 95% confidence intervals. For the comparison of the age of the
patients between the three groups, analysis of variance was conducted. In order to
85
Chapter 2.2
analyze the source of the reports a Chi-square test was used. All statistical
analyses were performed using SPSS 8.0.
Results
A total number of 5,503 reports were included in the study. Altogether 1,466
women used OCs but no itraconazole; 16 women used itraconazole but no OCs,
and 23 women used both drugs. Mean age, source and number of reports for
index and reference groups that were studied are presented in Table 1. There was
a difference between these groups concerning the age of the patients (p<0.001
analysis of variance), as well as the source of the reports ( p<0.001 Chi-square).
Table 1. Age and source of reporting of index groups, reference group and total group
Index groups Reference Total
A B C group
Only oral Only Itraconazole as Neither

contraceptive itraconazole well as oral itraconazole
contraceptive nor oral
contraceptives
Mean age 31.1 40.8 29.7 37.2 35.5

(years)
Source 71.6 43.7 56.5 77.3 75.6

(percentage
reporting by
physicians)
Number of 1,466 16 23 3,998 5,503

reports
Differences in reported ADRs between the three groups are shown in Table 2.
Delayed bleeding was reported by 10 patients who used itraconazole and OCs
concomitantly. The other 13 patients in this group did not mention this side effect.
If only OCs were used, 9 out of 466 patients mentioned ‘delayed bleeding’. If
only itraconazole was used none out of 16 patients reported the occurrence of this
effect.
86
Table 2. Reporting Odds Ratios (OR) concerning the occurrence of a ‘delayed bleeding’
of reference and index groups.
Cases Controls crude OR Adjusted* OR
n=39 n=5,464 (95% CI) (95% CI)
No itraconazole or OC 20 3978 1 1
OC/ no itraconazole 9 1457 1.2 (0.6-2.7) 0.8 (0.4-1.8)
Itraconazole/ no OC 0 16 0 (0-∞) 0 (0-∞)
Concomitant use of 10 13 153 (60-389) 85 (32-230)
itraconazole and OC
*adjusted for year of reporting, age and source
The ADR Reporting Odds Ratios and 95% confidence intervals for all
comparisons are shown in Table 2.
Adjusted for year of reporting, age and source, the ADR Reporting Odds Ratio for
the risk of a ‘delayed bleeding’ in reports of women who used itraconazole and
OCs concomitantly compared with the reports of women who used neither of
these drugs was 85 (95% CI 32-230). The ADR Reporting Odds Ratio adjusted
for year of reporting, age and source in the group where only OCs were used was
0.8 (95% CI 0.4-1.8). When only itraconazole was used, there were no cases.
Discussion
A major function of spontaneous reporting systems is generating new ‘signals’.7

For this purpose, all reports are individually reviewed and detection depends upon
the skills and memory of the professionals involved. Since this procedure is time
consuming and the number of reports is increasing, there is a growing need for an
automated method that facilitates detection of ADRs. A special case is the
detection of interactions, since they are often not reported. The analysis of
possible drug-drug interactions is based on the concept that a suspected ADR is
more often reported on the combination of two drugs compared to the situation
where either of these drugs has been used in absence of the other one. The level of
the odds ratio however, is of less importance, but provides an indication of the
degree of disproportionality. Odds ratios tend to overestimate the relative risk,8 so
interpretation is more difficult. However, the use of an odds ratio offers
87
Chapter 2.2
advantages since concomitant variables may be analysed by logistic regression

analysis.
Possible drug-drug interactions might be analysed in different ways. In the first
place, as was shown in this study, three index groups can be distinguished; two
groups where both drugs are used in absence of one each other and one group
where both drugs are used concomitantly. These index groups can be compared
with a reference group in which none of the suspected drugs is used. The
likelihood of a reaction being due to a suspected interaction is increased if the
95% confidence intervals of the group in which both drugs are used and the other
two index groups are mutually exclusive.
Secondly, the analysis could also have been done by using a logistic model in
which next to the use of both itraconazole and oral contraceptives as covariates
subsequently an interaction term for the concomitant use of both drugs is added.
In our example the logistic model for calculating the adjusted odds ratios would
then look like:
log(odds) = b0 + b2A + b3S + b4Y + b5I + b6OC + b7OC*I

(A = age, S= source, Y= reporting year, I= Itraconazole, OC= oral contraceptive)
The estimates resulting from this model does not differ from the previous method.
This model can be compared with the model in which no interaction term (b7-
OC*I) is present. A likelihood ratiotest can be used to evaluate the effect of
adding this term. Given the fact that the difference in –2 log likelihood follows a
chi-square distribution with one degree of freedom, adding the interaction term is
this case was statistically significant (p< 0.01).
The latter method can also be used for the automatic detection of drug
interactions. The use of a computer in the detection of adverse drug interactions
was already proposed by Amery.9 In his approach the computer is instructed to
look for pairs of a specific concomitant drug with a specific ADR. The fraction of
patients showing the ADR under consideration is subsequently compared with the
frequency of this ADR in all database patients. In our approach, the use of logistic
regression analysis offers the possibility for correcting for covariates, and
furthermore offers a possibility of analysing the use of interaction terms in more
detail. As preliminary experiments indicate, analysing the effect of adding an
88
interaction terms by a likelihood ratiotest is a promising tool in the automated

detection of drug interactions and will be further explored.
Internal validity
In general, as with the analysis of adverse drug reactions, selection bias might be
present, since certain ADRs are more likely to be reported than others. In our
example selection bias might be present in the index groups where OCs are used in
contrast to the index group where no OCs are used. In the latter situation, a delay in
the onset of the menstruation with a couple of days might be reported less likely
because of acceptance of the normal physiological irregularity of the menstruation
women might experience. A delay in the onset of the withdrawal bleeding during the
use of OCs, however, might be experienced as a more unusual situation. Despite the
fact that a delayed withdrawal bleeding rarely occurs with oral contraceptives this
‘event’ might be reported more frequently. This selection bias can cause an
overestimation of the odds ratio for the occurrence of a ‘delayed bleeding’ in reports
were OCs are used. Selection bias, however, is not likely to have played a more
prominent role in cases where both OCs and itraconazole were used than in cases
where only OCs were used.
Increased reporting due to a recent introduction of a drug or attention for an
adverse drug reaction in the media does not necessarily influence the Reporting
Odds Ratio, since non selective reporting bias has a similar effect on both
numerator and denominator.3 If there is a specific attention for an interaction
however, reporting bias might occur.
Another point of attention is confounding. For instance, a third drug may act as a
confounder when it is associated with one of the drugs or the concomitant use of
both suspected drugs. For instance analgesics such as naproxen may cause a delay
or interrupt menstruation in women who are not taking oral contraceptives.10
NSAIDs were not prescribed to any of the case patients reported to Lareb, but in
the Netherlands these drugs are also available for self-treatment. The indication
for the use of itraconazole, mycotic infections, is seldom painful. This implies that
there is little reason for the use of NSAIDs, which might influence the occurrence
of a withdrawal bleeding. Although it is unlikely that self-medication could have
acted as a confounder in these reports, the use of NSAIDs for self-treatment could
be definitely ruled out in only one case. None of the patients in the group who
used both drugs, also used NSAIDs.
89
Chapter 2.2
External validity
Like the analysis of ADRs in a spontaneous reporting system, detection and
confirmation of an association does not imply a causal relationship. Additional
studies need to be done to confirm a true pharmacodynamic or pharmacokinetic
interaction and to determine the strength of this association in daily practice.
Because of the various degrees of underreporting, that are inextricably bound up
with spontaneous reporting,11-13 it is difficult to estimate the true incidence of a
possible interaction. In our example where the signal power strongly points at a
true interaction, this estimation was not made, since data concerning the degree of
underreporting and prescription rates of drug are not available.
Itraconazole and oral contraceptives

The concomitant use of itraconazole and oral contraceptives might lead to an
increase in reports of a ‘delayed bleeding’. The chance of this ADR being
reported was not increased in women who used itraconazole or an OC alone. The
number of patients who used itraconazole in the absence of OCs is rather small.
Since there were no cases, the odds ratio must be zero, but the corresponding
confidence interval was large. Theoretically, the delay in withdrawal bleeding
might have been caused by the use of itraconazole, but the reporting rate is much
higher when oral contraceptives were used concomitantly. For this reason, the
strong association between the concomitant use of itraconazole and OCs and the
occurrence of a ‘delayed withdrawal bleeding’ therefore is suggestive of an
interaction.
The mechanism underlying this suspected interaction remains to be explained.
Itraconazole interacts with a variety of drugs.14 Triazole compounds such as
ketoconazole and itraconazole in vitro inhibit cytochrome P450-3A, an enzyme
system involved in the biotransformation of steroid hormones.15,16 An increase in
the levels of steroids might lead to a delay of the withdrawal bleeding. Of the ten
women who reported this suspected side effect, nine women used a desogestrel
containing OC, and one woman a cyproterone acetate containing OC. The reason
for the high number of desogestrel containing OCs is not clear.
For our analysis we focussed on a comparison of the chances of ‘events’ being
reported in the different groups. Nevertheless, different mechanisms may be
responsible for the ‘delayed bleeding’ in these two groups. If a woman uses
itraconazole without an oral contraceptive, the ‘delayed bleeding’ is in fact a
90
delayed menstruation. In women using oral contraceptives, it refers to a delayed

withdrawal bleeding. Since it is not clear if different mechanisms underlie these
suspected ADRs, both ADRs were treated as one variable in our analysis.
Conclusion
In a spontaneous reporting system the automatic analysis of reports is usually

limited to adverse drug reactions. In the method described, the index and
reference groups originating from the database can be used to signal possible
drug-drug interactions. In our example, the results strongly suggest that in women
using OCs and itraconazole concomitantly the withdrawal bleeding may be delayed
due to a drug interaction, and illustrates that the method described might be a useful
tool in analyzing possible interactions in spontaneous reporting systems.
References
1. De Koning GHP. A regionalized spontaneous surveillance program for adverse drug
reactions as tool to improve pharmacotherapy. (Thesis) Utrecht University, Utrecht, the
Netherland, 1994.
2. Finney DJ. Systematic signalling of adverse reactions to drugs. Meth Inform Med
1974;13:1-10.
1992;45:1177-84.
5. Van Puijenbroek EP, Feenstra J, Meyboom RHB. Verstoring van de pilcyclus tijdens het
gelijktijdig gebruik van itaconazol en orale anticonceptiva. Ned Tijdschr Geneeskd
1998;142:146-9.
6. Meyboom RHB, Van Puijenbroek EP, Vinks MHAM, Lastdrager CJ. Disturbance of
withdrawal bleeding during concomitant use of itraconazole and oral contraceptives. N Z
Med J 1997;110:300.
7. Carson JL, Strom BL, Maislin G. Screening for unknown effects of newly marketed drugs.
In: Pharmacoepidemiology. New York: John Wiley & Sons, 1994:431-47.
8. Davies HT, Crombie IK, Tavakoli M. When can odds ratios mislead? BMJ 1998;316:989-
91.
9. Amery W. Analysis of the information in a central ADE database. Intern J Risk Safety in
Med 1993;5:123.
91
Chapter 2.2
10. Meyboom RHB, Heymeijer GWJ, Van den Bemt PMLA, de Koning GHP. Disturbance of
menstruation as a side effect of nonsteroidal antiinflammatory drugs (NSAIDs).
11. Milstien JB, Faich GA, Hsu JP. Factors affecting physician reporting of adverse drug
reactions. Drug Inf J 1986;20:157-64.
1993;7:119-37.
13. Tubert-Bitter P, Begaud B, Moride Y, Chaslerie A, Haramburu F. Comparing the toxicity of
two drugs in the framework of spontaneous reporting: a confidence interval approach. J Clin
Epidemiol 1996;49:121-3.
14. Stockley IH. Drug interactions. Oxford: Blackwell Science, 1996.
15. Back DJ, Tjia JF. Azoles and allylamines: the clinical implications of interaction with
cytochrome P450 enzymes. J Dermatol Treatment 1990;1:11-3.
16. Baldwin SJ, Bloomer JC, Smith GJ, Ayrton AD, Clarke SE, Chenery RJ. Ketoconazole and
sulfaphenazole as the respective selective inhibitors of P450 3A and 2C9. Xenobiotica
1995;25:261-70.
92
2.3
DETECTING DRUG-DRUG INTERACTIONS USING A

DATABASE FOR SPONTANEOUS ADVERSE DRUG
REACTIONS: AN EXAMPLE WITH DIURETICS AND NON-
STEROIDAL ANTI-INFLAMMATORY DRUGS
EP van Puijenbroek1, ACG Egberts2,3, ER Heerdink2 and HGM Leufkens2
1
2
3
EUROPEAN JOURNAL OF CLINICAL PHARMACOLOGY 2000;56:733-8

Chapter 2.3
Summary
Objective Drug-drug interactions are relatively rarely reported to spontaneous

reporting systems (SRSs) for adverse drug reactions. For this reason, the
traditional approach for analysing data from SRS has major limitations for the
detection of drug-drug interactions. We developed a method that may enable
signalling of these possible interactions, which are often not explicitly reported,
utilising reports of adverse drug reactions in data sets of SRS. As an example, the
influence of concomitant use of diuretics and non-steroidal anti-inflammatory
drugs (NSAIDs) on symptoms indicating a decreased efficacy of diuretics was
examined using reports received by the Netherlands Pharmacovigilance
Foundation Lareb.
Methods Reports received between 1 January 1990 and 1 January 1999 of
patients older than 50 years were included in the study. Cases were defined as
reports with symptoms indicating a decreased efficacy of diuretics, non-cases as
all other reports. Exposure categories were the use of NSAIDs or diuretics versus
the use of neither of these drugs. The influence of the combined use of both drugs
was examined using logistic regression analysis.
Results The odds ratio of the statistical interaction term of the combined use of
both drugs was increased (adjusted odds ratio 2.0, 95% confidence interval 1.1-
3.7), which may indicate an enhanced effect of concomitant drug use.
Conclusion The findings illustrate that spontaneous reporting systems have a
potential for signal detection and the analysis of possible drug-drug interactions.
The method described may enable a more active approach in the detection of
drug-drug interactions after marketing.
94
Drug-drug interaction between diuretics and NSAIDs
Introduction
Since the early 1960s, spontaneous reporting systems (SRSs) have been used to
detect adverse drug reactions (ADRs) after marketing of drugs. Nowadays, these
reporting systems play a major role in pharmacovigilance.1 Since the size of data
sets is increasing, automated signal generation may be a promising tool for
selecting possible combinations of ADRs and drugs that might be worthwhile
analysing in more detail.2-4 For signal detection concerning possible unexpected
ADRs, various measures of disproportionality can be used, including Reporting
Odds Ratios (RORs).5,6
The basic principle of looking for disproportionality can be extended to the
detection of drug-drug interactions, which are generally more difficult to detect.
Usually a drug-drug interaction might be suspected in the event that similar
substances have proven to cause a similar interaction. If this is not the case,
detection becomes more complicated. In the event a drug-drug interaction is
unexpected given the previous knowledge, it is rarely reported to an SRS.
Furthermore, in individual patients, it is usually not clear whether ADRs arise
directly from the use of a certain drug or that the ADR concerned is in fact the
result of an underlying pharmacodynamic or pharmacokinetic drug-drug
interaction. In particular in the elderly, additional factors such as co-morbidity
and multiple drug use may be present, enhancing the number of possible
explanations of a certain unexpected clinical event. In pre-marketing trials,
patients with multiple drug use are usually excluded, which makes the detection
of drug-drug interactions in the post-marketing period even more important.
In the event of a drug-drug interaction, one drug influences the effect of another
drug. This may subsequently cause an increase or decrease in the number of
reported ADRs of the latter drug. By analysing individual reports, it is usually
difficult to recall whether specific concomitant medication was also used in
similar reports. An automated statistical approach may be helpful in analysing
large numbers of these reports and in revealing the existence of these complex
relationships. The influence of the combined use of drugs can be studied by
introducing a statistical interaction term in a logistic model for the calculation of
RORs, as was shown in a previous study by our group.7
As an example of the approach described, in reports received by the Netherlands
Pharmacovigilance Foundation Lareb, the concomitant use of both diuretics and
95
Chapter 2.3
non-steroidal anti-inflammatory drugs (NSAIDs) was associated with the

occurrence of symptoms that may point towards a reduction of the therapeutic
effects of diuretics, This drug-drug interaction has been described in several case-
reports and studies.8-21
Methods
Setting and design

The Netherlands Pharmacovigilance Foundation Lareb is the national centre for
submitting spontaneous reports of suspected ADRs originating from health care
professionals in the Netherlands.22 These reports are considered to be a reflection
of the ADRs that occur in daily practice, taking into account the various degrees
of underreporting that are an inherent attribute of spontaneous reporting.23-26 The
analysis of drug-drug interactions is based on the assumption that specific ADRs
may occur more frequently when both drugs are used concomitantly in
comparison with separate use.7,27 This specific increase in reports is assumed to be
reflected in the number of reports to Lareb. All reports submitted to Lareb
between 1 January 1990 and 1 January 1999 of patients older than 50 years were
included in the analysis. Data concerning ages and genders of the patients had to
be available.
Selection of cases and non-cases

A decrease in the efficacy of diuretics may express itself as the occurrence of
oedema or signs indicating the onset or worsening of congestive heart failure
(CHF). After being received by Lareb, the reported possible ADR is coded by a
qualified assessor using the World Health Organization (WHO) Adverse Drug
Reaction Terminology.28 The presence of one or more of the following WHO
preferred terms on the reports was therefore considered as an indication for this
situation: ‘oedema’, ‘oedema dependent’, ‘oedema generalised’, ‘oedema
peripheral’, ‘cardiac failure’, ‘cardiac failure left’, ‘cardiac failure right’,
‘pulmonary oedema’ and ‘oedema legs’. Reports that mentioned one of the
aforementioned ADRs were defined as cases. All other reports were defined as
non-cases.
96
Exposure categories
Information about concomitant drug use is requested on the reporting forms. For
the vast majority of the reports Lareb has the patients’ drug dispensing history
from community pharmacies. All drugs in use at the moment the ADR occurred
were considered possible causes of the ADR. If a drug was used to treat the ADR
it is not listed as concomitant medication. The reporting health professionals give
an indication which drug is considered the suspected drug. Usually, however,
only one drug is indicated. Since we were looking for drug-drug interactions, we
therefore made no distinction between suspected and non-suspected medication.
All medication that the patient was using according the medication history on the
calendar date of the event was considered.
Exposure categories were the use of NSAIDs (WHO Anatomical Therapeutic
Chemical (ATC) classification code M01A), or diuretics (ATC code C03) versus
the use of neither of these drugs. Covariates used in the analysis were: type of
health professional that reported the ADR (either pharmacist or physician), year
of reporting, age and gender of the patient involved, the use of antidiabetic drugs
(ATC code A10), cardiac glycosides (ATC code C01), antihypertensive drugs
(ATC code C02), peripheral vasodilatating drugs (ATC code C04), b-blocking
agents (ATC code C07), calcium channel blocking agents (ATC code C08) and
drugs acting on the renin angiotensin aldosterone system (RAAS, ATC code
C09).
Statistical analysis
For the analysis, the following logistic model was used:
log(odds) = b 0 + b1N + b2D + b3N*D + bn-xCn-x

where N= NSAIDs, D= diuretics, Cn-x = different covariates, i.e. age, source and reporting
year.
A statistically significant value of the interaction term b3 indicates an additional

effect of concomitant use of diuretics and NSAIDs. Probability (P) values of 0.05
or less were considered statistically significant. For all analyses the statistical
package SPSS 8.0 was used.
97
Chapter 2.3
Results
From January 1990 until January 1999, Lareb received 9,907 reports of patients
aged over 50 years. Eighty-five reports were excluded because age or gender of
the patient was not known. A total of 9,822 reports, which were separated in 305
cases and 9,517 non-cases, were included in the analysis. Characteristics of cases
and non-cases concerning age, gender, source of the reports and the use of several
cardiac drugs are provided in Table 1.
Table 1. Characteristics of cases and non-cases

Cases Non-Cases Odds ratio
n (%) n (%) (95% CI)
n=305 n=9517
Mean age (years) 67.4 65.9

Females 224 (73.4%) 5848 (61.4%) 1.73 (1.34-2.24)
Reports by physicians 210 (68.9%) 6220 (65.5%) 1.17 (0.92-1.50)
NSAIDs 67 (22.0%) 1546 (16.2%) 1.45 (1.11-1.91)
Diuretics 78 (25.6%) 1697 (17.8%) 1.58 (1.22-2.06)
Insulin and oral 17 (5.6%) 656 (6.9%) 0.79 (0.49-1.31)
antidiabetic drugs
Cardiac glycosides 43 (14.1%) 1185 (12.5%) 1.15 (0.83-1.60)
Antihypertensive drugs 14 (4.6%) 209 (2.2%) 2.14 (1.23-3.73)
Peripheral vasodilatating 3 (1.0 %) 79 (0.8%) 1.19 (1.37-3.78)
drugs
Beta-blocking agents 66 (21.6%) 1727 (18.1%) 1.25 (0.94-1.64)
Calcium antagonists 127 (41.6%) 1246 (13.1%) 4.74 (3.74-6.00)
Drug acting on the 49 (16.4%) 1717 (18.0%) 0.87 (0.64-1.19)
RAAS system
NSAIDs non-steroidal anti-inflammatory drugs; 95% CI 95% confidence interval; RAAS

renin angiotensin aldosterone system
Among the cases, the number of females and the age of the patients (p<0.01, t-
test) is significantly higher. Also the use of diuretics, NSAIDs, antihypertensive
drugs and calcium channel blocking agents was more frequent among the cases.
There were no differences concerning the use of insulin and oral antidiabetic
drugs, cardiac glycoside drugs, peripheral vasodilatating drugs, beta blocking
agents, and drugs acting on the RAAS in the medication history.
98
Among the cases, the following suspected ADRs were mentioned on the reporting
forms: oedema (n=68), oedema dependent (n=90), oedema generalised (n=14),
oedema peripheral (n=45), cardiac failure (n=14), cardiac failure left (n=6),
oedema pulmonary (n=3) and oedema legs (n=66). The WHO term ‘cardiac
failure right’ was not noted on the report forms. Since more than one ADR can be
attributed to one report, the total number of suspected ADRs exceeds the number
of cases.
The distribution of drugs present in cases and non-cases, stratified for the use of
NSAIDs and diuretics, is shown in Table 2. The corresponding odds ratios with
95% confidence intervals are also shown. If no NSAID was used, the crude odds
ratio for the use of diuretics was 1.29 (95% CI 0.95-1.77). If an NSAID was used
by the patient, the crude odds ratio was 3.03 (95% CI 1.82-5.08).
Table 2. Distribution of drugs present in cases and non-cases, stratified for the use of non
steroidal anit-inflammatory drugs (NSAIDs) and diuretics
Cases Non- Total OR
n=305 cases (95% CI)
n=9,517
NSAID not Diuretic not present 185 6527 6712

present
Diuretic present 53 1444 1497 1.29
(0.95-1.77)
Total 238 7971 8209
NSAID Diuretic not present 42 1293 1335

present
Diuretic present 25 253 278 3.03
(1.82-5.08)
Total 67 1546 1613
95% CI 95% confidence interval; OR odds ratio
Crude and adjusted odds ratios and 95% confidence intervals are shown in Table
3. The use of diuretics or NSAIDs alone is not statistically significant as an
increased risk for onset or worsening of symptoms of CHF or oedema. However,
the odds ratio of the statistical interaction term NSAIDs*diuretics is statistically
significant (adjusted odds ratio 2.0, 95% CI 1.1-3.7). This is an indication for an
99
Chapter 2.3
enhanced chance of cases being reported, associated with the combined use of
both drugs.
Table 3. Association between non-steroidal anti-inflammatory drugs (NSAIDs), diuretics

and covariates on the occurrence of symptoms of congestive heart failure (CHF) during
the use of NSAIDs and diuretics.
Covariates Adjusted Unadjusted
odds ratio* odds ratio
(95% CI) (95% CI)
NSAIDs (ATC code M01A) 1.22 (0.86-1.73) 1.15 (0.82-1.61)

Diuretics (ATC code C03) 1.05 (0.75-1.46) 1.29 (0.95-1.77)
Interaction term NSAIDs*Diuretics 2.00 (1.08-3.69) 2.35 (1.29-4.28)
Age (year) 1.01 (1.00-1.02)

Gender (female) 1.80 (1.38-2.34)
Source (physician) 0.93 (0.71-1.22)
Year of reporting 0.93 (0.88-0.98)
Antidiabetic drugs (A10) 0.75 (0.45-1.25)
Cardiac glycosides (C01) 0.68 (0.48-0.98)
Antihypertensives (C02) 1.94 (1.09-3.45)
Peripheral vasodilat. drugs (C04) 0.85 (0.25-2.86)
b-blocking agents (C07) 1.06 (0.79-1.41)
Calcium antagonists (C08) 5.25 (4.09-6.75)
Drugs acting RAAS system (C09) 0.83 (0.60-1.15)
*Odds ratio adjusted for age, gender, source, year of reporting, and the use of antidiabetic
drugs, cardiac glycosides, antihypertensive drugs, peripheral vasodilatating drugs, b-
blocking agents, calcium-antagonists, and drugs acting on the renin angiotensin
aldosterone system (RAAS). 95% CI 95% confidence interval; ATC anatomical
therapeutic chemical
Discussion
Methodology
In the event drug-drug interactions had not been previously observed clinically or
experimentally, the detection in the post-marketing phase is troublesome. We
developed a method by which drug-drug interactions, which were not reported
explicitly, can be detected using a spontaneous reporting system. Although this
method was developed to detect unknown drug-drug interactions, in this case the
100
well-known interaction between diuretics and NSAIDs was used to illustrate this
new approach.
In contrast to the exposure categories diuretics or NSAIDs, the odds ratio of the
statistical interaction term of the combined use of both drugs was increased,
which may indicate an enhanced effect of concomitant drug use. As shown in
Table 2, the ROR expressing the magnitude of the association between diuretics
and signs of oedema or CHF increased when NSAIDs were used among the
concomitant medications. This odds ratio can be also be calculated as the product
of the ROR of diuretics (1.29) and the ROR of the interaction term (2.35).
The analysis rests upon the assumption that the data set of the SRS is a close
representative of the event rate occurring in the population of drug users. In this
way, the calculated odds ratios approximate the incidence rate of the ADRs in the
population of drug users. The observation of a disproportionate association
relative to the whole database, even if different adjustments are made, does not
imply a causal relationship but suggests an association and serves as a starting
point for further analysis. The methodology is to be used for signalling
interactions and not for signal testing. Interpretation of the data should be done
with great trepidation due to the spontaneous character of the reports giving rise
to all possible sources of bias. If there is a specific interest in an interaction for
instance, a reporting bias might occur. None of the original reporting forms,
however, mentions a suspicion of a possible drug-drug interaction. Increased
reporting due to a recent introduction of a drug or attention for an ADR in the
media does not necessarily influence the Reporting Odds Ratio since non-
selective reporting bias has a similar effect on both numerator an denominator.5
Another point of attention is confounding bias. For instance, a third drug may act
as a confounder when it is associated with one of the drugs or the concomitant use
of both suspected drugs. The interaction term NSAIDs*diuretics also might have
been increased due to confounding if a statistical interaction existed between the
use of NSAIDs and a pre-existing CHF, for which diuretics were used to treat this
condition. Therefore, in a separate analysis, the interaction terms
NSAIDs*cardiac glycosides and NSAIDs*ACE-inhibitors were added to the
logistic model. The adjusted odds ratios for the latter interaction terms were not
significantly increased, whereas the adjusted odds ratio for the interaction term
NSAIDs*diuretics was 2.0 (95% CI 1.0-3.7). These findings are supportive for an
101
Chapter 2.3
actual interaction between diuretics and NSAIDs and not for confounding due to
an interaction between the use of NSAIDs and a pre-existing cardiac failure.
Different covariates were used in the analysis. Since pre-existing impairment of
kidney function, enhancing the risk of using NSAIDs, for instance in case of
diabetes mellitus and hypertension, cannot be excluded,29 both antidiabetic drugs
and antihypertensive agents were used as covariates. The use of NSAIDs alone (in
the absence of the use of diuretics) is also associated with signs of CHF20,30,31 and
oedema is a common ADR of these drugs.32 The use of calcium antagonists is
also associated with leg oedema.33 In our study, the influence of calcium
antagonists on the occurrence of symptoms of CHF, mainly oedema, was evident.
If an additional analysis is done after exclusion of all reports that mention the
presence of one of the calcium antagonists in the medication history, the adjusted
odds ratio of the interaction term NSAIDs*diuretics was still 2.6 (95% CI 1.2-
5.3).
The source of the reports was also used as a covariate, since physicians might
report signs of CHF more often in comparison to pharmacists, however,
pharmacists might be more familiar with the drug-drug interaction under
investigation. In the logistic model we corrected for age of the patient, since the
change on CHF increases in elderly patients. In this age group the use of NSAIDs
also might increase, i.e. because of rheumatological disorders.
Because of the various degrees of underreporting that are inherent to SRS, it is
difficult to estimate the true incidence of a possible interaction.
NSAIDs and diuretics

Several reports in literature suggest that concomitant use of diuretics and NSAIDs
might lead to a decrease in the effect of the diuretic drug.8-21 This applies to
sulphonamides as well as thiazide diuretics. Because of the small number of
reports with combined use of NSAIDs and diuretics, we did not made a
distinction between different classes of NSAIDs and diuretics. In larger databases
that also offers the possibility of the use of concomitant medication, studying the
different classes of NSAIDs would be possible.
In this study we used a limited number of ADR codes that might indicate a
decrease in the effect of diuretics. Since the starting point in the analysis was a
reduction in the effect of diuretics, the analysis was not restricted to oedema only.
Nevertheless, the number of cases that were coded as cardiac failure was rather
102
low (6.5%). This might be partially related to the original descriptions of the
reporting health professional, which sometimes hampers assigning an appropriate
diagnosis. In such cases only the symptoms can be coded. Since it was not always
clear when codes for dyspnoea and related pulmonary problems should have been
attributed to respiratory or cardiological problems, they can not be used for
automated signal detection. In our database, the majority of cases of dyspnoea is
used for shortness of breath resulting from a pulmonary cause. Using the code
dyspnoea in the analysis will therefore result in a dilution of the effect.
Practical considerations of detecting drug-drug interactions

Detecting drug-drug interactions in data sets of SRS implies a meticulous process.
In theory a large number of possible drug-drug interactions is possible. For
various combinations, either selected at random, or selected based on a related
drug-drug interaction, the interaction term of the covariates involved can be
derived. For this reason, analysis of drug-drug interactions should be developed as
an automated process. The first selection is based on the presence of a
disproportionate increase in the number of reports of a certain ADR in association
with a combined drug use (unadjusted odds ratio). This step might be executed
automatically by a computer programme. A second step is the refinement of the
signal using a dedicated correction for possible confounders present in the
database (adjusted odds ratio). In general, a sufficient number of cases where both
drugs are used concomitantly is a prerequisite for detecting interactions in a
database for ADRs. Furthermore, the drug-drug interaction should lead to an
increase in the number of reports, which implies a rather strong clinical effect. If
the association is monitored for in pharmacist or physician information systems,
there is a chance that it will be reported less frequently since it might be
considered as a well-known interaction.
The aim of the present study was to illustrate the possible use of an SRS as a tool
for signalling drug-drug interactions. Our results were indeed supportive for this
drug-drug interaction. Given the fact that this interaction was present in a
relatively small database, the use of larger databases may be promising. Although
in our example we strengthened a signal that was already known in literature,
SRS may also be used for the actual detection of drug-drug interactions, provided
high quality reports are available and concomitant drug use is filed in the
database. Due to an increase in the size and quality of the SRS databases,
103
Chapter 2.3
detection and analysis of drug-drug interactions clearly offers a major challenge

for pharmacovigilance in the near future. In contrast to the present way of
detecting drug-drug interactions, the method described provides an active
approach in the post marketing phase of drugs.
Acknowledgements
We acknowledge the Netherlands Organisation for Applied Scientific Research

(TNO), Prevention and Health subdivision, for partly funding this study (project
ZON 28-2632).
References
1998;54:315-21.
3. Lindquist M, Edwards IR, Bate A, Fucik H, Nunes AM, Ståhl M. From association to alert -
A revised approach to international signal analysis. Pharmacoepidemiol Drug Saf
1999;8:S15-S25.
4. Meyboom RH. Pharmacovigilance in perspective. Drug Saf 1999;21:429-47.
5. Stricker BHCh, Tijssen, JGP Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;45:1177-84.
7. Van Puijenbroek EP, Egberts ACG, Meyboom RHB, and Leufkens HGM. Signalling
possible drug-drug interactions in a spontaneous reporting system: Delay of withdrawal
bleeding during concomitant use of oral contraceptives and itraconazole. Br J Clin
Pharmacol 1999;47:689-93.
8. Patak RV, Mookerjee BK, Bentzel CJ, Hysert PE, Babej M, Lee JB. Antagonism of the
effects of furosemide by indomethacin in normal and hypertensive man. Prostaglandins
1975;10:649-59.
9. Rawles JM. Antagonism between non-steroidal anti-inflammatory drugs and diuretics. Scott
Med J 1982;27:37-40.
10. Symmons DP, Kendall MJ, Rees JA, Hind ID. The effect of flurbiprofen on the responses to
frusemide in healthy volunteers. Int J Clin Pharmacol Ther Toxicol 1983;21:350-4.
11. Daskalopoulos G, Kronborg I, Katkov W, Gonzalez M, Laffi G, Zipser RD. Sulindac and
indomethacin suppress the diuretic action of furosemide in patients with cirrhosis and
ascites: evidence that sulindac affects renal prostaglandins. Am J Kidney Dis 1985;6:217-21.
104
12. Brater C, Chennavasin P. Indomethacin and the response to bumetanide. Clin Pharmacol
Ther 1980;27:421-5.
13. Ahmad S. Indomethacin-bumetanide interaction: an alert. Am J Cardiol 1984;54:246-7.
14. Favre L, Glasson P, Riondel A, Vallotton MB. Interaction of diuretics and non-steroidal anti-
inflammatory drugs in man. Clin Sci 1983;64:407-15.
15. Baker DE. Piroxicam-furosemide drug interaction. Drug Intell Clin Pharm 1988;22:505-6.
16. Hartmann D, Kleinbloesem CH, Lucker PW, Vetter G. Study on the possible interaction
between tenoxicam and furosemide. Arzneimittelforschung 1987;37:1072-6.
17. Wa TC, Lawson M, Jackson SH, Hitoglou-Makedou A, Turner P. Interaction of ketoprofen
and frusemide in man. Postgrad Med J 1991;67:655-8.
18. Herchuelz A, Derenne F, Deger F, Juvent M, Van Ganse E, Staroukine M, Verniory A,
Boeynaems JM, Douchamps J. Interaction between nonsteroidal anti-inflammatory drugs
and loop diuretics: modulation by sodium balance. J Pharmacol Exp Ther 1989;248:1175-81.
19. Kaufman J, Hamburger R, Matheson J, Flamenbaum W. Bumetanide-induced diuresis and
natriuresis: effect of prostaglandin synthetase inhibition. J Clin Pharmacol 1981;21:663-7.
20. Feenstra J, Grobbee DE, Mosterd A, Stricker BH. Adverse cardiovascular effects of NSAIDs
in patients with congestive heart failure. Drug Saf 1997;17:166-80.
21. Heerdink ER, Leufkens HG, Herings RM, Ottervanger JP, Stricker BH, Bakker A. NSAIDs
associated with increased risk of congestive heart failure in elderly patients taking diuretics.
Arch Intern Med 1998;158:1108-12.
1993;7:119-37.
24. Martin RM, Kapoor KV, Wilton LV, Mann RD. Underreporting of suspected adverse drug
reactions to newly marketed ("black triangle") drugs in general practice: observational study.
BMJ 1998;317:119-20.
26. Lumley CE, Walker SR, Hall GC. The under-reporting of adverse drug reactions seen in
general practice. Pharm Med 1986;1:205-12.
risk & safety in Medicine 1993;5:123.
Monitoring, Uppsala, Sweden; 1995.
29. Blackshear JL, Davidman M, Stillman MT. Identification of risk for renal insufficiency from
nonsteroidal anti- inflammatory drugs. Arch Intern Med 1983; 143:1130-4.
30. Nevins M, Berque S, Corwin N, Lyon L. Phenylbutazone and pulmonary oedema. Lancet
1969;2:1358.
31. Tashima CK, Rose M. Letter: Pulmonary edema and salicylates. Ann Intern Med
1974;81:274-5.
32. Coles LS, Fries JF, Kraines RG, Roth SH. From experiment to experience: side effects of
nonsteroidal anti- inflammatory drugs. Am J Med 1983;74:820-8.
105
Chapter 2.3
33. Lim PO, MacDonald M. Antianginal and b-adrenoreceptor blocking drugs. In: Dukes MNG
(ed). Meyler's Side Effects of Drugs. Amsterdam: Elsevier Science, 1996:488-535.
106
2.4
ASSOCIATION BETWEEN TERBINAFINE AND

ARTHRALGIA, FEVER AND URTICARIA:
SYMPTOMS OR SYNDROME?
EP van Puijenbroek1, ACG Egberts2,3, RHB Meyboom1 and HGM Leufkens2
1
2
Utrecht Institute for Pharmaceutical Sciences, Utrecht, the Netherlands
3
PHARMACOEPIDEMIOLOGY AND DRUG SAFETY 2001;10:135-142

Chapter 2.4
Summary
Objective The antifungal agent terbinafine has been approved for marketing in
the Netherlands since 1992. Adverse drug reactions (ADRs) may occur in about
10 % of the patients, the majority being gastro-intestinal disorders and skin
reactions. Since the introduction of terbinafine, the Netherlands
Pharmacovigilance Foundation Lareb received eight reports of arthralgia during
the use of this drug. In four reports the additional presence of skin reactions was
mentioned, two of these reports concerned urticaria. Two patients who reported
arthralgia also had a fever. These reports were described in more detail, and
analysed statistically in order to determine whether symptoms are interrelated.
Methods All reports with known gender and a reporting date between 1 March
1992 and 1 January 1999, concerning patients older then 10 years, were included.
The extent to which the symptoms urticaria, fever and arthralgia were interrelated
was examined by logistic regression modelling.
Results Case series as well as the results of the statistical analysis show a
clustering of symptoms among reports of patients using terbinafine. Both urticaria
and arthralgia were statistically significantly associated with reports on
terbinafine compared to all other reports in the database.
Conclusion The findings might point towards a clustering of these symptoms in
patients using terbinafine. Possibly these symptoms have a shared aetiology,
presumably an immunological reaction.
108
Association between terbinafine and arthralgia, fever and urticaria
Introduction
The orally and topically active allylamine antifungal agent terbinafine has been
approved for marketing in the Netherlands since March 1992. The drug interferes
with the ergosterol biosynthesis via specific and selective inhibition of fungal
squalene epoxidase.1 Terbinafine is effective in the treatment of onychomycosis
and several other types of dermatomycotic infections.2,3 Its spectrum includes a
broad range of dermatophyte and some yeast species. Adverse drug reactions
(ADRs) may occur in about 10 % of the patients treated with terbinafine, the
majority being gastro-intestinal disorders (5%), and skin reactions ( 2-3%).4-6
Rarely, terbinafine has been associated with severe drug eruptions such as toxic
epidermal necrolysis,7,8 erythema multiforme8-10 and Stevens-Johnson syndrome.11
In general, frequently occurring ADRs are usually detected in pre-marketing
trials. In these trials, drugs are used by a well-defined group of a relative small
number of patients for a limited time span.12 After marketing, when the drug is
used by a large number of patients under different circumstances, previously
unknown ADRs are still being detected. An example is the detection of taste
disorders associated with terbinafine, which was described for the first time in
1992, after the international marketing of the drug.13,14 Reporting of ADRs by
pharmacists or physicians to ‘spontaneous reporting systems’ (SRS) play an
important role in the detection of ADRs. In most countries these systems are
maintained by so-called ‘national pharmacovigilance centres’. In the Netherlands,
the Pharmacovigilance Foundation Lareb is responsible for collecting and
analysing these reports. Since the introduction of terbinafine, Lareb received 294
reports of suspected adverse reactions to terbinafine. Eight reports concerned
arthralgia. In four of these reports the reporting physician or pharmacist also
mentioned the presence of skin reactions, including two reports of urticaria. Two
patients who reported arthralgia also had a fever. The reports suggest a clustering
of arthralgia, fever and urticaria. The association between terbinafine and
arthralgia is not yet mentioned in the Dutch Summary of Product Characteristics,
but will be in the future.
The objective of this article is to describe the clinical characteristics of these
reports and to analyse the clustering of these symptoms statistically in order to
determine the strength of the association. In case symptoms are interrelated, this
might be an indication for the existence of a syndrome.
109
Chapter 2.4
Methods
Setting
The Netherlands Pharmacovigilance Foundation Lareb is the National Centre for
spontaneous reports of suspected ADRs originating from health care professionals
in the Netherlands.15 These reports are considered to be a reflection of the ADRs
that occur in daily practice, taking into account the various degrees of
underreporting that are an inherent attribute of spontaneous reporting.16-19 After
being received by Lareb, the reported possible adverse drug reactions are
evaluated and coded by a qualified assessor according to the WHO Adverse Drug
Reaction Terminology.20
Case series
For all reports of terbinafine coded by ‘arthralgia’, additional information
concerning the clinical details was retrieved from the reporter by means of a
questionnaire. Special attention was paid to the symptoms involved, type and
distribution of arthralgia or arthritis and accompanying symptoms such as fever,
lympadenopathy, renal function and skin disorders. Furthermore, additional
questions were asked about the time of onset and course of the symptoms, a
history of allergic or rheumatological disorders, liver or renal function disorders
or a family history of rheumatological disorders. In the event an additional
radiological examination or laboratory testing was carried out, a copy of the
results was asked for. Questionnaires were sent to the reporting physician or
pharmacist together with a copy of the data already present in our database. After
3 weeks, a reminder was sent. If there was no response after an additional period
of 3 weeks, the reporting physician or pharmacist was contacted by telephone.
Analysis of clustering
Reports with a reporting date between 1 March 1992 and 1 January 1999 were
eligible for enrolment in the study. Since terbinafine is rarely used in children and
in our database no reports concerning patients younger than 13 years were
present, only reports concerning patients older than 10 years of age were
included. Reports were excluded where the gender of patients was not reported.
A report to an SRS can be considered as a set of data considering a single patient,
in combination with one ore more drugs and one or more ADRs. Statistical
110
methods, such as logistic regression analysis can be used to provide evidence for
a possible underlying relationship between these different covariates. In the SRS
databases the strength of the association between a drug and ADR can be studied
by calculating Reporting Odds Ratios.21,22 To study a possible relationship
between fever, urticaria and arthralgia, the ADRs were considered being
covariates and the presence of terbinafine as the suspected drug on the report form
to be the dependent variable. The extent to which the covariates are interrelated
can be examined by using statistical ‘interaction terms’ in a logistic regression
model. Since we were interested in expressing the presence of terbinafine as a
function of arthralgia, fever and urticaria, cases were defined as reports on which
terbinafine (oral administration) was mentioned as the suspected medication, non
cases were defined as all other reports. Reporting Odds Ratios (RORs) were
calculated, which were adjusted for age and gender of the patients, source of the
reports and year of reporting. In the logistic regression analysis a forward
stepwise approach was applied. SPSS 8.0 was used for all statistical assessments.
Results
Case series
Cases involved five men and three women. All patients were treated with
terbinafine 250 mg once daily. The median age was 42 years (range 34-66 years).
Time of onset of the reaction ranged from 1 or 2 days until 18 days. Seven
physicians and one pharmacist submitted the reports. Additional information was
received with regard to six out of the eight cases of arthralgia. One patient had
moved so additional information could not be provided. Clinical details of these
patients (A-H) are presented in Table 1 and the addendum of this thesis. No clear
pattern could be recognised in the type of joints involved and the distribution of
the arthralgia. Polyarthralgia was present in all eight cases. Only patient B had a
history of backache. Patients A-F and H were treated for onychomycosis. The
indication for use of patient G was not received. Patients A,B,D-F and H had no
history of joint disorders, traumata, liver or kidney function disorders,
111
Table 1. Clinical details of eight reports concerning arthralgia during the use of terbinafine
Patient; sex; Indication Concomitant Description of suspected Time for the Remarks
age (years) for use medication ADR ADR to occur
A; F; 44 Onycho- Doxycycline 100 mg Arthralgia of right wrist, one 18 days after Fully recovered after stopping
mycosis during 7 days (1 week week later followed by right start terbinafine. ESR, C-reactive
after start of knee. ADL impaired protein levels, streptococcal
terbinafine) antibody titre or rheumatoid
Ethinylestradiol/ factors not increased. Treated by
desogestrel ibuprofen 400 mg three times
daily.
B; M; 41 Mycosis Ibuprofen 400 mg, Arthralgia of shoulders, hips 10 days after Recovered after stopping
miconazole cream; and elbows. Symmetrical start terbinafine Lab: ESR, 5 mm; GGT,
Beclomethasone affected. No signs of arthritis. 11 mmol/l; Hb, 8.9 mmol/l;
inhaler ADL impaired. Backache in leukocyte count 5.4 x 109/l.
history. Treated with nabumetone 1 g twice
daily.
C; F; 42 Onycho- Pain of hands and feet 1-2 days after No additional information
mycosis start received.
D; M; 41 Onycho- Arthralgia. Pain on flexion 2 weeks after Recovered. No additional
mycosis and extension of wrists and start laboratory examination.
knees. No signs of arthritis.
E; F; 37 Onycho- ‘Joint complaints’, malaise, 10 days after Recovered after stopping
mycosis rash (pruritus, erythema) start terbinafine leukocyte count
7.2x109/l, ALT, GGT not
increased. In blood smear, some
atypical lymphocytes. After 1
week re-examination of blood
revealed no abnormalities.
112
F;M; 66 Onycho- Echinacea containing Urticarial rash, fever, 2 weeks after Treated with terfenadine and
mycosis drug headache, vomiting, diffuse start prednisolone. Lab: Hb, ESR,
arthralgia. leukocytes and blood smear within
No joints specified. normal limits.
G; M; 34 Unknown Miconazole cream Generalized urticaria and 10 days after Treated with terfenadine,
arthralgia. No joints specified start prednisolone and promethazine.
H; M; 50 Onycho- Budesonide inhaler Pain in almost every joint and Couple of days Recovered. Treated with
mycosis myalgia. No signs of arthritis. 1-2 diclofenac and domperidone.
One week after the start of
terbinafine, anorexia,
vomiting, abdominal pain,
smell and taste disorders.
Also rash, stomatitis and
peeling of the skin developed.
Fever of 39ºC leukocyte
count 4.8 x109/l; ESR, 5 mm;
Hb, 9.3 mmol/l; ALT, 1672
U/l; LDH, 869 U/l; GGT, 202
U/l; kidney function normal.
All patients were treated with terbinafine 250 mg once daily. Indication for use of patient G was not known. Patients A, B, D-F and H
had no history of rheumatological or allergic disorders, traumata, liver or kidney function disorders. No rechallenge was carried out
in either of the patients. These data could not be retrieved for patients C and G.
113
Chapter 2.4
rheumatological or allergic disorders. There were no joint disorders in the family

history of any of the patients concerned. These data could not be retrieved for
patients C and G. In none of the patients a rechallenge was carried out.
Analysis of clustering
Between 1 March 1992 and 1 January 1999, a total number of 16,776 reports
were received by the Netherlands Pharmacovigilance Foundation. Of 98 patients
sex or age was not known and 649 reports were excluded because the patients
were younger than 10 years of age. A total of 16,127 reports were included in the
analysis, of which 294 reports mentioned terbinafine as the suspected medication.
Joint complaints were reported 154 times, of which eight were associated with
terbinafine. The distribution of the ADRs urticaria, fever and arthralgia on reports
concerning terbinafine and other reports are presented in Table 2.
Table 2. Distribution of ADRs: urticaria, arthralgia and fever on terbinafine versus all
other drugs
Reports on terbinafine Reports on other
n (%) drugs
n (%)
Only arthralgia reported 5 (1.7%) 142 (0.9%)

Only urticaria reported 17 (5.8%) 345 (2.2%)
Only fever reported 2 (0.7%) 161 (1.1%)
Both arthralgia and urticaria reported 1 (0.3%) 2 (0.01%)
Both arthralgia and fever reported 1 (0.3%) 2 (0.01%)
Both fever and urticaria reported 2 (0.01%)
Urticaria, fever and arthralgia reported 1 (0.3%)
No arthralgia, urticaria of fever reported 267 (90.8%) 15,179 (95.9%)
Total 294 15,833
Both urticaria (adjusted ROR 1.72 (95% CI 1.35-2.18)) and arthralgia (adjusted
ROR 3.14 (95% CI 1.52-6.47)) were significantly associated with reports on
terbinafine (Table 3).
114
Table 3. Association between arthralgia, fever, urticaria and terbinafine

Unadjusted odds ratio (95% CI) Adjusted odds ratio (95% CI)*
Fever 1.06 (0.83-1.37) 1.06 (0.83-1.36)

Urticaria 1.75 (1.38-2.22) 1.72 (1.35-2.18)
Arthralgia 3.00 (1.46-6.18) 3.14 (1.52-6.47)
*Odds ratio adjusted for year of reporting, source of the reports, age and gender of the
patient
The covariates being the best predictors of the dependent variable were urticaria
(adjusted ROR 1.66 (95% CI 1.29-2.14)) as well as the interaction terms
arthralgia*fever (adjusted ROR 2.35 (95% CI 1.32-4.17)) and arthralgia*urticaria
(adjusted ROR 3.33 (95% CI 1.03-10.73)). The odds ratios and corresponding
95% confidence intervals are shown in Table 4.
Table 4. Association between terbinafine and arthralgia, fever and urticaria. Results of
logistic regression (forward stepwise approach)
. Odds ratio
(95% CI)
Arthralgia * fever 2.35 (1.32-4.17)

Arthralgia * urticaria 3.33 (1.03-10.73)
Urticaria 1.66 (1.29-2.14)
Discussion
The case series as well as the statistical analysis suggest a clustering of fever,
arthralgia and urticaria in association with the oral antifungal drug terbinafine. In
general, ADRs to terbinafine are often mild. The reported overall incidence of
ADRs in a large post marketing study involving 25,884 patients was 10.5%. Of
all events 55.9% were considered to be related to be ‘probably’ or ‘possibly’
related to terbinafine. The majority involved the gastrointestinal system and the
skin.5 Nevertheless more serious reactions may occur. Liver and kidney function
disorders may decrease the elimination of terbinafine.23,24 These patients are more
likely to experience ADRs due to an increase in plasma levels of terbinafine. For
115
Chapter 2.4
this reason we asked for liver and kidney function disorders in our questionnaire,
but they were not observed. Complaints of the musculo-skeletal system were
reported in 0.8% in the postmarketing study mentioned.24 In another post
marketing study among 10,361 patients, which were probably part of the previous
mentioned postmarketing study, 10.9% of the patients had pre-existing musculo-
skeletal system disorders. In this study, musculo-skeletal events were reported by
1.3% of the patients, of which only 25.2% were thought to be related to
terbinafine treatment. Arthralgia occurred in 0.4% of the patients, myalgia in
0.2%. Half of the cases of arthralgia and myalgia occurred within the first 2
weeks of treatment, and were considered as being possibly related to the use of
terbinafine. None of the musculo-skeletal events was serious and there was no
evidence that terbinafine exacerbated pre-existing muskulo-skeletal symptoms.
Dermatological events were reported by 341 patients (3.3%), of which urticaria
was reported 47 times (0,4%).25 Both studies had a descriptive character and no
control groups were used. Details on the simultaneous occurrence of different
symptoms were not provided.
In the literature anecdotal case reports have appeared, describing symptoms
similar to those in our study. It is not yet clear if these reports represent distinct or
related symptoms; in the latter case they might share a common aetiology. For
instance, terbinafine has been associated with the clinical diagnosis of a serum
sickness-like reaction in one case report. This concerned a well documented case
of an-81-year-old male, who developed exanthematous rash, fever, myalgia and
arthralgia 6 weeks after the start of terbinafine therapy.26 True serum sickness is
caused by an immune complex-mediated reaction to a foreign serum protein
occurring 1-3 weeks after exposure. Usually it presents with an urticarial rash,
followed or accompanied by pain and swelling of the joints (knees, ankles and
wrists) in a symmetrical distribution. Serum sickness-like reactions, with
arthralgia or arthritis, skin reaction or fever may occur with many drugs, but
penicillin is the most common cause.27-29 The diagnosis of a true serum sickness-
like reaction, however, requires information about complement factors C3 and
C4. Since this is not a parameter commonly tested by general practitioners, this
information was not available. For the majority of the reactions, reported to
Lareb, the time of onset is compatible with a serum sickness-like reaction. Patient
A also used doxycycline, which was incidentally associated with a serum
sickness-like reaction.28,30 Involvement of doxycycline in this patient cannot be
116
fully excluded. Alternatively, urticaria, joint pain and sometimes swelling of the
joint can occur together as 'urticarial arthralgia', in which attacks of severe
urticaria and joint pain occur coincidentally due to an urticarial reaction of the
synovia. Usually it involves a type I hypersensitivity reaction, although some
food-allergic reactions also might be immune complex-mediated.31 In another
case history in the literature, terbinafine has been associated with a
hypersensitivity syndrome consisting of fever, pruritic eruption, lymphadenopathy
and hepatitis 10 days after the start of oral terbinafine.32 No arthralgia was
mentioned in this patient. Another hypersensitivity syndrome was described in a
66-year-old male who developed a cutaneous eruption, fever, lymfadenopathy and
hepatic dysfunction after 4.5 weeks of therapy.33 Also our patient H had abnormal
liver function tests, and a skin reaction with erythema and peeling. Terbinafine
has in incidental case reports also been associated with severe drug eruptions such
as toxic epidermal necrolysis,7,8 erythema multiforme,8-10 Stevens-Johnson
Syndrome11 and acute generalised exanthematous pustulosis.34,35 Also drug-
induced psoriasis,4,36 cutaneous lupus erythematosus37,38 and the exacerbation of a
systemic lupus erythematosus39 were reported. In our patients, the symptoms were
not suggestive of a SLE-like syndrome, but this could not be ruled out, since no
antinuclear antibodies were checked.
Other rare skin reactions, associated with the use of terbinafine are a fixed drug
eruption40 and an erythema anulare centrifugum-like psoriatic drug eruption.41
The description of the symptoms of patient H might resemble a Stevens-Johnson
syndrome or erythema multiforme. Next to mucosal lesions, a Stevens-Johnson
syndrome may be accompanied by extracutaneous manifestations, like arthralgia,
liver and kidney disorders and fever.42,43 In a Stevens-Johnson syndrome, an
immune mechanism might be involved in the process. Circulating antibody-
antigen complexes and complement activation have been demonstrated in the
blood and skin lesion of patients.42 Not only terbinafine, but also other antifungal
agents have also been associated with joint disorders. The antifungal antibiotic
amphotericin B may cause severe pains in muscles and joints along with many
other side-effects.27 Itraconazole was also associated with a serum sickness-like
reaction in a 53-year-old female.44 As far as we know, there are no signs from
mycotic infections that resemble the symptoms presented in our patients.
Confounding by indication is therefore unlikely.
117
Chapter 2.4
In an additional analysis the clustering of the separate symptoms was analysed in

more detail. In logistic analysis, next to urticaria two interaction terms were found
to be the strongest predictor of the dependent variable. This indicates a clustering
of arthralgia and fever as well as arthralgia and urticaria in reports on terbinafine.
In the control group, however, no report was present in which all three symptoms
were mentioned together. For this reason, the corresponding interaction term
indicating the clustering of all three symptoms could not be taken into account in
the analysis. All eight patients in our analysis had ‘arthralgia’. Because the
distinction between arthralgia and arthritis is sometimes difficult to make, we also
included the WHO term ‘arthritis’ in a separate analysis. Also in this situation the
variables urticaria and the interaction term arthralgia*fever were statistically
significant. The findings of the statistical analysis are in support of a clustering of
the symptoms involved, but cannot be considered as conclusive evidence. For
methodological reasons, suspicions of new or unexpected ADRs originating from
reporting systems only have a signal generating value. In general, further
epidemiological studies are necessary to confirm a possible causal relationship
between a suspected drug and suspected ADRs.
Spontaneous reporting systems are primarily designed for the detection of
individual symptoms in relation with drug use. These signals can be detected by
means of the analysis of individual reports. Also more complex relationships can
be studied in SRS databases, like signalling possible drug-drug interactions.45,46
When two or more adverse events occur in a patient, it is of major interest, but
often difficult to determine whether they both have been caused by the same or by
a different pharmacological or pathological mechanism, i.e. if they are related
(part of a syndrome) or coincidental. In this study we have used statistical
reasoning to address the question if terbinafine-associated arthralgia, urticaria and
fever are likely to be related or separate events. We propose that a similar
approach may be more widely used in identifying drug-induced syndromes in
spontaneous reporting databases.
Conclusion
Reports to the Netherlands Pharmacovigilance Foundation Lareb are suggestive

of an association between the use of the antifungal agent terbinafine and the
occurrence of arthralgia, fever and urticaria. Case series as well as the results of
118
the statistical analysis show a clustering of symptoms among reports of patients

using terbinafine. These findings might point towards a shared aetiology of these
symptoms, presumably an immunological reaction. If arthralgia develops, it is
important to inquire about involvement of other organ systems and to have
additional laboratory examination done. Considering the possibility of an
immunological reaction, it is advisable that the use of terbinafine in these patients
should be stopped.
Acknowledgements
We acknowledge physicians and pharmacist of the patients concerned for their

help, and the Netherlands Organisation for Applied Scientific Research (TNO),
subdivision Prevention and Health, for partly funding this study (project ZON 28-
2632).
References
1. Balfour JA, Faulds D. Terbinafine. A review of its pharmacodynamic and pharmacokinetic
properties, and therapeutic potential in superficial mycoses. Drugs 1992;43:259-84.
2. Baudraz-Rosselet F, Rakosi T, Wili PB, Kenzelmann R. Treatment of onychomycosis with
terbinafine. Br J Dermatol 1992;126(Suppl 39):40-6.
3. Villars V, Jones TC. Clinical efficacy and tolerability of terbinafine (Lamisil)-a new topical
and systemic fungicidal drug for treatment of dermatomycoses. Clin Exp Dermatol
1989;14:124-7.
4. Gupta AK, Sibbald RG, Knowles SR, Lynde CW, Shear NH. Terbinafine therapy may be
associated with the development of psoriasis de novo or its exacerbation: four case reports
and a review of drug-induced psoriasis. J Am Acad Dermatol 1997;36:858-62.
5. Hall M, Monka C, Krupp P, O'Sullivan D. Safety of oral terbinafine: results of a
postmarketing surveillance study in 25,884 patients. Arch Dermatol 1997;133:1213-9.
6. Abdel-Rahman SM, Nahata MC. Oral terbinafine: a new antifungal agent. Ann
Pharmacother 1997;31:445-56.
7. White SI, Bowen-Jones D. Toxic epidermal necrolysis induced by terbinafine in a patient on
long-term anti-epileptics. Br J Dermatol 1996;134:188-9.
8. Carstens J, Wendelboe P, Sogaard H, Thestrup-Pedersen K. Toxic epidermal necrolysis and
erythema multiforme following therapy with terbinafine. Acta Derm Venereol 1994;74:391-
2.
9. McGregor JM, Rustin MH. Terbinafine and erythema multiforme. Br J Dermatol
1994;131:587-8.
119
Chapter 2.4
10. Todd P, Halpern S, Munro DD. Oral terbinafine and erythema multiforme. Clin Exp
Dermatol 1995;20:247-8.
11. Rzany B, Mockenhaupt M, Gehring W, Schopf E. Stevens-Johnson syndrome after
terbinafine therapy. J Am Acad Dermatol 1994;30:509.
12. Post-marketing Surveillance In: Stephens MDB (ed). Detection of New Adverse Drug
Reactions. Hampshire: MacMillan Publishers Ltd, 1992:183-5.
13. Ottervanger JP, Stricker BH. Loss of taste and terbinafine. Lancet 1992;340:728.
14. Juhlin L. Loss of taste and terbinafine. Lancet 1992;339:1483.
1993;7:119-37.
17. Martin RM, Kapoor KV, Wilton LV, Mann RD . Underreporting of suspected adverse drug
reactions to newly marketed ("black triangle") drugs in general practice: observational study.
BMJ 1998;317:119-20.
19. Lumley CE, Walker SR, Hall GC. The under-reporting of adverse drug reactions seen in
general practice. Pharm Med 1986;1:205-12.
21. Stricker BHCh, Tijssen, JGP Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;45:1177-84.
22. Egberts AC, Meyboom RH, De Koning FH, Bakker A, Leufkens HG. Non-puerperal
23. Jensen JC. Clinical pharmacokinetics of terbinafine (Lamisil). Clin Exp Dermatol
1989;14:110-3.
24. Gupta AK, Shear NH. Terbinafine: an update. J Am Acad Dermatol 1997;37:979-88.
25. O'Sullivan DP, Needham CA, Bangs A, Atkin K, Kendall FD. Postmarketing surveillance of
oral terbinafine in the UK: report of a large cohort study. Br.J Clin Pharmacol 1996;42:559-
65.
26. Kruczynski K, Balter MS. Serum sickness-like reaction associated with oral terbinafine
therapy. Can J Clin Pharmacol 1995;2:129-30.
27. Hart FD. Drug-induced arthritis and arthralgia. Drugs 1984;28:347-54.
28. Immune-complex diseases. In: Frank MM, Lawley TJ. Isselbacher KJ, Braunwald E, Wilson
JD, Martin JB, Fauci AS, Kasper DL (eds). Principles of Internal Medicine. New York:
McGraw-Hill, 1994:1638-43.
29. Assem E-SK. Drug allergy and tests for its detection. In: Davies DM (eds). Textbook of
Adverse Drug Reactions. Oxford: Oxford University Press, 1991:689-713.
30. Vidal PC, Gonzalez QA. Doxycycline-induced parotitis. Postgrad Med J 1991;67:313-4.
31. Golding DN. Is there an allergic synovitis? J R Soc Med 1990;83:312-4.
32. Gupta AK, Kopstein JB, Shear NH. Hypersensitivity reaction to terbinafine. J Am Acad
Dermatol 1997;36:1018-9.
120
33. Gupta AK, Porges AJ. Hypersensitivity syndrome reaction to oral terbinafine. Austral J
Dermatol 1998;39:171-2.
34. Dupin N, Gorin I, Djien V, Helal H, Zylberberg L, Leibowitch M, Escande JP. Acute
generalized exanthematous pustulosis induced by terbinafine. Arch Dermatol
1996;132:1253-4.
35. Condon CA, Downs AM, Archer CB. Terbinafine-induced acute generalized exanthematous
pustulosis. Br J Dermatol 1998;138:709-10.
36. Wilson NJ, Evans S. Severe pustular psoriasis provoked by oral terbinafine. Br J Dermatol
1998;139:168.
37. Murphy M, Barnes L. Terbinafine-induced lupus erythematosus. Br J Dermatol
1998;138:708-9.
38. Brooke R, Coulson IH, al-Dawoud A. Terbinafine-induced subacute cutaneous lupus
erythematosus. Br J Dermatol 1998;139:1132-3.
39. Holmes S, Kemmett D. Exacerbation of systemic lupus erythematosus induced by
terbinafine. Br J Dermatol 1998;139:1133.
40. Munn SE, Russell J. Terbinafine and fixed drug eruption. Br J Dermatol 1995;133:815-6.
41. Wach F, Stolz W, Hein R, Landthaler M. Severe erythema anulare centrifugum-like psoriatic
drug eruption induced by terbinafine. Arch Dermatol 1995;131:960-1.
42. Nethercott JR and Choi BC. Erythema multiforme (Stevens-Johnson syndrome)- chart
review of 123 hospitalized patients. Dermatologica 1985;171:383-96.
43. Denman AM. Stevens-Johnson syndrome followed by persistent recurrent severe arthralgia.
Br J Reumatology 1990; Br J Rheumatol 1990; 29:214.
44. Park H, Knowles S, Shear NH. Serum sickness-like reaction to itraconazole. Ann
Pharmacother 1998;32:1249.
45. Van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens HGM. Signalling possible
drug-drug interactions in a spontaneous reporting system: Delay of withdrawal bleeding
during concomitant use of oral contraceptives and itraconazole. Br J Clin Pharmacol
1999;47:689-93.
46. Van Puijenbroek EP, Egberts ACG, Leufkens HGM. Detecting drug-drug interactions using
a database for spontaneous reports of adverse drug reactions: an example with diuretics and
non-steroidal anti-inflammatory drugs. Eur J Clin Pharmacol 2000;56:733-9.
121
Part 3
IMPLEMENTATION OF THE QUANTITATIVE APPROACH
In the third part of this thesis the implementation of quantitative signal detection
at the Netherlands Pharmacovigilance Foundation Lareb is discussed. Detection
of new adverse drug reactions (ADRs) after marketing is often based on a manual
review of reports sent to a Spontaneous Reporting System (SRS). Among the
many potential signals that are identified, only a limited number are important
enough to require further attention. Goal of the study of Chapter 3.1 is to gain
insight into factors contributing to the selection and dissemination of possible
signals from the Netherlands Pharmacovigilance Foundation Lareb to the Dutch
Medicines Evaluation Board and the contribution of quantitative signal detection
in this process.
Although in various centres quantitative techniques are being developed, they are
generally not routinely implemented. The practical aspects of the implementation
of quantitative signal detection are described in Chapter 3.2. At the Netherlands
Pharmacovigilance Foundation quantitative signal detection has been used in
various ways for a number of years. In this chapter, the role and position of
quantitative signal detection and the way it is applied is described.
3.1
DETERMINANTS OF SIGNAL SELECTION IN A

SPONTANEOUS REPORTING SYSTEM FOR
ADVERSE DRUG REACTIONS
EP van Puijenbroek1, AC van Grootheest1, WL Diemont1, HGM Leufkens2,

and ACG Egberts2,3
1
2
3
BRITISH JOURNAL OF CLINICAL PHARMACOLOGY (IN PRESS)

Chapter 3.1
Summary
Objective Detection of new adverse drug reactions (ADR) after marketing is

often based on a manual review of reports sent to a Spontaneous Reporting
System (SRS). Among the many potential signals that are identified, only a
limited number are important enough to require further attention. Goal of this
study is to gain insight into factors contributing to the selection and dissemination
of possible signals originating from the SRS maintained by the Netherlands
Pharmacovigilance Foundation.
Methods In a case control design, all signals (n=42) disseminated to the
Medicines Evaluation Board from the second quarter of 1997 until the third
quarter of 2000, which could be expressed as a combination of a single ATC code
and a single WHO preferred term, were included. For each case, four controls
were matched in time. Logistic regression analysis was used to investigate the
influence of various factors, such as the fact whether the ADR or drug is new, the
strength of the association, the seriousness of the reaction and the documentation
of the reports.
Results Multivariate analysis showed that the presence of a ‘serious report’ (Odds
Ratio 3.8, 95% CI 1.3-11), a WHO ‘critical term’ (OR 4.7, 95% CI 1.8-13), the
ADR being unlabelled (OR 6.1, 95% CI 2.3-16) and the presence of a
disproportionate association (OR 3.5, 95% CI 1.4-8.4) were all independently
associated with signal selection. The number of reports and the time after
marketing of the drug had no influence.
Conclusion This study showed that selection of signals is based on both
qualitative and quantitative aspects. Knowledge of these factors may improve the
efficiency of the underlying signal selection process.
126
Determinants of signal selection
Introduction
After marketing of a drug, close monitoring for unexpected adverse drug

reactions (ADRs) remains necessary due to the limited size of pre-marketing
trials, the selection of the patients involved and the limited duration of the trials.1
For this reason, post marketing surveillance aims at a timely detection of either
new ADRs or an increase of the frequency of ADRs which are already known to
be associated with the drugs involved. Spontaneous reporting systems (SRSs) still
play an important role in providing early signals concerning suspected ADRs. A
signal can be defined as reported information on a possible causal relation
between an adverse event and a drug, the relation being previously unknown or
incompletely documented.2 The detection of signals from SRSs generally results
from a systematic manual review of every incoming report sent to
pharmacovigilance centres. In the present way of signal detection many potential
signals are identified. However, relatively few of them are important enough to
require further attention. For this reason, criteria need to be formulated to help to
determine if there is a need for further attention concerning a particular signal.
Several factors may help to determine whether or not a particular signal is worth
further investigation. In general, four categories may be identified, which are for
instance represented by the acronym ‘SNIP’, used by the Medicines Control
Agency. These categories refer to the Strength of a signal, whether or not the
issue is New, the clinical Importance as judged by the seriousness of the reaction
and severity of the cases and finally the potential for Preventive measures by the
regulatory authorities.3 Most pharmacovigilance centres, however, do not
systematically apply such criteria.
At the Netherlands Pharmacovigilance Foundation Lareb, maintaining the SRS in
the Netherlands, signal selection is carried out by a review of every incoming
report in a weekly meeting between trained assessors. Working on behalf of the
Dutch Medicines Evaluation Board (MEB), this agency is informed periodically
about generated signals. On a three-monthly basis the most relevant signals are
published in a so-called quarterly report to the MEB. Since the goal of this report
is to give an early warning to the authorities, the signals described have a
premature character. Criteria for selecting interesting signals about which the
MEB should be informed, however, are not predefined and depend on the
knowledge and experience of the assessors involved. Although in a later stage the
127
Chapter 3.1
signals described may be superseded, these quarterly reports give a unique chance
for studying the early stages of the process of signal selection and dissemination
of information from SRSs. The goal of this study is to gain insight in the various
factors that play a role in the subjective process of signal selection. Knowledge of
these factors may improve the efficiency of the underlying signal selection
process.
Methods
Setting
The Netherlands Pharmacovigilance Foundation Lareb collects and analyses
reports of suspected ADRs reported by Dutch physicians and pharmacists. A
report may concern one or more suspected ADRs and one or more suspected
drugs. The reports are evaluated, coded and filed in a database. To the reporting
health professional a letter of confirmation is sent together with information
regarding the reported association. All ADRs are coded according the WHO
adverse drug reaction terminology.4 In this respect, possible ADRs are assigned to
a so-called ‘preferred term’, which gives a detailed description of the clinical
event. Preferred terms are linked to ‘high-level terms’, which provides a code for
qualitatively similar conditions. As an example, the preferred terms ’anxiety’ and
‘nervousness’ share the same high-level term ‘anxiety’. In this way clustering of
ADRs for analysing purposes is possible. The suspected drugs as well as
concomitantly used drugs are coded according to the ATC classification.
Design
Associations described in the quarterly reports to the MEB may represent possible
new signals of possible ADRs, drug-drug interactions but also reviews of drugs
that were marketed recently in the Netherlands. In a case control design, factors
associated with the selection and dissemination of signals to be reported to the
MEB were evaluated. All signals described in the quarterly reports from the first
publication in the second quarter of 1997 until the third quarter of 2000 were
reviewed. Cases were defined as associations that could be expressed as a
combination of a single ATC code and a single WHO preferred term. The
following exclusion criteria were applied: 1) associations concerning a group of
related substances or related ADRs, like ‘neuropsychiatric ADRs’ or ‘allergic
128
reactions’, 2) ‘review’ articles 3) drug-drug interactions, and 4) associations

concerning a single ATC code and single preferred term but associated with an
additional factor, for example gender or a special circumstance. An example of
the latter one is for instance a possible withdrawal syndrome during the use of
labetolol in a newborn infant.
The specific report that triggered the particular association to be selected as a
signal was called the ‘index report’. For each case, four controls were selected.
Controls consisted of those associations, not being published in the quarterly
reports, which were described in reports received by Lareb, 10 and 20 reports
previously and 10 and 20 reports later than the index report. In this way cases and
controls were matched in calendar time. In case a physician or a pharmacist
described two or more associations on the reporting form, the first association that
was mentioned was chosen. Similar to the ‘index report’ the reports that were
used to select the control associations are called ‘control reports’.
Factors
Factors possibly related to the selection of signals were distinguished in four
different types, namely the fact whether the association or drug is new, factors
related to the strength of the association, factors related to the seriousness of the
reaction involved and factors related to the documentation of the reports to the
SRS. The following factors were studied:
The fact whether the association or drug is new
-In the event the suspected ADR is mentioned in the Dutch text books
‘Farmacotherapeutisch Kompas’ or the ‘Informatorium Medicamentorum’, being
annually updated standard works often consulted by Dutch physicians or
pharmacists in the Netherlands, the ADR was considered labelled, otherwise
unlabelled.5,6
-Period of marketing of the drug involved. A distinction was made between the
drugs that have been marketed for less than five years on the Dutch market.
Factors related to strength of the association
-The total number of reports received on this particular association at the time the
index or control report was received. Three categories were defined: 1-2 reports,
3-4 reports and more than 4 reports.
129
Chapter 3.1
-The presence of a disproportionate association. For every association,

disproportionality was calculated.7 This was done by creating a 2x2 contingency
table for every association involved (Figure1). For calculating the number of
reports concerning the suspected drug, the full ATC code was used, for
calculating the number of ADRs, the preferred term was used. As a measure of
disproportionality, the corresponding Reporting Odds Ratio (ROR) was
calculated.8-10 The ROR can be calculated as a*d/b*c (see Figure 1) and expressed
as a point estimate with the corresponding 95% confidence interval.
Reported suspected ADR

Suspected ADR of the Other suspected
association involved ADRs
Drug Suspected drug of the

association involved a b
Other suspected drugs c d
Figure 1. 2x2 contingency table used for the calculation of ADR Reporting Odds Ratios
An association was considered to be disproportionate in case that the lower limit

of the 95% confidence interval was greater than 1. In the event the denominator of
the fraction is zero, the ROR can not be calculated. This may occur in the event of
a rare ADR. In this situation, the association involved was considered to be
reported more frequently than its background frequency and having the same
impact on the assessors as a statistically significant ROR.
Factors related to the seriousness of the reaction involved
-The index or control was considered serious in the event of death, a life
threatening situation, (prolonged) hospital admission, disability, or congenital
malformations.
-The presence of a so-called ‘critical term’. Critical terms are a subset of the
WHO preferred terms, being indicative of serious disease states, which can be
regarded as important to follow up.11 For this reason, critical terms may be of
particular interest for signal generation.
Factors related to documentation of the reports.
-The dechallenge of the index or control report is positive, i.e. the ADR
disappeared after cessation of the drug.
130
-The rechallenge of the index or control report is positive, i.e. the ADR
reappeared after the drug was used again.
The fact that the reports was sent by the physician (in attendance) of the patient.
Analysis
Logistic regression analysis was used to analyse the influence of the various
factors. In the first part of the study a univariate analysis was carried out. Odds
ratios were expressed as point estimates with corresponding 95% confidence
intervals. Factors that were statistically significantly associated with the selection
of signals were analysed in the second part of the study in a multivariate analysis.
Finally, the independent statistically significant factors resulting from the
multivariate analysis were subsequently analysed in more detail. With respect to
the selection of signals, the performance expressed as sensitivity, specificity,
positive and negative predictive value of both the separate factors as well as the
combination of these factors, was calculated in respect to the manual selection of
the associations in the signal selection process as the gold standard. For logistic
regression analysis the statistical package SPSS 10.0 was used, for calculating the
parameters of performance Microsoft Excel 97 was used.
Results
Between the second quarter of 1997 and the third quarter of 2000 a total number
of 76 associations were published in the quarterly reports to the MEB. Sixteen
associations were excluded, because they concerned a group of related substances
or ADRs, 7 associations were excluded because they were published as a ‘review’
article, another 7 associations were excluded because they concerned possible
drug-drug interactions, and finally 4 associations were excluded because an
additional factor was involved. A total number of 42 signals were included in the
analysis and were matched with 168 controls. An overview of the selected signals
is shown in Table 1.
131
Chapter 3.1
Table 1. Selected signals from the quarterly reports to the MEB from the 2nd quarter of
1997 until the 3rd quarter of 2000
Association Date of publication
norfloxacin-fixed drug eruption 2nd quarter 1997
oxybutinin - hallucination 2nd quarter 1997
losartan - taste disorder 2nd quarter 1997
mefloquin - convulsions 3nd quarter 1997
paroxetin - restless legs syndrome 3nd quarter 1997
losartan - angiooedema 4nd quarter 1997
cisapride - QT prolongation 4nd quarter 1997
lamotrigin - death 4nd quarter 1997
terbinafin - arthralgia 1nd quarter 1998
lithium - decrease in libido 1nd quarter 1998
miconazol - influence op protrombin time 1nd quarter 1998
tramadol - micturition disorder 2nd quarter 1998
irberstartan - angiooedema 2nd quarter 1998
miconazol oral gel - chocking 2nd quarter 1998
rulizole - trombopenia 2nd quarter 1998
vigabatrin - visual field defect 3nd quarter 1998
tolcapone - leucopenia 3nd quarter 1998
nefazodone - priapism 4nd quarter 1998
olanzapin - death 4nd quarter 1998
fexofenadin - QT prolongation 4nd quarter 1998
sildenafil - death 4nd quarter 1998
itraconazol - dyspnoea 1nd quarter 1999
diclofenac - anaphylactic reaction 1nd quarter 1999
quetiapine - leucopenia 1nd quarter 1999
diclofenac - haemolytic anaemia 2nd quarter 1999
oral budesonide - anaphylactic reaction 2nd quarter 1999
atorvastatin - rhabdomyolysis 2nd quarter 1999
interferon alfa 2B - Raynauds syndrome 2nd quarter 1999
alendronate - alopecia 3nd quarter 1999
lamotrigin - sialoadenitis 3nd quarter 1999
valproic acid - parkinsonism 3nd quarter 1999
metronidazol - hepatitis 4nd quarter 1999
valproic acid - polycystic ovary syndrome 4nd quarter 1999
acitretin – tast loss 4nd quarter 1999
simvastatin - eczema 1nd quarter 2000
loperamide - urinary retention 1nd quarter 2000
co-trimoxazole - tremor 2nd quarter 2000
lamotrigin - Stevens-Johson syndrome 2nd quarter 2000
minocycline - interstitial pneumonia 2nd quarter 2000
clopidrogel - trombotic trombocytopenic purpura 3nd quarter 2000
rofecoxib - death 3nd quarter 2000
pergolide - pulmonary fibrosis 3nd quarter 2000
132
Univariate analysis
Table 2 shows the differences between cases and controls and the results of the
univariate analysis. The absolute number of reports concerning the association
does not contribute in the selection of signals. However, the proportion of
associations with a statistically significant ROR, is higher among the cases (Odds
Table 2. Differences between cases and controls and the results of the univariate analysis
Cases Controls Odds ratio
n (%) n (%) (95% CI)
n=42 n=168
How new is the association or the

drug?
ADR unlabelled 27 (64) 67 (40) 2.7 (1.3-5.5)

Suspected drug shorter than 5 years 17 (41) 39 (23) 2.2 (1.1-4.6)
marketed
Factors related to strength of the

association
Absolute number of reports

1 or 2 reports (ref. category) 22 (52) 69 (41) 1 (ref)
3 or 4 reports 8 (19) 28 (17) 0.9 (0.4-2.3)
more than 4 reports 12 (29) 71 (42) 0.5 (0.2-1.2)
ROR full ATC code / preferred
term statistically significant (lower
limit 95% CI >1) 23 (55) 46 (27) 3.2 (1.6-6.4)
Factors related to the seriousness

of the reaction involved
presence of a ‘serious’ ADR 21 (50) 20 (12) 7.4 (3.5-16)

WHO critical term present 26 (62) 30 (18) 7.5 (3.6-16)
Factors related to documentation of

the reports to the SRS
dechallenge positive 17 (41) 60 (36) 1.2 (0.6-2.5)

rechallenge positive 4 (9.5) 16 (9.5) 1.0 (0.3-3.2)
Source of reports: number of 31 (74) 107 (64) 1.6 (0.8-3.4)
reports by physicians
133
Chapter 3.1
Ratio 3.2, 95% CI 1.6-6.4). Also the proportion of unlabelled ADRs is statistically
significant higher among the cases (OR 2.7, 95% CI 1.3-5.5), as well as the
proportion of drugs that are shortly marketed than five years (OR 2.2, 95% CI
1.1-4.6).
The impact of a ‘critical’ term (OR 7.5, 95% CI 3.6-16) present among the cases
is comparable with the presence of a ‘serious’ (OR 7.4, 95% CI 3.5-16) report.
Information regarding the quality of the documentation of the reports like the
presence of a positive de- or rechallenge, or the source of the reports does not
make a statistically significant contribution to the selection of possible signals.
Multivariate analysis
In the second part of the study, the influence of factors that were positively
associated with signal selection in the first analysis was analysed in a multivariate
logistic regression analysis. The results are shown in Table 3. The presence of a
‘serious report’ (OR 3.8, 95% CI 1.3-11), a WHO ‘critical term’ (OR 4.7, 95% CI
1.8-13), the ADR being unlabelled (OR 6.1, 95% CI 2.3-16) and the presence of a
disproportionate association (OR 3.5, 95% CI 1.4-8.4) were all independently
associated with signal selection. The time since marketing of the suspected drug
apparently did not make an additional contribution to the selection of signals.
Table 3. Results of the multivariate logistic regression analysis

Odds ratio
(95% confidence interval)
ROR full ATC code / preferred term statistically 3.5 (1.4-8.4)

significant
Index or control report concerns a ‘serious’ ADR 3.8 (1.3-11)
Critical term present 4.7 (1.8-13)
ADR unlabelled 6.1 (2.3-16)
Suspected drug shorter than 5 years marketed 1.4 (0.5-3.5)
Performance of separate factors

Table 4 shows sensitivity, specificity, positive and negative predictive values for
the factors contributing significantly to the selection of cases. Sensitivity is
highest for the ADR being unlabelled (0.64), but this factor also has the lowest
134
specificity (0.60) and lowest positive predictive value (0.28). With respect to the
negative predictive value all factors are comparable.
Table 4. Sensitivity, specificity, positive predictive value and negative predictive value of
independent factors contributing to the signal selection process
Sensitivity Specificity Positive Negative
predictive predictive
value value
Individual factors
ROR full ATC code / preferred 0.55 0.73 0.33 0.87

term statistically significant
Index or control report concerns a 0.50 0.88 0.51 0.88
‘serious’ ADR
Critical term present 0.62 0.82 0.46 0.90
ADR unlabelled 0.64 0.60 0.28 0.87
Combined factors
One or more factors present 1 0.29 0.26 1

Two or more factors present 0.72 0.80 0.48 0.92
Three or more factors present 0.52 0.95 0.53 0.89
All four factors present 0.07 0.99 0.6 0.81
Discussion
Our study showed that the seriousness of the reaction, the presence of a critical
term, a disproportionate number of associations in the database and the fact that
the suspected ADR is unlabelled, all play a role in the signal selection process.
Although slight differences exist, all four independent factors resulting from this
analysis are more or less comparable with respect to the performance.
Factors responsible for the selection of signals were divided in four categories,
being the fact if the association or the drug involved was new, factors related to
the strength of the association, the seriousness of the association and factors
related to the documentation of the reports. Except for the latter one, all types
seem to have influence on the selection of these signals for dissemination to the
135
Chapter 3.1
MEB. Except for the factors related to the documentation, the aforementioned
categories are also part of the ‘SNIP’ criteria.3 A retrospective analysis of
pharmacovigilance topics published between 1973 and 1990 in the Netherlands
revealed that 46% of the topics was new and 57% referred to established
products.12 Although the definition of the various factors slightly differ, the
results of our study are in accordance with these findings. In our study, 64% of
the ADRs were unlabelled and 60% referred to established products.
The quality of the reports is essential for an optimal assessment and for
maintaining a high quality of the database. Factors indicating the presence or
absence of information about a possible dechallenge or rechallenge in the index or
control report and the source of the reports, however, were not statistically
significant. The way the reports are presented seems not to be decisive in the
selection of signals. Nevertheless, the level of documentation of a report also
involves other aspects, like the completeness of the medication history or the
absence or presence of detailed clinical information. Therefore, further research
will be necessary to gain insight in the contribution of the level of documentation
of the reports in the selection of possible signals.
Not only the information present in the index report may have contributed to the
selection of an association, but also other reports that were received previously
may have contributed to the selection. For this reason, we analysed the
contribution of information concerning the fact whether a report was ‘serious’, the
presence of a critical term, and the presence of a positive dechallenge or
rechallenge in any other report than the index or control report. None of these
factors, however, had an additional contributed to the selection of signals
compared with the information present in the index of control report.
Spontaneous Reporting Systems for ADRs have been used over the past 40 years.
Since the amount of data is increasing there is a growing need to additional
quantitative signal detection techniques. Napke used a so-called pigeonhole
system in which every incoming report was colour coded.13 The optical impact
focussed the attention to specific associations. In the mid 70’s the first ideas for
computer programmes were developed.14,15 The development and implementation
of these techniques, which are mainly based on searching for a disproportionate
number of associations, is still increasing.9,16 Recently new approaches have been
developed like the Bayesian Confidence Propagation Neural Network.17,18 Also
techniques for the identification of possible drug-drug interactions and syndromes
136
are being developed.19 Quantitative techniques, however, cannot replace the

traditional case-by-case approach, but serve as an additional tool in signal
analysis. Although information technology may be helpful in identifying possible
signals, possibilities for including clinical information in this decision making
process are not yet available. The results of our study showed that, in contrast to
the absolute number of reports sent to the SRS, disproportionality is a predictive
factor for selecting possible signals. However, in the event the numerator is zero,
a ROR can not be calculated. This is a drawback for the use of this measure, but
since the assessors are familiar with the interpretation of the ROR, it was also
used in our analysis. The ROR could not be calculated for two index cases and
one control case. In all three cases the poison probability was calculated, for
which p<0.001. For this reason, in the event a ROR could not be calculated, the
association was considered to be disproportionate present in the database.
In the present study, the ROR was calculated with respect to the full ATC code
and the preferred term, but the ROR can be calculated in various other ways. By
choosing another level of aggregation with respect to the suspected drugs or
ADRs, different information can be obtained. For instance, in the event the ROR
is calculated based on the first five positions of the ATC code (ATC5) and the
preferred term, information is provided concerning the occurrence of the
suspected ADR in chemically related substances in the database. Similarly, the
Reporting Odds Ratio can be calculated concerning the full ATC code (ATC7)
and the high level term, which provides information concerning the occurrence of
the related ADRs associated with the suspected drug in the database. Finally, the
Reporting Odds Ratio concerning the first five positions of the ATC code (ATC5)
and the high level term can be calculated, which provides information concerning
the occurrence of the related ADRs in chemically related substances. All these
various approaches, however, yielded similar results as the calculation of the
ROR based on the full ATC code and the preferred term, indicating that these
other approaches did not provide any additional information on top of the normal
approach.
In preparation for the weekly assessment meeting where the selection of possible
signals take place, concise information concerning the reports is presented to the
assessors on an overview form. This concerns information about the gender and
age of the patient, the source of the reports, a description of the event, coding of
the suspected drug and suspected ADR, time of onset of the ADR and the fact
137
Chapter 3.1
whether the suspected ADR is labelled or unlabelled. Furthermore, quantitative

information is provided like the number of associations reported, the number of
reports on the ATC code, the number of reports on the high-level term and the
standard residual value as a measure of disproportionality. The assessors may
refer to the database for additional information like a more extensive description
of the clinical event and the concomitant medication the patient used. Since 1999
also information concerning the extent of disproportionality in the Lareb database
is presented as a ROR with corresponding 95% confidence interval. Data
concerning the presence of the association in the WHO database are available but
are not presented in advance. Although for instance information about the extent
of disproportionality, the fact if the ADR is labelled and the seriousness of the
report is available, selection bias may be present, but this reflects the present
procedure of selection and dissemination of signals.
To make a distinction between ‘unlabelled’ and ‘labelled’ associations, two
standard works that are frequently used in daily practice by physicians and
pharmacist were used. Historically, coding of the reports at Lareb always required
the use of these two standard works and not the official SPC text. Although slight
differences with the official SPC occasionally may exist, this coding was also
used in this study.
Signals published in the quarterly reports have a preliminary character. After all,
the goal is to inform the MEB about possible new signals in a rather early stage.
This implies that the causality of the signals mentioned in Table 1 does not
necessarily have to be proven. Selected cases serve as a point of attention for
further research or attention for other cases that might be reported. Media
attention of previous publication of a case report might raise the number of
reports on an association, which may cause reporting bias. Although this can not
completely be ruled out in individual reports, this does not necessarily have to be
unfavourable, since this may lead to an early warning of the association under
concern.
The results of this study may be used to improve the signal selection process, by
making a pre-selection of associations based on the presence of one of these
factors. However, the performance of these factors is based on calculations on the
dataset of the Netherlands Pharmacovigilance Foundation. On other datasets
differences of the performance may exist.
138
Conclusion
This study showed that selection of signals generated by the subjective review of
data sent to a spontaneous reporting system for adverse drug reactions is based on
both qualitative and quantitative aspects. Both the extent of disproportionality as
well as the seriousness of the reports, the presence of a ‘critical term’ and the fact
whether or not the association under concern is labelled, seem to have a
comparable impact in the process of the selection signals to be disseminated. The
results of this study revealed which factors are primarily responsible for signal
selection and dissemination. A better understanding of these factors may
eventually improve the efficiency of the signal selection process.
References
1. In: Inman WHW (ed). Monitoring for Drug Safety. Lancaster: MTP press, 1986.
Lancet 2000;356:1255-9.
3. A view from regulatory agencies. In: Graham DJ, Waller P, Kurz X. Strom BL (eds).
Pharmacoepidemiology. New York: John Wiley & Sons ltd, 2000:109-24.
5. van der Kuy A (ed). Farmacotherapeutisch Kompas 2000/2001. Commissie Farmaceutische
Hulp van het College voor Zorgverzekeringen, Amstelveen, 2000.
6. Informatorium Medicamentorum 2000 Wetenschappelijk Instituut Nederlandse Apothekers,
's Gravenhage, 2000.
7. Van Puijenbroek EP, Bate A, Leufkens HGM, Lindquist M, Orre R, Egberts ACG. A
comparison of measures of disproportionality for signal detection in spontaneous reporting
systems for Adverse Drug Reactions. (submitted for publication)
8. Stricker B.H.Ch and Tijssen J.G.P. Serum sickness-like reactions to cefaclor. J Clin
Epidemiol 1992;45:1177-84.
9. Egberts AC, Meyboom RH, De Koning FH, Bakker A, Leufkens HG. Non-puerperal
1997;44:513-8.
11. Lindquist M, Edwards IR, Fucik H, Nunes HM, Ståhl M. From association to alert - A
revised approach to international signal analysis. Pharmacoepidemiology and Drug Safety
1999;8:S15-S25.
139
Chapter 3.1
12. Meyboom RHB, Gribnau FWJ, Hekster YA, de Koning GHP, Egberts ACG. Characteristics
of topics in pharmacovigilance in the Netherlands. Clin Drug Invest 1996;12:207-19.
13. Napke E. Canada. In: Inman WHW (eds). Monitoring for Drug Safety. Lancaster: MTP
press, 1986:65-6.
10.
15. Finney DJ. Statisical logic in the monitoring of reactions to therapeutic drugs. Methods Inf
Med 1971;10:237-45.
16. Egberts ACG, Van der Hofstede JW, Meyboom RHB, et al. In: Egberts ACG,
Transformation of a database of spontaneously reported suspected adverse drug reactions
and its use as a tool in signal detection. Pharmacoepidemiologic approaches to the evaluation
of antidepressant drugs (Thesis) Utrecht University, Utrecht, the Netherlands 1997:111-24.
1998; 54:315-21.
18. Orre R, Lansner A, Bate A, and Lindquist M. Bayesian neural networks with confindence
93.
19. Van Puijenbroek EP, Egberts ACG, Meyboom RHB, and Leufkens HGM. Signaling
possible drug-drug interactions in a spontaneous reporting system: Delay of withdrawal
bleeding during concomitant use of oral contraceptives and itraconazole. Br J Clin
Pharmacol 1999;47:689-93.
140
3.2
PRACTICAL APPLICATION OF QUANTITATIVE SIGNAL

DETECTION IN THE SPONTANEOUS REPORTING SYSTEM
FOR ADVERSE DRUG REACTIONS IN THE NETHERLANDS
Chapter 3.2
Introduction
Despite extensive research and large pre-marketing trials, only limited

information concerning possible adverse drug reactions (ADRs) of a drug is
available at the time of marketing. For this reason, there is a clear need for a
continuous and systematic surveillance of drugs for any unexpected ADRs. In the
last few decades spontaneous reporting systems (SRSs), which are used to
monitor the safety of drugs after marketing, have earned a reputation for offering
a fast and reasonably efficient (and cost-effective) way of detecting ADRs.1
Health professionals (physicians and pharmacists) report suspected associations
between ADRs and drugs to an SRS on a voluntary basis. The process of signal
detection, i.e. the search for unexpected associations, is usually carried out by a
systematic manual review of all reports sent to an SRS. This procedure is time
consuming and more complex associations between patient characteristics, the
reported ADR(s) and suspected drug(s) are difficult to establish. Methods have
been developed to support the process, like the statistical quantitative analysis of
the data set of an SRS. Such quantitative analyses can provide additional
information concerning a possible relationship between a suspected ADR and a
drug. Several approaches were introduced, for instance the signalling of changes
in trend or the ‘reaction proportion signalling’ introduced by Finney, involving
the comparison of records of a particular drug and a specific suspected ADR with
those of a larger set of drugs and reactions.2-4 Over the years, modifications of the
latter approach have been developed and implemented, such as the Proportional
ADR Reporting Ratio used by the Medicines Control Agency5 and the Reporting
Odds Ratio.6-8 Currently, several such (modified) quantitative approaches are
being used by a number of SRS centres as well as the WHO Collaborating Centre
for International Drug Monitoring (the Uppsala Monitoring Centre).5,8-11 Despite
the increasing popularity of these new approaches, they are as yet not routinely
applied. At the Netherlands Pharmacovigilance Foundation Lareb quantitative
signal detection and subsequent evaluation of signals has been used in various
ways for a number of years. In this chapter, the role and position of quantitative
signal detection and the way it is applied at Lareb will be described.
142
Application of quantitative signal detection
Quantitative signal detection as applied by the Netherlands

Pharmacovigilance Foundation Lareb
Each report received by the Netherlands Pharmacovigilance Foundation Lareb,

which institute manages and maintains the national SRS on behalf of the Dutch
Medicines Evaluation Board, is subjected to review by qualified assessors. Data
concerning the suspected ADR(s) and drug(s) are coded using the WHO Adverse
Drug Reaction terminology12 and the Anatomical Therapeutic Chemical (ATC)
classification system, respectively, and subsequently filed in a database. If the
suspected ADR is mentioned in the Dutch textbooks ‘Farmacotherapeutisch
Kompas’ and/or the ‘Informatorium Medicamentorum’ the ADR is considered
labelled, and if this is not the case it is classified as unlabelled.13,14 All new reports
are discussed on a case-by-case basis in weekly assessment meetings. Here,
potentially relevant associations (signals) are selected for dissemination to the
Medicines Evaluation Board and/or publication in national and international
journals. Every incoming report represents an association between one or more
suspected ADRs and one or more suspected drugs, along with patient
characteristics. For the analysis of disproportionality a case/non-case design is
used. The extent to which the association between ADR and suspected drug
stands out with respect to its background frequency in the database a Reporting
Odds Ratio (ROR) is calculated.6-8 Based on the classical 2x2 contingency table,
for each association of drug and ADR, the following values are calculated (see
Figure 1).
Reports with the suspected Reports without the

ADR suspected ADR
Reports with the suspected drug a b

All other reports c d
Figure 1. The 2x2 contingency table for calculation of disproportionality.
Cell a represents the number of reports in which the same combination of

suspected drug and suspected ADR was mentioned. Cell b represents the number
of reports concerning the suspected drug but now in association with other
possible ADRs and cell c the number of reports concerning the suspected ADR
143
Chapter 3.2
associated with other drugs. Finally, cell d reflects the number of reports
concerning other drugs associated with other ADRs. The ROR is defined as the
product of the exposure odds among the cases with respect to the exposure odds
among the non-cases and is calculated by (a*d)/(b*c). The 95% confidence
interval is calculated by:
æ 1 1 1 1 ö
) ± 1 . 96 ç + + + ÷
95 % CI = e ln( ROR è a b c d ø
At the Netherlands Pharmacovigilance Foundation the ROR is used to analyse the

extent of disproportionality. Several approaches for calculation of the extent of
disproportionality are currently available, all having their own specific
applicability. For instance the WHO Centre for International Drug Monitoring
(Uppsala Monitoring Centre (UMC) uses a so-called Bayesian Confidence
Propagation Neural Network (BCPNN) analysis to isolate those associations that
stand out as being different to the generality of the database.15 Although several
measures of disproportionality can be used, the results of the various approaches
do not essentially differ, when applied to the Lareb dataset, if more than three
reports are included (Chapter 1.1). Since there is no golden standard concerning
the various measures of disproportionality,16 validation studies are few and far
between. Recently a validation study had been carried out concerning the
aforementioned BCPNN analysis of the WHO. When used for the analysis of the
WHO database, this approach yielded a positive predictive value of 44% and a
negative predictive value of 85% in the detection of signals as compared to
reference literature sources.11
The purpose of the quantitative signal detection as applied at the Netherlands
Pharmacovigilance Foundation is threefold. First, the results of the statistical
analysis are used as an aid in the traditional case-by-case review. Secondly, for
datamining purposes, a list of associations of ADRs and suspected drugs that are
disproportionately present in the database is generated periodically. Finally, on an
ad hoc basis, quantitative signal detection is used for evaluation purposes to study
signals in more detail, for example by adjusting for covariates or analysing more
complex relationships, like drug-drug interactions and drug-related syndromes.
144
Quantitative signal detection as an aid in the traditional

case-by-case review
In the case-by-case review the results of the quantitative approach serve as an

additional source of information. For the weekly assessment meetings a first
impression of the drug-ADR associations is compiled using overview forms
containing concise information on those reports that have been received in a
particular period. Figure 2 shows an example of such an overview form. In
addition to information regarding the patient’s age and gender, the suspected
drug, suspected ADR and the source of the report (either physician or
pharmacist), the results of the quantitative analysis are presented. Routinely, two
different RORs are calculated for this quantitative signal detection. Firstly, an
ROR is calculated on the basis of the number of reports with the generic
compound (expressed as the entire ATC code) and the suspected ADR. In
addition, a second ROR is calculated based on the first five positions of the ATC
code, representing the extent of disproportionality for chemically-related
substances, for instance groups of beta-blocking agents or ACE inhibitors. In the
overview form depicted in Figure 2 the results of the quantitative signal detection
of the Lareb database are shown. The first ROR shows the extent to which
‘enalapril’ is associated with reports of angioedema. This ROR is statistically
significant (11.0, 95% CI 7.0-17.2), implying that angioedema is significantly
associated with enalapril in reference to other reports in the database. The second
ROR refers to the association between angioedema and ACE inhibitors in general
and is also statistically significant (12.0, 95% CI 8.3-17.2), implying that ACE
inhibitors are indeed generally associated with this ADR.
To allow a more in-depth discussion of the reports, additional information may be
generated if the information on the overview forms is not sufficient to enable an
optimal assessment. This information generally concerns more detailed data
derived from the original reports sent in by the physicians or pharmacists,
additional information regarding the association between drugs and ADR(s)
reported in the literature and details with respect to concomitant medication used
by the patient.
145
Chapter 3.2
Report number: 30406 Date of entry: Jan 25th 2001 Region: Central districts of the Netherlands
Source: General Practitioner Gender: F Age: 51 Seriousness: not reported
Drug 1: RENITEC TABLET 10MG ATC code: C09AA02ENALAPRIL/ENALAPRILATE

Event 1: swelling of cheek and lips: angioedema WHO preferred term: ANGIOEDEMA
Drug 1 - Event 1: Labelled: yes
(a/b/c) ROR (95% confidence interval)

ATC 7-WHO preferred term: 59 / 708 / 336 10.99 (7.04-17.15)
ATC 5-WHO preferred term: 114 / 1476 / 281 11.97 (8.34-17.17)
Figure 2. Example of an overview form used in the assessment meetings at Lareb. Among
other data, two Reporting Odds Ratios with the corresponding numbers of the cells (a, b
and c) of the contingency table are shown.
Besides the results of the calculations based on the data set of the Netherlands
Pharmacovigilance Foundation, the outcome of the quantitative signal detection
on the UMC data set can be utilised. On a quarterly basis, the quantitative data of
each drug-ADR combination received by the UMC are sent to the national
pharmacovigilance centres.17 The UMC database contains summarised
information regarding all reports on combinations of drugs and ADRs, expressed
as the ATC code and the ‘preferred term’ of the WHO ADR terminology,
respectively. In this way the Lareb database is supplemented with quantitative
data from a second source, which thus provides additional insight into the
existence of possible signals. The reports from Lareb are in turn sent to the
Uppsala Monitoring Centre, so the reports are filed in both databases. However,
the total number of Lareb reports in the WHO-UMC database is relatively small
(approximately 30,000) compared to the total number of reports in the WHO
database (approximately 2,000,000). For this reason, it is unlikely that the Dutch
reports will have a great influence on the calculations based on the data set of the
UMC, which implies that the two sources of information can be regarded as
independent.
Quantitative signal detection as a datamining tool
Apart from having a function in case-by-case analyses, quantitative signal

detection can also be used more actively as a datamining tool. By screening the
database, overviews with disproportionate associations can be generated. A
disproportionate association may represent a signal and should consequently be
146
analysed in more detail.3,15 This approach can be used to identify disproportionate

associations that may have been missed by the traditional case-by-case approach.
With the datamining technique one or more additional filters can be applied. For
example, filtering the unlabelled cases allows a selection to be made of
associations that need to be monitored more closely because they may represent
unknown ADRs. Other filters may, for instance, be used to trace a specific drug or
a suspected ADR. As an illustration (see Table 1), Lareb has filtered out all
reports received between January 1st 1990 and November 1st 2000 in which a
suspected association with angioedema was mentioned. The associations are
sorted by the lower limit of the confidence interval of the ROR, in descending
order. In addition to the drug and suspected ADR the number of reports in the
cells of the aforementioned contingency table are shown, as well as the fact
whether the association is labelled or unlabelled. As expected, angioedema was
found to be mainly associated with ACE inhibitors and angiotensin II inhibitors.
The analysis of more complex relations such as drug-drug

interactions and drug-related syndromes
A report sent to an SRS contains information about one patient, one or more
(suspected) drugs and one or more suspected ADRs. However, all these factors
may be interrelated and since these complex relationships may not always be
transparent statistical analyses are often indispensable. An example is the
clustering between drugs in the search of possible drug-drug interactions. Logistic
regression analyses can be used to analyse this clustering by evaluating
interaction terms in the logistic model.18,19 Another example is the analysis of
syndromes, in which, similar to the detection of drug-drug interactions, the
clustering of suspected ADRs is analysed.20 These approaches have been
described in more detail in Chapters 2.2 to 2.4. Such statistical analyses can not
only be used for the evaluation of signals, but can also be used for the routine
screening of the database for drug-drug interactions and syndromes. For each
combination of two drugs it can be calculated whether the number of observed
ADRs exceeds the number of expected ADRs (Figure 3A), which may represent a
possible drug-drug interaction.
147
Table 1. Example of the datamining approach
Drug Labelled / ROR Lower limit Upper limit a b c
unlabelled 95% CI 95% CI
fosinopril labelled 32.8 14.2 75.4 8 19 307

enalapril with diuretics labelled 22.6 9.7 52.9 7 24 308
enalapril/enalaprilate labelled 12.9 9.3 17.7 50 346 265
lisinopril labelled 8.5 5.1 14.2 17 159 298
losartan with diuretics labelled 14.4 4.2 49.5 3 16 312
quinapril/quinaprilate labelled 10.1 4.0 25.9 5 38 310
captopril labelled 6.7 3.8 11.7 14 165 301
losartan labelled 4.8 2.8 8.2 15 246 300
irbesartan unlabelled 7.4 2.3 24.4 3 31 312
paracetamol labelled 5.5 2.0 15.2 4 56 311
isosorbide dinitrate unlabelled 5.6 1.7 18.2 3 41 312
rofecoxib labelled 4.3 1.6 11.9 4 71 311
acetylsalicylic acid labelled 4.2 1.3 13.4 3 55 312
The table shows disproportionality expressed as the ROR with corresponding 95% confidence interval and the number of reports in
cells a, b and c of the contingency table of various combinations of angioedema and reported drugs. In this example, the associations
are sorted by the lower limit of the confidence interval in descending order.
148
Although the reporting health professionals usually will make a distinction

between suspected and concomitant medication, in the analysis of drug-drug
interactions all drugs that are used at the moment of occurrence of the event are
included in the calculations. These calculations can be carried out for the various
ADRs in the database, represented by the cross-sections in Figure 3A. It is
important, however, to choose an ADR that is frequently associated with both
types of drugs. Similar to the detection of drug-drug interactions, the approach
can be applied to screen the database for drug-related syndromes. In this analysis
the calculations in the cross-sections concern the combination of two ADRs in
association with one particular drug that is more frequently reported than usual.
ADR Drug
Drug A ADR A
Drug B ADR B
A B
Figure 3. Datamining approach for (A) the detection of drug-drug interactions and (B)
drug-related syndromes
In collaboration with the Prevention and Health department of the Netherlands

Organisation for Applied Scientific Research (TNO), a graphical interface has
been developed that facilitates a fast and efficient screening of the database for
possible drug-drug interactions or drug-related syndromes. The interface displays
the cross-section of the cubes represented in Figure 3. An example of output
generated by this interface is shown in Figure 4.
149
Chapter 3.2
Combination of
itraconazole and
oral contraceptives
Itraconazole only
Figure 4. Example of the outcome of a drug-drug interaction analysis performed with the
datamining interface. The strength of the interaction between the two types of drugs is
represented by the intensity of the colour.
For various combinations of two types of drugs, the chance of a delayed

withdrawal bleeding was examined. The intensity of the colour in the cells
represents the level of statistical magnitude of the interaction. The intensity of the
colour in the marginal cells reflects the level of statistical significance for each
drug associated with the ADR separately. In addition, information concerning the
various calculations is displayed. The details of the possible itraconazole-
contraceptive interaction in association with delayed withdrawal bleeding are
described in more detail in Chapter 3.2. The concept of the detection of drug-drug
interactions may also be used to determine possible risk factors for developing
certain ADRs. If an apparent drug-drug interaction is in fact caused by an
increased susceptibility for developing an ADR in certain patients, other drugs
will possibly show the same type of drug-drug interaction in this category of
patients since the same confounding factor, i.e. patients with a similar condition,
is involved in both cases. Distinguishing between a drug-drug interaction and the
increased risk of developing an ADR due to the underlying disease may be rather
difficult.
150
Tracheal
sympaticomimetc drugs
Other tracheal drugs
Non-steroidal anti-
inflammatory drugs
Figure 5. Example of an interaction analysis of possible risk factors, using the datamining
interface. The strength of the interaction between the drugs is indicated by the intensity of
the colour.
However, if the analysis of several chemically or pharmacologically unrelated

drugs show an apparent ‘drug-drug interaction’ in relation to one or more ADRs,
it is more likely that a specific type of patient has an increased risk of developing
one or more of these ADRs. An example is the increased risk of developing
respiratory ADRs in patients with asthma taking non-steroidal anti-inflammatory
drugs (NSAIDs).21 The results of such an analysis produced by means of the
above-mentioned computer interface are shown in Figure 5. The concomitant use
of inhalation sympaticomimetic drugs, as well as other inhaled medication, and
NSAIDs will in all probability increase the number of reported ADRs involving
the respiratory system. It is unlikely, however, that an actual drug-drug interaction
between NSAIDs and both groups of unrelated drugs does exist. Thus, the
analysis described above constitutes a valuable contribution to the detection of the
151
Chapter 3.2
increased danger of respiratory ADRs occurring in patients suffering from asthma

or Chronic Obstructive Pulmonary Diseases (COPD) when using NSAIDs. The
aforementioned concept can be extended to the identification of other risk factors,
such as age, gender and, possibly in the near future, genotype.
Integrating information
Finney stated that the essence in ADR monitoring is to collect facts that
individually tell little but that collectively form a fabric of clues to drug dangers.2
Subsequently, an intelligent interpretation of the various combined aspects is
required before a conclusion about a possible relationship between suspected drug
and ADR can be drawn. Bringing together information from different sources is
of great benefit to the analysing process. The results of a quantitative approach
should be considered as additional information that is to be interpreted in
combination with other sources, such as clinical information and findings reported
in the literature (see Figure 6). Following this line of reasoning, it can be stated
that an association that is not disproportionately present in the database can still
represent a true signal while a disproportionately present association on the other
hand may not indicate a true signal. Human analysis and interpretation of the data
still is the decisive factor in the signal detection process. The method of statistical
analysis of SRS data sets, or quantitative signal detection, is currently gaining
ground. At the Netherlands Pharmacovigilance Foundation a quantitative
approach is routinely applied in various ways. By choosing different reference
groups comparisons with similar drugs or related ADRs can easily be made, thus
providing information from different perspectives. Again, the results of the data-
set analyses should be regarded as additional sources of information and
complementary to the traditional analysis techniques. Integration with clinical and
pharmacological assessments still remains an important principle in signal
detection.
152
Literature
Health
professional
Reports sent
to SRS
Medicines
Evaluation
Board
Quantitative
signal Overview Possible
detection forms and
signals
detailed
Science
reports
WHO Signal
combinations detection
database
Figure 6. Overview of information sources used for signal detection, and signal
dissemination routes at Lareb.
References
1. Rossi AC, Knapp DE, Anello C, O'Neill RT, Graham CF, Mendelis PS, Stanley GR.
Discovery of adverse drug reactions. A comparison of selected phase IV studies with
spontaneous reporting methods. JAMA 1983;249:2226-8.
2. Finney DJ. The detection of adverse reactions to therapeutic drugs. Stat Med 1982;1:153-61.
10.
4. Finney DJ. Statistical logic in the monitoring of reactions to therapeutic drugs. Methods Inf
Med 1971;10:237-45.
209.
7. Stricker BHC, Tijssen JGP Serum sickness-like reactions to cefaclor. J Clin Epidemiol
1992;45:1177-84.
1997;44:513-8.
153
Chapter 3.2
9. Egberts ACG, Van der Hofstede JW, Meyboom RHB. Transformation of a database of
spontaneously reported suspected adverse drug reactions and its use as a tool in signal
detection In: Egberts, ACG. Pharmacoepidemiologic approaches to the evaluation of
antidepressant drugs (Thesis) Utrecht University, Utrecht, the Netherlands,1997:111-24.
FDA spontaneous reprting system. The American Statistician 1999;53:177-89.
13. van der Kuy A (ed). Farmacotherapeutisch Kompas 2000/2001. Commissie Farmaceutische
Hulp van het College voor Zorgverzekeringen; Amstelveen, 2000.
14. Informatorium Medicamentorum 2000. Wetenschappelijk Instituut Nederlandse Apothekers;
's Gravenhage, 2000.
1998;54:315-21.
16. van Puijenbroek EP, Bate A, Leufkens HGM, Lindquist M, Orre R, Egberts ACG. A
comparison of measures of disproportionality for signal detection in spontaneous reporting
systems for adverse drug reactions. (submitted).
17. Lindquist M, Edwards IR, Fucik H, Nunes HM, Ståhl M. From association to alert - A
revised approach to international signal analysis. Pharmacoepidemiol Drug Saf 1999;8:S15-
S25.
18. van Puijenbroek EP, Egberts AC, Heerdink ER, Leufkens HG. Detecting drug-drug
interactions using a database for spontaneous adverse drug reactions: an example with
diuretics and non-steroidal anti-inflammatory drugs. Eur J Clin Pharmacol 2000;56:733.-8.
19. van Puijenbroek EP, Egberts AC, Meyboom RH, Leufkens HG. Signalling possible drug-
drug interactions in a spontaneous reporting system: delay of withdrawal bleeding during
concomitant use of oral contraceptives and itraconazole. Br J Clin Pharmacol 1999;47:689-
93.
20. van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens HGM. Association between
terbinafine and arthralgia, fever and urticaria: Symptoms or syndrome? Pharmacoepidemiol
Drug Saf 2001;10:135-142.
21. Oates JA, FitzGerald GA, Branch RA, Jackson EK, Knapp HR, Roberts LJ. Clinical
implications of prostaglandin and thromboxane A2 formation (1). N Engl J Med
1988;319:689-98.
154
GENERAL DISCUSSION
General discussion
Introduction
Quantitative signal detection has, in some form or another, always played a role in
pharmacovigilance. In the dawning of pharmacovigilance, information from
spontaneous reporting systems (SRSs) concerning associations between drug(s)
and unwanted clinical events was expressed in terms of the absolute number of
reports or crude estimates. Gradually, the need for a more considered
interpretation of the data became more urgent, which necessitated adjustments for
covariates such as age, gender, the source of the reports or the number of
prescriptions dispensed. This was illustrated by Rossi et al. who, in their 1987
study, showed that differences in upper gastrointestinal complaints associated
with the use of various NSAIDs reported to the Food and Drug Administration
were considerably reduced after adjustment for the underlying reporting rates.1
Although the first theoretical concepts of the quantitative approach to SRS data
were already described in the mid 1960s, it was not until recently that
developments in information technology enabled the routine application of these
techniques. In the past two decades, a limited number of SRSs and the WHO
Monitoring Centre for International Drug Monitoring have acquired experience
with quantitative or statistical approaches, although this has mainly been confined
to the association of a single drug and one particular adverse drug reaction
(ADR).
The Netherlands Pharmacovigilance Foundation Lareb was officially founded in
1991, but the first steps towards its establishment had already been taken as early
as 1984. Lareb runs and maintains a regionalised spontaneous reporting system.
One of the characteristics of Lareb is the short communication lines it maintains
with the reporting health professionals. Besides its function as an ‘early warning’
system, the Lareb methodology serves as a tool facilitating the improvement of
pharmacotherapy by providing the reporting physicians and pharmacists with
customised feedback. The initial concepts and procedures of Lareb have been
described in more detail by De Koning.2 As a result of new European regulations,
Lareb, acting on behalf of the Medicines Evaluation Board (MEB), was appointed
as the national spontaneous reporting centre for ADRs in the Netherlands in 1995.
In the mid 1990s Egberts had already laid a firm basis for a quantitative approach
at Lareb in which period he also reported his findings on studies into the
possibilities for the use of additional sources of information in the signal detection
156
General discussion
process.3 The qualitative theoretical concepts of pharmacovigilance were

subsequently elaborated by Meyboom.4 Due to their combined efforts the
Netherlands Pharmacovigilance Foundation now had a solid basis for further
enhancements of their signal detection process. More recently, additional
quantitative techniques have been developed and applied to supplement the
classical case-by-case approach in order to increase the sensitivity of the system
in picking up new and relevant signals. Furthermore, these quantitative
approaches can be more sensitive in detecting complex patterns within large sets
of reported suspected drugs and ADRs than people whose mental capacity is too
limited to recognise these hidden relationships. The studies presented in this
thesis have shown that the quantitative approach is an essential part of the
analysis procedures applied by the Netherlands Pharmacovigilance Foundation. It
does not only facilitate the detection and analysis of unknown associations
between a single drug and a single event, it also offers possibilities for the
analysis of more complex relations such as drug-drug interactions and drug-
related syndromes.
The first part of this thesis focused on methodological and statistical aspects of
quantitative signal detection. The results of the quantitative approaches applied in
various centres appeared to be broadly comparable when more than three cases
per combination of suspected drug and ADR were collected. Additionally, we
showed that non-selective underreporting does not influence the analysis of drug-
drug interactions, syndromes and covariate-drug interactions. In Part Two
examples were given concerning the analysis of the relationship between a single
drug and a suspected ADR, the detection of drug-drug interactions and drug-
related syndromes. Finally, in Part Three the main determinants of signal
detection were analysed and the application of quantitative signal detection at the
Netherlands Pharmacovigilance Foundation Lareb was described. The selection of
signals generated by the subjective review of data in a case-by-case analysis
appeared to be based on both qualitative and quantitative aspects. The extent of
disproportionality, however, proved to be more important than the number of
reports received per combination of suspected drug and ADR. In this fourth and
final part, additional methodological considerations and practical implications of
the use of quantitative signal detection in pharmacovigilance will be discussed.
Additionally, suggestions for further research are made.
157
General discussion
The added value of quantitative signal detection
Quantitative approaches are becoming increasingly important in signal detection.

This is not to say that they detract from case-by-case analyses, which method has
been an important tool in the analysis of SRSs for many years. Meyboom et al.
stated that each method applied in pharmacovigilance (e.g. spontaneous reporting,
prescription event monitoring or case control surveillance) follows a more or less
individual approach to signal detection.5 This not only applies to the classical
case-by-case analysis, in which each report is individually assessed as to whether
or not the reported association represents a signal, but also to quantitative signal
detection. Case reports or case series resulting from the former approach are
highly sensitive in picking up qualitative signals. On the other hand, they are
limited in their ability to provide quantitative information. Quantitative
approaches in signal detection unify the qualitative and quantitative aspects of
detection. However, it is exactly this dualistic character that is responsible for
both their strengths and their weaknesses. A quantitative approach is objective,
since calculations are not hampered by subjective information. Moreover, current
information technology allows calculations to be made in a short time span and
additionally permits concurrent adjustments for various covariates like age,
gender and concomitant medication. The main disadvantage of a quantitative
approach is the fact that clinical information can only be taken into account to a
limited extent. When used as a datamining tool, it should be considered as a
filtering mechanism to focus attention on those associations that are most likely to
represent true signals. For the analysis of possible signals resulting from a case-
by-case analysis it merely provides additional information. In other words, as the
next step in the detection and analysis of signals, thoughtful interpretation will
continue to be of vital importance.
As often with the introduction of new approaches, there is the risk that they are
regarded as a panacea for the limitations inherent to the prevailing approach.
However, a quantitative approach should be viewed as an additional source of
information, not as a substitute for the classical case-by-case analysis.
158
General discussion
Methodological considerations
As stated previously, clinical trials are limited in their power to detect ADRs.6
This is, among other reasons, caused by the small number of patients that are
studied before the drug is marketed and by the frequent exclusion of patients that
may be more susceptible to developing ADR(s), e.g. the elderly and those with
significant concurrent diseases. Additionally, the structured nature of trials, i.e.
specific doses of the drug under investigation are prescribed for limited periods of
time, makes them unsuited for the detection of ADRs with a long latency time.7,8
The likelihood of an ADR being detected depends on the frequency of the ADR,
the number of patients included in the trial and the background incidence of the
ADR. For example, with a sample size of 5,000 patients the certainty of finding at
least one case of an ADR with a true incidence rate of 1 in 1,000 would be more
than 99%. In the event the ADR cannot be clearly differentiated from the natural
background incidence, however, the sample size needed to detect this ADR would
increase dramatically.9 Although spontaneous reporting systems suffer from
certain methodological problems, and covariates are generally difficult to control
for, these systems have proven to be relatively efficient in the surveillance of
drugs for possible harmful effects.
Validity
With respect to the interpretation of the outcome of quantitative analyses of SRS
databases by means of measures of disproportionality, three different populations
can be distinguished (see Figure 1).
159
General discussion
Drug using population X
ADR experiencing population X'
ADR reported population X''
k1 incidence ADR
k2 reporting ratio
k1 k2
Figure 1. The different populations relevant for the interpretation of spontaneous

reporting
The largest population represents patients that actually use the suspected drug
(X). The second population consists of a smaller proportion of this group of
patients, namely those who experience an ADR (X'). Finally, the third and by far
smallest population is constituted by the number of patients of the latter group of
whom the suspected ADR is actually reported (X''). The incidence of the drug is
represented in Figure 1 by k1, reporting of the drug is represented by k2. In
quantitative analysis, the measure of disproportionality expresses the relationship
between drugs and suspected ADRs within the ADR reported population (X'').
Based on these calculations, an estimation of the level of disproportionality in the
ADR experiencing population (X') can be made. Still, for the practising health
professional it may be important to have information about the occurrence of an
ADR among patients in the drug using population (X) at his or her disposal. The
SRS, however, is primarily designed to generate hypotheses of these possible
160
General discussion
effects, and it cannot provide accurate information about the incidence of the
ADR. In other words, the results of the calculations cannot always be generalised
to the population of patients using the drug (X). This does not necessarily have to
be a problem, since the main goal of an SRS is to detect possible signals and not
to estimate the incidence of the ADRs involved.
An example may illustrate the misinterpretation that may occur if results of the
data of SRS are considered indicative of the incidence of ADRs. If we assume
that two drugs have a similar pattern of ADRs, calculations based on measures of
disproportionality would yield no differences, even if different incidences of the
ADRs exist in the patient population that is using the drugs.
In quantitative signal detection, the association between various factors is
expressed as a point estimate with corresponding confidence interval. This
information should not be considered as having an exact meaning, like the
incidence of the ADR, but rather as a tool that facilitates the interpretation of the
data in the context of other reports and co-factors in the database. Therefore,
interpretation of the quantitative results based on data sets of spontaneous
reporting systems should be carried out by assessors who are experienced in this
field and who have an expert knowledge of the composition of the database.
By calculating a measure of disproportionality, like the Reporting Odds Ratio, the
frequency of an adverse reaction in association with a certain drug is being
compared with the other reports in the database.10 The rationale for using this
approach is that the composition of different types of reactions in the data set of
an SRS is relatively constant.11 Other reports in the database are being used as a
measure of the ‘background incidence’ of the relevant events. All other reports in
the database are commonly used for reference purposes. Although this is a large
group, which has the advantage that smaller confidence intervals can be
calculated, its composition is also heterogeneous. If, for instance, liver-function
disorders are over-represented in a certain pharmacological group, being part of
the reference group, the number of ADRs needed for the drug in the index group
to yield a statistically significant result will also increase. There is a chance that
under these circumstances rare liver-function disorders associated with a new
drug will be less easily detected. This implies that great care should be taken in
choosing the reference group.
Only a small proportion of the ADRs occurring in daily practice will be reported.
This problem of so-called underreporting refers to factor k2 in Figure 1, and may
161
General discussion
be the cause of selection bias. Stephens gave an overview of the various factors
responsible for this phenomenon.9 Examples are factors associated with the drug,
e.g. the length of time the drug is on the market,12 or factors associated with the
ADR (e.g. the severity of the reaction or the unusual nature of the events). Also
miscellaneous factors like legal implications may be involved or the doctor’s
familiarity with the regulations.13 Estimations of the degree of reporting,
therefore, vary widely. For instance, a Swedish study showed that 80% of the
cases of osteitis occurring after BCG vaccination were reported.14 A study by
Moride et al., however, showed that general practitioners tend to only report 1 out
of 24,322 ADRs to the pharmacovigilance centre. In the latter study, the
underreporting was lowest for serious and unlabelled ADRs and for newly
marketed drugs.15 Underreporting is inherent to spontaneous reporting and,
unfortunately, no general indication of the extent to which it occurs can be given.
The best estimate may be obtained by an in-depth study of a sample of the
relevant population.9 Underreporting can either be non-differential or differential.
As shown in Chapter 1.2, calculations of RORs are not necessarily influenced by
non-differential underreporting but may be influenced by differential
underreporting. This holds for the analysis of associations between ADR and
suspected drug, drug-drug interaction, drug-related syndromes and drug-covariate
interactions. An example of non-differential bias is media attention for a specific
drug, which has a similar influence on both the numerator and denominator of the
ROR.16 If media attention is focussed on the specific association of the drug and
the ADR, however, this differential underreporting or bias will influence the
ROR.
Several other types of selection bias may also jeopardise the validity of
conclusions, for example the possibility that the characteristics of reported cases
differ from the non-reported cases in terms of severity of the ADR, time to onset
and the presence of risk factors.17 In the analysis of drug-drug interactions for
instance, patient characteristics or the concomitant medication may act as a
confounding factor. This does not necessarily have to be unfavourable, since it
may contribute to the detection of risk factors for developing ADRs as was
discussed in Chapter 3.2.
Reports sent to an SRS do not always represent true ADRs (so-called
misclassification). They may also represent symptoms or diseases incurred
because of other reasons that cannot be distinguished from true ADRs on clinical
162
General discussion
grounds. For this reason some drug-event combinations may be purely

coincidental18 and false positive signals (i.e. decreased specificity) may occur.
Bégaud showed that the probability of a significant number of coincidental
associations between a drug and suspected ADR is nil or remains low, unless the
symptom is very common. For rare ADRs, like type B effects, it is highly unlikely
that more than three coincidental reports will be received. Due to the problem of
underreporting, however, even for more common effects the number of
coincidental reports remains low.17 Therefore, underreporting does not pose a
fundamental threat to signal detection.
Practical considerations
Every new case sent to an SRS concerns a possible association between suspected
clinical events and drug(s), for which it should be decided whether it represents a
possible signal. This decision is based on weighing clinical interpretation of the
reports (and their context), the results of the quantitative analysis and other data
available (e.g. findings in the literature; see Figure 2). As was shown in Chapter
3.1, these various factors also play a role in the selection and dissemination of
signals to the MEB.
Clinical information
Other data Quantitative analysis
Figure 2. Factors involved in the analysis of possible signals
An SRS should be considered as a diagnostic tool for health authorities, health

professionals, and the industry, the results of which can be used to weigh benefit
and safety of the drugs involved. A balance must be found between the
sensitivity, specificity, and false positive and false negative signals in order to
establish an optimal signal-to-noise ratio. False negative signals are potentially
harmful since important ADRs may be missed. A large number of false positive
signals on the other hand may indirectly cause harm in that they may withhold
163
General discussion
patients from an effective treatment. For this reason, the threshold in quantitative
signal detection must be established very precisely.10
Similar to the detection of ADRs, drug-drug interactions may only be detected
after drugs have been marketed, which may have serious implications for the
drug(s) involved. An example is mibefradil, which was introduced in 1997. From
premarketing studies it was already known that mibefradil induced cytochrome
P450 3A4 and 2D6. Because of the characteristics of the population in which
mibefradil was used, frequent drug-drug interactions occurred, one of which was
with HMG-co-A inhibitors.19 Eventually, these frequent interactions caused this
drug to be withdrawn from the market.
Due to the growing availability of more potent drugs acting on fundamental
biochemical pathways or receptors, the number of drug-drug interactions is likely
to increase.20 The elderly are at particular risk of adverse drug-drug interactions,
and furthermore, the risk increases with the number of drugs prescribed and taken
concurrently.21,22 Apart from the ‘prescription-only’ drugs, in the Netherlands
many others are available as ‘over the counter’ medication. Examples are
NSAIDs or H2 blocking drugs that may interact with prescribed drugs. Since the
relative number of elderly people is growing and since an increasing number of
younger patients are taking combinations of drugs, as is the case in the treatment
of AIDS or the eradication of Helicobacter Pylori, the number of possible drug-
drug interactions occurring is also likely to rise. The studies reported in Chapter
2.2 and 2.3 have shown that data sets of spontaneous reporting systems can be
used in the detection and analysis of drug-drug interactions. Since it has been
estimated that between 6 and 30 percent of all ADRs are due to drug-drug
interactions,23 the investigation of this type of interactions constitute a challenging
new task for pharmacovigilance. Ideally, drug-drug interactions should be
searched for in large databases that contain reliable information on all drugs being
used by a particular patient. When a drug-drug interaction is already known and is
monitored for in the pharmacist or physician information systems he or she is
unlikely to report the interaction and thus selective underreporting may occur.
Another aspect that needs particular attention in the detection of drug-drug
interactions is the choice of the suspected ADR(s). Generally, a distinction is
made between various types of ADRs.24 Type A ADRs are related to the
pharmacological effects of the drug, are dose-related and tend to be fairly
common. They represent the largest group of ADRs. Type B effects are related to
164
General discussion
patient characteristics. They may concern allergic or idiosyncratic reactions and

occur only in a minority of the patients. Type C effects refer to the change in
frequency of natural diseases. An example is the possible association between
oral contraceptives and breast tumours. In the detection of drug-drug interactions,
it is essential to choose a relatively common ADR. The choice should be well
founded to ensure that the ADR is dose-related with the pharmacological effects
of the drugs involved, and secondly to ensure that the combination of both drugs
is mentioned in a sufficient number of reports. For this reason, type B and type C
effects are less suited for the detection of drug-drug interactions.
In a drug-drug interaction, both a pharmaceutical and a pharmacokinetic
interaction as well as a pharmacodynamic interaction may be involved, each of
which may lead to either an increase or decrease in the effect of one of the
interacting drugs. Although the mechanism of action is important for an
understanding of the nature of the interaction, in the detection of ADRs in a
spontaneous reporting system it is of less significance. The net effect of the
possible interaction, however, is essential, since it determines the choice of the
ADR(s) used for the detection of the interaction. If an increased effect of one of
the drugs is expected, the ADR should be associated with that particular drug. An
example is the possible interaction between oral contraceptives and the use of
itraconazole (see Chapter 2.2), in which a delayed withdrawal bleeding is taken to
be indicative of a possible interaction.
In the event that a possible drug-drug interaction causes a decreased effect of one
of the drugs, the ADR used to identify the drug-drug interaction should be
indicative of a lack of efficacy of one of the drugs involved. This was illustrated
in Chapter 2.3 where the influence of concomitant use of diuretics and non-
steroidal anti-inflammatory drugs (NSAIDs) on symptoms indicating a decreased
efficacy of diuretics was examined.
In Chapter 2.4 we showed that SRS data sets can be used in the detection and
analysis of possible drug-related syndromes. A syndrome can be seen as a
complex of signs and symptoms that, together, constitute the picture of a
disease.25 Examples of drug-related syndromes are the oculomucocutaneous
syndrome associated with the use of practolol26 and the possible association of
Reye’s syndrome and aspirin use.27 Another example is the serotonin syndrome
associated with the use of various drug agents affecting the serotonergic system,
among which the serotonin re-uptake inhibitors.28
165
General discussion
Only a small part of the ADR clusters present in the database actually represent
distinct clinical syndromes. Rather than being part of a certain clinical syndrome,
an apparent clustering of symptoms may occur due to fact that the symptoms
themselves are related. This is the case with nausea and vomiting or abdominal
pain and diarrhoea, which symptoms, though closely interrelated, do not represent
a particular clinical entity.
Recommendations for further research
Studies into the validation of quantitative approaches are still sparse, although
recently such a validation study has been conducted on the Bayesian Propagation
Neural Network analysis of the WHO.29 The problem with validation studies is
that they are hampered by the fact that there is, as yet, no true gold standard
available. Additionally, the validity of a quantitative approach strongly depends
on the data set to which it is applied. Our study, in which the various methods of
quantitative signal detection were compared, has shown that the results of the
various approaches are broadly comparable if associations are based on more than
three reports (Chapter 1.1). This does not necessarily imply, however, that similar
results to those of the validation study on the WHO database will be found for the
case/non-case design of the Lareb database. Therefore, validation of the
quantitative signal detection within the Lareb data set and those of other SRSs
needs to be corroborated by additional, more elaborate studies. One of the
possible approaches could be comparison with case reports of new ADRs
published in the literature. However, publication of case reports in national and
international journals is not primarily intended to give an early warning.29 This
implies that the goals of publication and signal detection in an SRS differ, which
hampers the validation process. Another possibility would be setting up an expert
panel that is to decide whether or not the associations reported do indeed
represent possible signals. The decisions of this panel could subsequently be used
as the validation standard.
The quantitative analyses of the data sets of SRSs give an impression about the
occurrence of ADRs within the population of patients experiencing an ADR
during the use of a certain drug. Adjusting for various covariates is already
possible. Nevertheless, also the implementation of data on drug utilisation would
be useful to correct for the number of prescriptions of the drugs involved.
166
General discussion
In the majority of reports sent to the Netherlands Pharmacovigilance Foundation

information on the concomitant drug use is included. For the analysis of
associations between ADR(s) and drugs, and for the analysis of drug-related
syndromes, however, only the suspected medication was used in the studies
presented in this thesis because the reporting health professionals did not attribute
the suspected ADR(s) to the concomitant medication. If we would have included
the concomitant medication in our investigations, which might then have yielded
different results. This approach was for instance used by Moore in his analysis of
the possible association between hypoglycaemia and the use of ACE inhibitors.30
The differences of these two approaches should be studied in more detail.
The extent to which clinical information can be taken into account is limited in
the current method of quantitative signal detection. At present, clinical
information first has to be coded, for instance by using the WHO ADR
terminology, before statistical analyses can be carried out. In the classical,
empirical assessment methods, however, clinical information has a more
prominent place. In order to improve the signal detection process, integration of
clinical details in the quantitative approach is recommended.
In the approaches discussed in the present thesis, the ATC code was used to code
the suspected drug. With the ATC classification chemically related drugs can also
be clustered to a certain extent. This may reveal associations between ADRs and
groups of chemically related drugs. However, in addition to the chemical
relationships, drugs may also be pharmacologically related even though they do
not have a similar ATC code. For instance, despite the fact that selective serotonin
reuptake inhibitors and triptans (serotonin agonists) belong to different ATC
classes, they may still share the same ADRs because of their similar
pharmacological activity. The use of different methods of clustering in
quantitative signal detection may provide a deeper insight into the relationship
between the chemical or pharmacological properties of the drugs involved and the
occurrence of certain types of ADRs. Concerning the analysis of drug-drug
interactions, aspects of metabolism of the drug under investigation can be studied,
such as the metabolism by the various cytochrome subtypes.
Finally, in Chapter 4.2 concepts of identifying risk factors in patients, like age,
gender or concomitant diseases, was described. As with the analysis of drug-drug
interactions, the relatively small size of the data set may hamper the analysis of
167
General discussion
these interactions and they should therefore preferably be carried out on larger
data sets.
Final remarks
Although the first initiatives of spontaneous reporting date from the middle of the
previous century, pharmacovigilance is continuously being refined. In addition to
the classical approach, where for each case it is assessed if the reported
association represents a new ADR, quantitative techniques are being developed to
facilitate the signal detection process. These new techniques enable a more
detailed analysis of patient characteristics, drugs and the ADRs involved, thus
enhancing the quality and widening the scope of pharmacovigilance. Given the
rapid advances in information technology, it is expected that quantitative signal
detection will establish itself as a standard source of information in spontaneous
reporting, which will clearly benefit the study into and the monitoring of the
safety of drugs after marketing.
References
1. Rossi AC, Hsu JP, Faich GA. Ulcerogenicity of piroxicam: An analysis of spontaneously
reported data. Br Med J 1987;294:147-50.
2. de Koning GHP. A regionalized spontaneous surveillance program for adverse drug
reactions as a tool to improve pharmacotherapy. (Thesis) Utrecht University, Utrecht, the
Netherlands, 1994.
3. Egberts ACG Pharmacoepidemiologic approaches to the evaluation of antidepressant drugs.
(Thesis) Utrecht University, Utrecht, the Netherlands, 1997.
4. Meyboom RHB. Detecting adverse drug reactions - pharmacovigilance in the Netherlands.
(Thesis) Catholic University of Nijmegen, Nijmegen, the Netherlands, 1998.
5. Meyboom RHB. Pharmacovigilance in perspective. Drug Saf 1999;21:429-47.
6. Rawlins MD, Thomson JW. Mechanisms of adverse drug reactions. In: Davis DM (ed).
Textbook of adverse drug reactions. Oxford: Oxford University Press, 1991:18-45.
7. Amery WK. Why there is a need for pharmacovigilance. Pharmacoepidemiol Drug Saf
1999;8:64.
8. Talbot JC, Nilsson BS. Pharmacovigilance in the pharmaceutical industry. Br J Clin
Pharmacol 1998;45:427-31.
9. Stephens MDB. Introduction. In: Stephens MDB, Talbot JCC, Routledge PA (eds).
Detection of new adverse drug reactions. New York: Grove's Dictionaries inc, 1998.
10.
168
General discussion

209.
12. Weber JCP. Epidemiology of Adverse Reactions to Nonsteroidal Anti-inflammatory Drugs
In: Rainsford KD, Velo GP (eds). Advances in Inflammatory Research. New York: Raven
Press, 1984:1-7.
13. Griffin JP, Weber JC. Voluntary systems of adverse reaction reporting-Part II. Adverse Drug
React Acute Poisoning Rev 1986;5:23-55.
14. Bottiger M, Romanus V, de Verdier C, Boman G. Osteitis and other complications caused
by generalized BCG-itis. Experiences in Sweden. Acta Paediatr Scand 1982;71:471-8.
15. Moride Y, Haramburu F, Requejo AA, Begaud B. Under-reporting of adverse drug reactions
in general practice. Br J Clin Pharmacol 1997;43:177-81.
1992;45:1177-84.
17. Begaud B, Moride Y, Tubert-Bitter P, Chaslerie A, Haramburu F. False-positives in
spontaneous reporting: should we worry about them? Br J Clin Pharmacol 1994;38:401-4.
18. Venning GR. Validity of anecdotal reports of suspected adverse drug reactions: the problem
of false alarms. Br Med J 1982;284:249-52.
19. Po AL, Zhang WY. What lessons can be learnt from withdrawal of mibefradil from the
market? Lancet 1998;351:1829-30.
20. Routledge PA. Adverse drug reactions and interactions: Mechanisms, risk factors, detection,
management and avoidance. In: Stephens MDB, Talbot JCC, Routledge PA (eds). Detection
of New Adverse Drug Reactions. New York: Grove's Dictionaries inc, 1998.
21. D'Arcy PF. Drug reactions and interactions in the elderly patient. Drug Intell Clin Pharm
1982;16:925-9.
22. Veehof L, Stewart R, Haaijer-Ruskamp F, Jong BM. The development of polypharmacy. A
longitudinal study. Fam Pract 2000;17:261-7.
23. Pirmohamed M, Orme ML. Drug interactions of clinical importance In: Davies DM, Ferner
RE, de Glanville H (eds). Davies's Textbook of Adverse Drug Reactions.London: Chapman
& Hall Medical, 1998:888-912.
25. Dirckx J (ed). Stedman's concise medical & allied health dictionary. Baltimore: Willams and
Wilkins, 1997.
26. Wright P. Untoward effects associated with practolol administration: oculomucocutaneous
syndrome. Br Med J 1975;1:595-8.
27. Halpin TJ, Holtzhauer FJ, Campbell RJ, Hall LJ, Correa-Villasenor A, Lanese R, Rice J,
Hurwitz ES. Reye's syndrome and medication use. JAMA 1982;248:687-91.
28. Ashton C, Young A. Drug-induced psychiatric disorders. In: Davies DM, Ferner RE , de
Glanville H (eds). Davies's Textbook of Adverse Drug Reactions. London: Chapman & Hall
Medical, 1998:669-731.
169
General discussion

1997;44:513-8.
170
Summary
The development of a new drug is a time-consuming process, in the course of

which attention is paid to quality, efficacy as well as safety. At the time of
marketing of the drug, the information concerning the safety is generally limited
and usually originating from a rather limited number of patients and studies. In
general, these studies are restricted to a selected group of patients. Furthermore,
the size as well as the duration of the trials is limited. As some adverse drug
reactions (ADRs) are rare or have a long latency period, and others only occur
after prolonged use of the drug or are restricted to specific patient groups, they are
generally difficult to detect or predict at this stage. The 1960s brought an
awareness of the importance of a systematic surveillance of drugs for ADRs, also
after marketing, with the detection of the association between thalidomide, a
hypnotic drug, and severe congenital malformations of the limbs of children
whose mothers had used this drug in the early stages of pregnancy. One of the
ways of post-marketing surveillance is ‘spontaneous reporting’. Health
professionals are urged (in most countries) or obliged by law to report their
suspicions of possible associations between drugs and ADRs to the centres
managing spontaneous reporting systems (SRSs). In the Netherlands, and on
behalf of the Medicines Evaluation Board, the Pharmacovigilance Foundation
Lareb is responsible for the management and maintenance of the SRS concerning
drugs that have been approved for the Dutch market. One of the aims of the SRSs
is the detection of signals indicating a possible association between a suspected
drug and reported ADR. Therefore, the individual reports are all being analysed
by experienced assessors. Next to this classical case-by-case analysis, also the use
of statistical approaches or quantitative signal detection may be an efficient tool
to analyse reports sent to an SRS. The application of numerical procedures in
pharmacovigilance, or quantitative signal detection, can not only be used for
signal detection and analysis of associations between a single ADR and a single
drug, but also for more complex relations like drug-drug interactions, drug-related
syndromes, risk factors in patients and time trends after marketing of the drug
under investigation. Statistical analyses provide an answer to the question if the
171
Summary
number of observed cases exceeds the number of expected cases or, worded
differently, if the association has been reported disproportionally. The objective
of this thesis is to explore the value and possibilities of quantitative signal
detection in spontaneous reporting systems and to add further conceptual insight
both into the analysis of the relationship between adverse drug reactions and a
suspected drug as well as into complex relationships like drug-drug interactions
and drug-related syndromes.
The first part of this thesis consists of two studies focussing on methodological
aspects of quantitative signal detection. In various centres, different measures are
used to quantify the extent to which an ADR is reported disproportionally to a
certain drug compared to the generality of the database. The objective of the study
in Chapter 1.1 is to examine the level of concordance of the various measures
when applied to the dataset of the Netherlands Pharmacovigilance Foundation.
Therefore, the use of the Reporting Odds Ratio, Proportional Reporting Ratio,
Yule’s Q, the Poisson probability and Chi-square test are compared with the use
of the Bayesian Confidence Propagation Neural Network analysis (BCPNN) of
the WHO Collaborating Centre for International Drug Monitoring. In general,
sensitivity is high and the specificity is fairly low with respect to the reference
measure. Additionally, this study shows that the different measures used are
broadly comparable when four or more cases per combination have been
collected.
In the studies presented in this thesis the ROR is frequently used as a measure of
disproportionality. One of the advantages of the use of this measure is the fact that
in the analysis of an association between a suspected drug and reported ADR, so-
called non-selective underreporting does not influence the height of the ROR. The
ROR can also be used in a logistic model in the analysis of clustering between
drugs for studying possible drug-drug interactions as well as in the analysis of
clustering of ADRs while studying possible drug-related syndromes. The study in
Chapter 1.2 shows that also in the analysis of complex relationships like drug-
drug interactions and drug-related syndromes, non-selective underreporting does
not influence the height of the ROR.
In the second part of this thesis, examples of the application of quantitative signal
detection are given. These examples concern the analysis of the association of
172
Summary
drug and an ADR as well as the analysis of drug-drug interactions and clustering
of ADRs. In Chapter 2.1 an example is given of the analysis of a possible
association between a drug and possible ADR. It is generally known that Non-
Steroidal Anti-inflammatory Drugs (NSAIDs) may rarely cause an anaphylactic
reaction. The objective of this study is to investigate whether the risk of
anaphylactic reactions being reported during the use of various NSAIDs is greater
than with other classes of drugs and if differences among NSAIDs exist. The
results of this study strengthen previous findings concerning the relative high risk
of developing an anaphylactic reaction during the use of NSAIDs, particularly
diclofenac, ibuprofen and naproxen. The study in Chapter 2.2 shows that datasets
of SRS can be use to signal possible drug-drug interactions, which are not
explicitly reported. In the event of a drug-drug interaction, one drug influences
the effect of another drug. This may subsequently cause an increase or decrease in
the number of reported ADRs of the latter drug. As an example of this approach
we analyse the ADR ‘delayed withdrawal bleeding’ during the concomitant use of
oral contraceptives and the antimycotic drug itraconazole. In a logistic model, the
combined use of both drugs is examined. The results strongly suggest that in
women using oral contraceptives and itraconazole concomitantly, the withdrawal
bleeding may be delayed due to a drug-drug interaction. Although this method has
been developed to detect unknown drug-drug interactions, in the study in Chapter
2.3 the known interaction between diuretics and NSAIDs is used to illustrate this
approach. In the dataset of the Netherlands Pharmacovigilance Foundation in
reports indicating the use of both types of drugs, the number of observed cases of
a decreased efficacy of diuretics exceed the number of expected cases, which
points at the existence of an actual drug-drug interaction. The studies in both
chapters illustrate that spontaneous reporting systems have a potential for signal
detection and the analysis of possible drug-drug interactions. The method
described may enable a more active approach in the detection of drug-drug
interactions after marketing.
Like the analysis of concomitant use of drugs in the analysis of drug-drug
interactions, datasets of SRS can be used to study the possibility of clustering of
symptoms indicating the existence of a possible drug-related syndrome, as is
illustrated in Chapter 2.4. Since the introduction of the antimycotic drug
terbinafine, the Netherlands Pharmacovigilance Foundation received eight reports
of arthralgia during the use of this drug. In four reports the additional presence of
173
Summary
skin reactions is mentioned, two of these reports concern urticaria. Two patients
who report arthralgia also have a fever. These reports are described in more detail,
and analysed in a logistic model in order to determine whether symptoms are
interrelated. The statistical analysis shows a clustering of these symptoms among
reports of patients using terbinafine. Presumably these symptoms have a shared
aetiology, possibly an immunological reaction.
In the third part of this thesis the implementation of quantitative signal detection
at the Netherlands Pharmacovigilance Foundation is discussed. Goal of the study
of Chapter 3.1 is to gain insight into factors contributing to the selection and
dissemination of possible signals from the Netherlands Pharmacovigilance
Foundation Lareb. Among the many potential signals that are identified, only a
limited number of signals are important enough to inform the Dutch Medicines
Evaluation Board (MEB) in a special procedure. In a case-control design, signals
disseminated to the MEB were compared with controls consisting of associations
between drugs and ADRs, which were not disseminated, to the MEB. Logistic
regression analysis is used to investigate the influence of various factors, such as
the fact whether the ADR or drug is new, the strength of the association, the
seriousness of the reaction and factors related to the documentation of the reports.
The analysis shows that the presence of a ‘serious report’, a ‘critical term’ of the
WHO ADR terminology, the ADR being unlabelled and the presence of a
disproportionate association are all independently associated with the selection of
a signal. Additionally, the findings show that the presence of, for example, two of
the aforementioned factors have a sensitivity of 0.72 and a specificity of 0.82 for
the association to be selected for dissemination to the MEB. Obviously, the
selection of signals is based on both qualitative and quantitative aspects. In
Chapter 3.2 the role and position of quantitative signal detection and the way it is
applied at Lareb is described. First, the results of the statistical analyses are used
as an aid in the traditional case-by-case review. Secondly, for datamining
purposes, a list of associations of ADRs and suspected drugs that are
disproportionately present in the database, is generated periodically.
Subsequently, these disproportionate associations needs to be analyse in more
detail. Finally, on an ad hoc basis, quantitative signal detection is used to study
signals in more detail, for example by adjusting for covariates or analysing more
complex relationships, like drug-drug interactions and drug-related syndromes.
174
Summary
In Chapter 4, the General discussion, some additional methodological

considerations and practical implications of quantitative signal detection are
elaborated. The studies presented in this thesis show that the quantitative signal
detection is an essential part of the analysis procedures applied by the Netherlands
Pharmacovigilance Foundation Lareb. It does not only facilitate the detection and
analysis of unknown associations between a single drug and a single event, but
also offers possibilities for the analysis of more complex relations such as drug-
drug interactions and drug-related syndromes. The studies presented in this thesis
show that the quantitative approach can be regarded as an additional source of
information and not as a substitute for the classical case-by-case analysis. SRSs
are primarily designed to generate hypotheses of possible ADRs, and they cannot
provide accurate information about the incidence of the ADR. Additionally, the
extent to which clinical information can be taken into account is limited in the
current method of quantitative signal detection. For this reason, every signal that
has been generated by the quantitative approach should be analysed in more
detail. Given the rapid advances in information technology, it is expected that
quantitative signal detection will establish itself as a standard source of
information in spontaneous reporting. Finally some suggestions are made for
further research.
175
Samenvatting
De ontwikkeling van een geneesmiddel is een langdurig proces waarin aandacht

besteed wordt aan kwaliteit, werkzaamheid en veiligheid. Op het moment waarop
een geneesmiddel op de markt toegelaten wordt, is de informatie omtrent de
veiligheid van geneesmiddelen in de regel beperkt en veelal afkomstig uit een
naar verhouding gering aantal onderzoeken en patiënten. Omdat deze
onderzoeken bovendien in de regel betrekking hebben op een geselecteerde groep
patiënten en ze beperkt zijn in zowel omvang als tijd, is het niet mogelijk een
volledig overzicht te hebben van alle bijwerkingen van het desbetreffende
geneesmiddel op het moment dat het op de markt toegelaten wordt. Met name de
kennis over zeldzamere bijwerkingen, bijwerkingen die pas na langere tijd
optreden en bijwerkingen bij patiënten die meerdere geneesmiddelen tegelijk
gebruiken of lijden aan chronische aandoeningen, is in dit stadium nog beperkt.
De dramatische ervaringen met thalidomide (Softenon) in de jaren zestig, een
slaapmiddel dat ernstige aangeboren afwijkingen kan veroorzaken bij kinderen
wiens moeder het middel in de eerste maanden van de zwangerschap gebruikt, liet
de noodzaak zien van een systematische bewaking van bijwerkingen ook nadat
deze op de markt toegelaten zijn. Een van de vormen van geneesmiddelen-
bewaking is de zogenaamde ‘spontane rapportage’. Artsen en apothekers kunnen
vermoedens van bijwerkingen op vrijwillige basis melden aan de hiervoor
aangewezen instanties. In Nederland wordt deze taak voor wat betreft
geregistreerde geneesmiddelen uitgevoerd door de stichting Landelijke Registratie
Evaluatie Bijwerkingen (Lareb) in opdracht van het College ter Beoordeling van
Geneesmiddelen. Een van de belangrijkste doelstellingen van spontane rapportage
is het zo snel mogelijk genereren van signalen die kunnen wijzen op een
mogelijke samenhang tussen een geneesmiddelgebruik en het optreden van de
gemelde bijwerking. Iedere melding van een mogelijke bijwerkingen wordt
hiertoe na ontvangst afzonderlijk geanalyseerd door ervaren beoordelaars. Naast
deze methode kan ook het gebruik van statistische technieken een efficiënt middel
zijn om de binnengekomen meldingen te analyseren in relatie tot eerder gedane
meldingen. De toepassing van statistische methoden in de geneesmiddelen-
177
Samenvatting
bewaking, ook wel kwantitatieve signaaldetectie genoemd, kan niet alleen

gebruikt worden voor de detectie en analyse van associaties tussen bijwerkingen
en geneesmiddelen, maar bovendien kunnen ook complexere relaties onderzocht
worden zoals geneesmiddelinteracties, geneesmiddel gerelateerde syndromen en
risicofactoren voor het ontwikkelen van bijwerkingen. Bij deze statistische
analyses wordt gekeken of meldingen vaker voorkomen dan op grond van het
toeval verwacht kan worden. Anders gezegd, hierbij wordt geprobeerd een
antwoord te geven op de vraag of de associatie disproportioneel vaak gemeld
wordt. Het doel van de in dit proefschrift beschreven onderzoeken is de waarde en
mogelijkheden van kwantitatieve signaaldetectie te exploreren en daarnaast meer
inzicht te krijgen in de analyse van complexe relaties zoals bijvoorbeeld
mogelijke geneesmiddelinteracties en de samenhang tussen verschillende gemelde
bijwerkingen als uiting van een geneesmiddel-gerelateerd syndroom.
In de eerste deel van dit proefschrift wordt een tweetal onderzoeken beschreven
die ingaan op enkele methodologische aspecten van kwantitatieve signaaldetectie
in geneesmiddelenbewaking. Verschillende centra voor geneesmiddelbewaking
hebben ieder een eigen methode van statistische analyse. In het in Hoofdstuk 1.1
beschreven onderzoek wordt nagegaan wat de mate is van overeenstemming van
de verschillende methoden, indien ze toegepast worden op de dataset van de
Stichting Lareb. Hiervoor worden de toepassing van de Reporting Odds Ratio
(ROR), de Proportional Reporting Ratio, Yules’ Q, de Chi-kwadraat toets en de
Poisson probability vergeleken met de toepassing van de Bayesian Propagation
Neural Network analysis (BCPNN) die door het WHO Collaborating Centre for
International Drug Monitoring gebruikt wordt. Bij vergelijking van de
verschillende methoden met de BCPNN blijkt dat de sensitiviteit hoog en de
specificiteit naar verhouding laag is. Bovendien laat dit onderzoek zien dat de
resultaten van de verschillende methoden grote overeenkomsten met elkaar
vertonen indien vier of meer meldingen per combinatie van geneesmiddel en
bijwerking ontvangen zijn.
Bij de onderzoeken in dit proefschrift wordt veelal gebruik gemaakt van de ROR
als maat voor het bestaan van disproportionaliteit. Een van de voordelen van het
gebruik van de ROR is dat bij de analyse van de relatie tussen geneesmiddel en
gerapporteerde bijwerking zogenaamde niet-selectieve onderrapportage geen
invloed heeft op de hoogte van deze ROR. Deze maat kan in een logistisch model
178
Samenvatting
ook gebruikt worden voor de analyse van clustering tussen geneesmiddelen bij
onderzoek naar mogelijke geneesmiddelinteracties en tussen gerapporteerde
bijwerkingen bij de analyse van mogelijke syndromen. In Hoofdstuk 1.2 wordt
aangetoond dat ook bij deze meer complexe analyses niet-selectieve
onderrapportage geen invloed heeft op de hoogte van de ROR.
In het tweede deel van dit proefschrift wordt een aantal praktische voorbeelden
gegeven van de toepassing van kwantitatieve signaaldetectie. Deze voorbeelden
hebben zowel betrekking op de analyse van een verband tussen geneesmiddel en
mogelijke bijwerking, als de analyse van meer complexere relaties zoals
mogelijke geneesmiddel-interacties en clustering van mogelijke bijwerkingen.
In Hoofdstuk 2.1 wordt een voorbeeld gegeven van de analyse van een mogelijke
relatie tussen een geneesmiddel en vermeende bijwerking. Het is bekend dat Niet-
steroïde anti-inflammatoire geneesmiddelen (NSAIDs) in zeldzame gevallen een
anafylactische reactie kunnen veroorzaken. In dit onderzoek wordt nagegaan of de
kans op het melden van een anafylactische reactie tijdens het gebruik van
NSAIDs hoger is dan bij het gebruik van andere geneesmiddelen en of er
onderlinge verschillen tussen de NSAIDs bestaan. De resultaten laten zien dat met
name voor diclofenac, ibuprofen en naproxen een naar verhouding groot aantal en
disproportioneel aantal meldingen ontvangen zijn. Dit zou kunnen wijzen op een
daadwerkelijk verhoogd risico op het krijgen van een anafylactische reactie bij het
gebruik van deze geneesmiddelen.
Het onderzoek uit Hoofdstuk 2.2 laat zien dat datasets van spontane
meldingsystemen gebruikt kunnen worden om mogelijke geneesmiddel-
interacties, die niet expliciet zijn gemeld, te detecteren. Uitgangspunt hierbij is dat
in het geval van een mogelijke geneesmiddel-interactie een of meer interacterende
geneesmiddelen meer bijwerkingen kunnen veroorzaken. Men kan verwachten dat
deze vervolgens ook meer gemeld worden. Als illustratie van deze methode is de
bijwerking ‘vertraagde onttrekkingsbloeding’ tijdens het gelijktijdig gebruik van
orale anticonceptiva en het antimycoticum itraconazol geanalyseerd. Met behulp
van een logistisch model is de invloed van het gecombineerd gebruik van beide
geneesmiddelen onderzocht. Het gelijktijdig gebruik van deze twee middelen
blijkt inderdaad geassocieerd te zijn met een verhoogd aantal meldingen van een
vertraagde onttrekkingsbloeding, hetgeen een aanwijzing kan zijn voor het
daadwerkelijk bestaan van geneesmiddel-interactie tussen deze stoffen. Een
179
Samenvatting
tweede voorbeeld wordt gegeven in Hoofdstuk 2.3 en laat zien dat ook een reeds
bekende geneesmiddel-interactie teruggevonden kan worden aan de hand van de
meldingen in de database. Het gelijktijdig gebruik van NSAIDs en diuretica kan
leiden tot een verminderd effectiviteit van laatstgenoemde middelen. Ook bij de
meldingen in de Lareb databank waar zowel NSAIDs als diuretica gebruikt
worden, komen meldingen betreffende een mogelijke verminderde effectiviteit
van diuretica vaker voor dan op grond van het toeval verwacht mag worden. De
onderzoeken uit beide hoofdstukken laten zien dat spontane rapportagesystemen
mogelijkheden bieden voor zowel de detectie als analyse van geneesmiddel-
interactie.
Niet alleen geneesmiddelen kunnen in samenhang met elkaar gemeld worden.
Ook de bijwerkingen zelf kunnen vaak in combinatie met elkaar gemeld worden,
hetgeen zou kunnen wijzen op het bestaan van een syndroom. Een voorbeeld
hiervan is nader uitgewerkt in het onderzoek dat in Hoofdstuk 2.4 besproken
word. Bij de Stichting Lareb zijn sinds de introductie van het antimycoticum
terbinafine acht meldingen ontvangen van het optreden van arthralgieën. Bij vier
meldingen werden eveneens huidreacties gemeld. Twee patiënten hadden
eveneens koorts. In het onderzoek worden de klinische details verder uitgewerkt
en wordt aan de hand van een logistisch model gekeken of de genoemde
symptomen met elkaar gerelateerd zijn. De statistische analyse versterkt de indruk
dat de genoemde symptomen inderdaad verband met elkaar lijken te houden.
Mogelijk ligt een immunologische reactie aan deze bijwerkingen ten grondslag.
In het derde deel wordt de plaats van de kwantitatieve benadering bij Lareb nader
beschreven. Doel van het in Hoofdstuk 3.1 beschreven onderzoek is inzicht te
krijgen in de factoren die een rol spelen bij de selectie en verspreiding van
signalen die voortkomen uit het meldingssysteem van Lareb. Van de vele signalen
die geïdentificeerd worden, zijn slechts enkele belangrijk genoeg om in een
speciale procedure doorgegeven te worden aan het College ter Beoordeling van
Geneesmiddelen (CBG). In een case-control studie werden signalen die aan het
CBG doorgegeven zijn, vergeleken met een controlegroep die bestond uit
associaties die niet in een speciale procedure aan het CBG doorgegeven werden.
Verschillende factoren werden onderzocht, waaronder het feit dat een
geneesmiddel nieuw was; of de bijwerking bekend was; de sterke van de
associatie en aspecten die betrekking hebben op de documentatie van de
180
Samenvatting
meldingen. De multivariate analyse toont aan dat de aanwezigheid van een

‘ernstige’ melding; het feit dat een bijwerking gecodeerd was met een
zogenaamde ‘critical term’ uit de bijwerkingenterminologie van de WHO; het feit
dat een bijwerking vaker gemeld is dan op grond van het toeval verwacht kon
worden, alsmede het feit of een bijwerking bekend was, allen onafhankelijk
geassocieerd waren met de selectie van een signaal. Daarnaast bleek dat
bijvoorbeeld de aanwezigheid van een combinatie van twee van de
bovengenoemde factoren een sensitiviteit van 0,72 en een specificiteit van 0,80
heeft voor het feit dat het signaal al dan niet geselecteerd wordt. Blijkbaar spelen
zowel kwantitatieve als kwalitatieve factoren een rol bij de selectie van signalen.
In Hoofdstuk 3.2 wordt een overzicht gegeven van de wijze waarop
kwantitatieve signaal detectie bij Lareb toegepast worden. Bij de afzonderlijke
evaluatie van de verschillende binnengekomen meldingen vervult de
kwantitatieve signaaldetectie een ondersteunende rol. Daarnaast is deze methode
ook een middel om actief te zoeken naar associaties in de databank die nadere
aandacht verdienen. Voor dit doel worden lijsten van geneesmiddel en mogelijke
bijwerking gegenereerd. Hierop staat mate van disproportionaliteit van
verschillende associaties vermeld. Een associatie die vaker voorkomt dan op
grond van het toeval verwacht wordt, kan wijzen op een mogelijk signaal en dient
daarom nader geanalyseerd te worden. Tot slot wordt de kwantitatieve benadering
bij Lareb gebruikt om meer complexe relaties te analyseren zoals geneesmiddel-
interacties en geneesmiddel-gerelateerde syndromen.
In de Discussie worden enkele aanvullende methodologische overwegingen

gegeven. De onderzoeken in dit proefschrift laten zien dat kwantitatieve
signaaldetectie een essentieel onderdeel vormt van het analyseproces bij Lareb.
Niet alleen worden hierdoor de detectie en analyse van onbekende associaties
tussen een geneesmiddel en mogelijke bijwerking verbeterd, maar het biedt
mogelijkheden ook meer om complexe relaties zoals geneesmiddel-interacties en
geneesmiddel-gerelateerde syndromen te onderzoeken. Spontane rapportage
systemen zijn primair ontworpen om hypothesen te genereren over het bestaan
van mogelijke bijwerkingen. Ze kunnen echter geen precieze informatie geven
over de incidentie van een bijwerking. Bovendien kan klinisch inhoudelijke
informatie op dit moment nog niet in het analyseproces betrokken worden. De
kwantitatieve benadering moet in de eerste plaats gezien worden als een
181
Samenvatting
additionele benadering en dient ieder gegenereerd signaal nader inhoudelijk

geanalyseerd te worden. Gezien de snelle ontwikkelingen in de
informatietechnologie mag verwacht worden dat kwantitatieve signaaldetectie een
standaardbron van informatie in de geneesmiddelenbewaking zal worden. Tot slot
worden voorstellen gedaan voor verder onderzoek.
182
Addendum to Chapter 1.2
Integrating underreporting into loglinear models
In this appendix we present our results in a more general way, set out for the
situation of two variables and an ADR. These two variables can be either
two specific drugs as in section 4, or one covariate and one drug in section 5,
or possibly two covariates.We denote the variables as follows: for the
specific ADR we use A, with levels a (a =1,2), and the other two variables
are X with levels x (x = 1,2) and Y with levels y (y =1,2). As usual, we
denote the true frequencies in the ADR- suspected population by txya, and the
expected frequencies in the SRS population by mxya.
We begin by defining a general loglinear model for the true frequencies
t
m xya in the
ADR experiencing population as
log t xya = u + u xX + uYy + u aA + u xy
XY XA
+ u xa + uYA XYA
ya + u xya (14)
where the parameters add up to zero over each index. By taking exponents
this equation
can be reparametrized as
X Y A XY XA YA XYA
t xya = bb x b y b a b xy b xa b ya b xya , (15)
X
where, for example, u = b , exp u xX = b x and so on. The identifying
restrictions in (15) are
X Y A XY XY XA XA
Õ x b x = Õ y b y = Õ a b a = Õ x b xy =Õ y b xy = Õ x b xa = Õ a b xa =
YA YA XYA XYA XYA
Õ y b ya = Õ a b a = Õ x b xya = Õ y b xya = Õ a b xya = 1.
We now write a general equation for the overall underreporting of the ADR
experiencing population by the term kxya. We use, for example, the notation
X
c x for the underreporting of levels x = 1 and x = 2 of variable X. This leads
to
X Y A XY XA YA XYA
k xya = cc x c y c a c xy c xa c ya c xya , (16)
183
with restrictions similar to those for the b-parameters:

Õ x c xX = Õ y cYy = Õ a c aA = Õ x c xyXY =Õ y c xyXY = Õ x c xa
XA XA
= Õ a c xa =
Õ y cYA YA XYA XYA XYA
ya = Õ a c a = Õ x c xya = Õ y c xya = Õ a c xya = 1.
In sections 3, 4 and 5, these restrictions are satisfied for the underreporting

constants because, for example, in Tables 5, and 6 c a ´ c -a 1 = 1.
The expected frequencies mxya for the SRS are related to the true frequencies
txya in the ADR experiencing population by
mxya = k xya t xya
( )(
= (cb ) c xX b x cYy b y caA b a c xy
X YXY
)(
b xy c xa
XA A
)(
b xa cYA
XY
)(
ya b ya c xya b xya
XYA XA
)( YA
)( XYA
)
X Y A XY XA YA XYA
= gg x g y g a g xy g xa g ya g xya .
(17)
where the g-parameters are restricted by
X Y A XY XY XA XA
Õ x g x = Õ y g y = Õ a g a = Õ x g xy = Õ y g xy = Õ x g xa = Õa g xa =
YA YA XYA XYA XYA
Õ y g ya = Õa g a = Õ x g xya = Õ y g xya = Õa g xya = 1.
Note that the g-parameters specify a multiplicative (i.e. loglinear) model for
the expected values in the SRS population. Analysis of the observed
frequencies of the SRS will yield estimates of the g-parameters. But equation
(17) illustrates the fact that, in principle, every g-parameter is biased w.r.t. its
corresponding β-parameter by its corresponding underreporting constant c.
XYA XYA XYA
For example, g xya is biased w.r.t. b xya with a factor c xya . In section 4.1 we
showed how this biases the odds-ratios in equations (6) to (11).
Equation (17) also makes clear what happens when certain underreporting
constants or certain loglinear β-parameters vanish by being equal to one.
First we distinguish three situations. In passing we indicate how these
situations relate to the special cases that we discussed in section 4. In our
discussion we only deal with so-called hierarchical models, i.e. if a higher-
order term is included, all the lower-order terms with indices that are a
subset of the set of indices of the higher-order term are included.
In the first situation, some of the higher-order underreporting constants
vanish where the corresponding β-parameters do not. The result is that, in
the expected frequencies in the SRS the higher-order γ-parameters are equal
184
to the β-parameters in the ADR suspected population. In this situation, in

estimating a loglinear model on the observed frequencies from the SRS, the
higher-order γ -parameters are unbiased w.r.t. to the corresponding β -
parameters of interest. This is discussed in detail in section 4.2, where
XYA
interest centred on the situation where c xya = 1 (there denoted by cdea), but
XYA
b xya ¹ 1 (there denoted by udea. This situation is worked out in observations
1 to 3 of section 4.2).
In the second situation the corresponding β-parameters and underreporting
constants vanish. Thus the corresponding γ -parameters vanish. In section
XYA
3.2, observation 4, this was discussed in detail for the vanishing of b xya and
XYA
c xya (there denoted as udea and cdea) (they vanish by assumption), and this
XYA
leads to the vanishing of g xya (they vanish by analysis). We refer to section
4.2 for a further discussion of this point, and the possible interpretation of
odds-ratios estimated from SRS data. In the third situation, some of the β-
parameters are equal to one, but none of the underreporting constants vanish.
The result is that, if a loglinear analyses is conducted on the observed
frequencies in the SRS, the estimates of γ -parameters are in fact estimates of
the corresponding underreporting constants. For example, if in the ADR
XA XYA
experiencing population in the true frequencies b xa = b xya = 1 , then
XA XYA XYA
g xa = g xya = 1 and c xya = 1. This situation is not discussed in section 4. If the
higher-order underreporting constants dominate the corresponding higher-
order β-parameters, SRS data are not useful for the assessment of causal
links between drugs and ADRs.
185
Reports of arthralgia during the use of terfinafine
Patient A is a 44-year-old woman who developed arthralgia of the right wrist

18 days after the start of terbinafine, one week this extended to her right
knee. Concomitant medication used was doxycycline 100 mg (indication for
use unknown) 7-14 days after the start of terbinafine. An oral contraceptive
ethinyl estradiol/desogestrel was used concomitantly. The complaints
hampered her daily activities. Laboratory examination revealed no
abnormalities. C-reactive protein levels, ESR, streptococcal antibody titre
and rheumatoid factor were within normal limits. The use of terbinafine was
stopped and she was treated with ibuprofen 400 mg three times daily. She
fully recovered. There was no history of joint disorders, traumata, liver of
kidney function disorders, rheumatological of allergic disorders. There were
no joint disorders in the family history. No rechallenge was carried out.
Terbinafine had not been used previously.
Patient B is a 41-year-old male who developed arthralgia of shoulders, hips
and elbows, 10 days after the start of terbinafine. All joints were
symmetrical affected. There were no additional signs of arthritis. There was
no fever, lymfadenopathy or skin disorders. Laboratory examination
revealed: ESR, 5 mm; GGT, 11 mmol/l; Hb, 8.9 mmol/l; leukocyte count
5.4x109/l. The use of terbinafine was stopped and he was treated with
nabumetone 1 g twice daily. He fully recovered. Concomitant medication
used was miconazole cream, beclomethasone nasal inhaler and ibuprofen
400 mg three times daily when needed. The latter drug was used because of
chronic backache.
Patient C is a 42-year-old female, who developed pains of hands and feet 1-2
days after the start of terbinafine. She used no concomitant medication.
Unfortunately, no additional information was received.
Patient D is a 41-year-old male who developed pain on flexion an extension
of wrists and knees, 2 weeks after the start of terbinafine. There were no
187
additional signs of arthritis. No additional laboratory testing was done. He

fully recovered immediately after stopping the use of terbinafine.
Patient E is 37-year-old female, who developed signs of ‘generalized’
arthralgia (no joints specified), 10 days after the start of terbinafine. On
physical examination no abnormalities were found. She was treated with
ibuprofen 400 mg. On laboratory examination, leukocyte count was
7.2x109/l; ALT and GGT had not increased. In the blood smear, some
atypical lymphocytes were found. After one week re-examination of blood
revealed no abnormalities. She fully recovered after 1 month.
Patient F is 66-year-old male who developed urticarial rash, fever, headache,
vomiting and diffuse arthralgia (no joints specified) 2 weeks after starting
terbinafine. He used a homeopathic drug as concomitant medication. He was
treated with terfenadine and prednisolone. Laboratory examination revealed
no abnormalities (Hb, ESR, leukocytes, and blood smear). After stopping
terbinafine, he fully recovered.
Patient G is a 34-year-old male, who developed signs of arthritis (no joints
specified) and generalized urticaria 10 days after he started using terbinafine.
Concomitant medication used was miconazole cream. He was treated with
terfenadine, prednisone and promethazine. Unfortunately, additional
information could not be retrieved, since the patient had moved and his
former GP did not have his present medical history.
Patient H is a 50-year-old male, who developed pains in almost all joints and
myalgia a 1-2 days after the start of terbinafine. The arthralgia was
symmetrically distributed. There were no signs of arthritis. One week after
the start of terbinafine, he developed anorexia, vomiting, and abdominal pain
and complained of smell and taste disorders. Also a rash, stomatitis and
peeling of the skin developed. He had a fever of 39oC. Laboratory tests
revealed the following abnormalities: leukocyte count 4.8 x109/l. ESR, 5
mm; Hb, 9.3 mmol/l; ALT, 1672 U/l; LDH, 869 U/l; GGT, 202 U/l; kidney
function normal. Concomitant medication used was budesonide inhaler. He
was treated with diclofenac and domperidone and fully recovered after 9
weeks.
All patients were treated with terbinafine 250 mg once daily. Patients A-E
and H were treated for onychomycosis. The indication for use of patient G
was not known. Patients A, B, D-F and H had no history of joint disorders,
188
traumata, liver or kidney function disorders, rheumatological or allergic

disorders. Neither were there any joint disorders in the family history in any
of the patients. These data could not be retrieved for patients C and G. No
rechallenge was done in any of the patients.
189
Dankwoord
Het is een goede traditie om op de laatste pagina’s van een proefschrift diegenen
te bedanken die direct of indirect bijgedragen hebben aan de totstandkoming
ervan. Ik wil daarom beginnen mijn dank uit te spreken aan mijn beide
promotoren Bert Leufkens en Toine Egberts. Bert verstaat de kunst om gericht en
kritisch naar je werk te kijken. Daarbij weet hij ook altijd openingen te creëren
voor oplossingen van mogelijke problemen. Zijn stimulerende en enthousiaste
houding combineert hij met een grote vakinhoudelijke kennis. Bovendien bleek
hij steeds in staat de soms wat engere visie die ik had wanneer ik me al te zeer
met klein detail bezig hield, in te ruilen voor een breder perspectief. Toine heb ik
reeds leren kennen in 1994 toen wij beiden op dezelfde dag bij de Stichting Lareb
kwamen werken. Zijn betrokkenheid, didactische vaardigheden en een
uitstekende vakinhoudelijke kennis zijn een ideale combinatie van eigenschappen
die ertoe bijgedragen hebben dat dit proefschrift ook inderdaad tot stand kwam.
Hij mag daarom met recht de drijvende kracht achter dit project genoemd worden.
Kees van Grootheest heeft als directeur van de Stichting Lareb de mogelijkheden
gecreëerd om binnen het kader van mijn werkzaamheden dit onderzoek uit te
kunnen voeren. Zijn politieke visie heeft ervoor gezorgd dat het werk van Lareb
in een nog bredere kring erkenning heeft gekregen en zodoende een solide basis
vormt van waaruit ideeën verder ontwikkeld kunnen worden. Hij hielp ervoor te
zorgen dat de onderzoeken in het proefschrift nauw aansluiten bij de
doelstellingen van de Stichting Lareb.
Collega’s en oud-collega’s bij de Stichting Lareb, AIO’s en andere medewerkers
van de vakgroep farmaco-epidemiologie en farmacotherapie in Utrecht hebben
laten zien dat een goede ontspannen werksfeer ervoor zorgt dat je met veel plezier
kunt werken en optimaal kunt presteren.
Verschillende co-auteurs hebben een belangrijke inhoudelijke bijdrage geleverd
aan dit proefschrift. Jacques van der Hofstede, Peter van der Heijden en Stef van
Buuren van de afdeling Preventie en Gezondheid van TNO waren in staat vanuit
een andere invalshoek hun licht te laten schijnen op de kwantitatieve
signaaldetectie. Ron Meyboom heeft met zijn grote kennis van de genees-
191
Dankwoord
middelenbewaking en beschouwende houding een essentiële bijdrage geleverd

aan de kwaliteit van het proefschrift. De levendige discussies met Andrew Bate
van het Uppsala Monitoring Centre hebben laten zien dat onderlinge
samenwerking tussen verschillende centra die zich bezig houden met spontane
rapportage van grote toegevoegde waarde is.
De leescommissie waarin zitting hadden: Prof. dr. AW Broekmans, Prof. dr. IR
Edwards, Prof. dr. YA Hekster, dr. RHB Meyboom en Prof. dr. BHC Stricker
waren in staat het proefschrift binnen korte termijn te beoordelen en bovendien
waardevolle aanwijzingen te geven om het geheel verder te verbeteren.
Zonder onze melders geen meldingen aan de Stichting Lareb en dus ook geen
proefschrift. Artsen en apothekers die de moeite nemen om hun vermoedens van
bijwerkingen te melden doen dit veelal in een relatieve anonimiteit. Uit eigen
ervaring weet ik dat het niet altijd vanzelfsprekend is een meldformulier in te
vullen en te verzenden. Enthousiasme en geloof in het uiteindelijke doel van het
melden van bijwerkingen zijn nodig om dit in de dagelijkse gang van zaken in te
passen.
Mijn werk als huisarts is voor mij van grote waarde geweest de realiteit van alle
dag niet uit het oog te verliezen. De prettige werksfeer waarin ik samen met mijn
collega’s in Vught kan werken hebben me steeds laten zien wat het uiteindelijke
doel is van onze exercities in de geneesmiddelenbewaking: een verbetering van de
zorg voor de individuele patiënt.
Dank ook aan al diegenen die in de laatste fase van het schrijven van het
proefschrift een helpende hand geboden hebben met het corrigeren van foutjes,
het opmaken van de tekst en het verzenden van de proefschriften. Mijn
paranimfen Marcel Mutsaerts en Willem Diemont wil ik danken voor hun
ondersteuning en hulp bij de organisatie van het feest.
Mijn grootste dank gaat uit naar Emmy en onze dochtertjes Linde en Sterre. De
laatste maanden is er weinig tijd voor elkaar geweest. Hun steun en liefde hebben
het mogelijk gemaakt dat de ruimte om dit proefschrift te schrijven er toch was.
192
List of publications
Articles marked with an asterix (*) relate to work described in this thesis.
Crohbach MJJS, Toorenburg-Beijer B, van der Wal J, van Leeuwen JT, van Puijenbroek
EP, van der Laan JR, Geijer RMM. NHG-standaard allergische en hyperreactieve
rhinitis. Huisarts en Wetenschap 1995;38:216-17.
Van Puijenbroek EP, Meyboom RHB. Slaapstoornissen tijdens het gebruik van
mefloquine. Pharmaceutisch weekbl 1995;130:477-9.
Van Puijenbroek EP. Stemklachten als bijwerking van geneesmiddelen. Pharmaceutisch
Weekbl 1996;131:551-4.
Van Puijenbroek EP, Van Amerongen CA. Is het eerste-uitgifte signaal van nut voor
postmarketing surveillance? Resultaten van een pilot study. Pharmaceutisch
Weekbl 1996;16:459-62.
Van Puijenbroek EP, Du Buf-Vereijken PW, Spooren PF, van Doormaal JJ. Possible
increased risk of rhabdomyolysis during concomitant use of simvastatin and
gemfibrozil. J Intern Med 1996;240:403-4.
Van Puijenbroek EP, Egberts ACG, Krom HJ. Visual hallucinations and amnesia
associated with the use of zolpidem. Int J Clin Pharmacol Ther 1996;34:318.
Meyboom RHB, Verduijn MM, Steenvoorden MG, Dekens-Konter JA, van Puijenbroek
EP. Reversibele tandverkleuring tijdens oraal gebruik van antibiotica. Ned Tijdschr
Geneeskd 1996;140:207-9.
Van Puijenbroek EP, Egberts ACG, Trooster JF, Zomerdijk J. Reduction of migrainous
headaches during the use of acenocoumarol. Headache 1996;36:48.
Meyboom RHB, van Puijenbroek EP, Vinks MHAM, Lastdrager CJ. Disturbance of
withdrawal bleeding during concomitant use of itraconazole and oral
contraceptives. N Z Med J 1997;110:300.
Van Puijenbroek EP, van Grootheest AC. Bijwerkingen, een verantwoordelijkheid van
artsen? Medisch Contact 1997;52:1156-8.
In 't Veld BA, van Puijenbroek EP, Stricker BHCh. Overgevoeligheidsracties bij gebruik
van mebeverine. Ned Tijdschr Geneeskd 1997;141:1392-5.
Bouvy M, Gerts BMF, Smakman-Nossbaum E, van Puijenbroek EP. Fixed drug eruption
bij gebruik van norfloxacine. Huisarts en Wetenschap 1997;40:595-7.
193
List of Publications
Van Puijenbroek EP, Feenstra J, Meyboom RHB. Verstoring van de pilcyclus tijdens het
gelijktijdig gebruik van itraconazol en orale anticonceptiva. Ned Tijdschr Geneeskd
1998;142(3):146-9.
Van Puijenbroek EP, Metselaar HJ, Berghuis PH, Zondervan PE, Stricker BHC. Acute
levercelnecrose bij gebruik van ketoconazol. Ned Tijdschr Geneeskd
1998;142:2416-8.
Heeringa M, van Puijenbroek EP. Reversible dysgeusia attributed to losartan. Ann Intern
Med 1998;129:72.
In 't Veld BA, Kwee-Zuiderwijk WJM, van Puijenbroek EP, Stricker BHC.
Neuropsychiatrische bijwerkingen toegeschreven aan het gebruik van oxybutynine.
Nederl Tijdschr Geneeskd 1998;142:590-2.
Van Puijenbroek EP, Meyboom RHB. Semen-like urethral discharge during the use of
mazindol. Int J Eat Disord 1998;24:111-3.
Heeringa M, Pulles-Heintzberger CFM, van Puijenbroek EP, Verduijn MM. Verlenging
van de QTc-tijd bij een pasgeborene tijdens gebruik van cisapride. Ned Tijdschr
Geneeskd 1999;143:2018-111.
*Van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens HGM. Signalling
possible drug-drug interactions in a spontaneous reporting system: delay of
withdrawal bleeding during concomitant use of oral contraceptives and
itraconazole. Br J Clin Pharmacol 1999;47:689-93.
Eland IA, van Puijenbroek EP, Sturkenboom MJ, Wilson JH, Stricker BHC. Drug-
associated acute pancreatitis: twenty-one years of spontaneous reporting in The
Netherlands. Am J Gastroenterol 1999;94:2417-22.
Diemont WL, van Puijenbroek EP, Hekster YA. Toeval of toeval? Epilepsie en
bijwerkingen. Ned Tijdschr voor Epileptologie 2000;28:134-44.
Grootheest AC van, van Puijenbroek EP. Artsen en apothekers kennen Lareb-is dat zo?
Pharmaceutisch Weekbl 2000;135:489-91.
*Van Puijenbroek EP, Egberts ACG, Heerdink ER, Leufkens HG. Detecting drug-drug
interactions using a database for spontaneous adverse drug reactions: an example
with diuretics and non-steroidal anti-inflammatory drugs. Eur J Clin Pharmacol
2000;56:733-8.
Pijlman AH, Kuck EM, van Puijenbroek EP, Hoekstra JB. Acuut delier, waarschijnlijk
uitgelokt door claritromycine. Ned Tijdschr Geneeskd 2001;145:225-8.
Van der Linden PD, van Puijenbroek EP, Feenstra J, Veld BA, Sturkenboom MC, Herings
RM, Leufkens HG, Stricker BH. Tendon disorders attributed to fluoroquinolones: a
study on 42 spontaneous reports in the period 1988 to 1998. Arthritis Rheum
2001;45:235-9.
*Van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens HGM. Different risks for
NSAID-induced anaphylaxis. Ann Pharmacother: accepted for publication.
194
List of publications
*Van Puijenbroek EP, Egberts ACG, Meyboom RHB, Leufkens. Association between
terbinafine and arthralgia, fever and urticaria: symptoms or syndrome?
*Van Puijenbroek EP, van Grootheest AC, Diemont W, Egberts ACG, Meyboom RHB,
Leufkens HGM. Determinants of signal selection in a spontaneous reporting system
for adverse drug reactions Br J Clin Pharmacol: accepted for publication.
195
Curriculum vitae
Eugène van Puijenbroek was born in Tilburg, the Netherlands on September 28,
1960. Following the completion of pre-university education at the
Rijksscholengemeenschap Koning Willem II in Tilburg (HAVO and Atheneum
B) in 1979, he started his medical study at the University of Nijmegen in the same
year. He obtained his medical degree in 1987 after which he fulfilled his military
service as a physician in the Dutch army. From 1989 till 1991 he worked as a
medical adviser at the ‘Sociaal Fonds Bouwnijverheid’. He was trained as general
practitioner at the University of Maastricht from 1991 till 1993. In 1994 he started
his work as a research associate and subsequently as head of the analysing
department of the Netherlands Pharmacovigilance Foundation Lareb.
Additionally he works as a general practitioner in the township of Vught. In
November 1998 he started the research project described in this thesis in
collaboration with the Department of Pharmacotherapy and
Pharmacoepidemiology, Utrecht Institute for Pharmaceutical Sciences (Head:
Prof. dr. HGM Leufkens) at the University of Utrecht. He is married with Emmy
Konst. They have two daughters: Linde and Sterre.
197

Book of Pharmacovigilance

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Book of Pharmacovigilance

Hochgeladen von

Copyright:

Verfügbare Formate

Quantitative Signal Detection

Eugène P. van Puijenbroek

Title: Quantitative Signal Detection in Pharmacovigilance

Printed by: Print Partners Ipskamp, Enschede

Kwantitatieve Signaaldetectie in de Geneesmiddelenbewaking

(met een samenvatting in het Nederlands)

Ter verkrijging van de graad van doctor aan de Universiteit Utrecht

Eugène Paul van Puijenbroek

Prof. dr. HGM Leufkens

Prof. dr. ACG Egberts

The work presented in this thesis was performed at the Netherlands

To Emmy, Linde and Sterre

PART 1 METHODOLOGICAL ASPECTS OF QUANTITATIVE SIGNAL DETECTION 21

1.1 A comparison of measures of disproportionality for signal detection

1.2 On the assessment of adverse drug reactions from spontaneous

PART 2 PRACTICAL APPROACHES OF QUANTITATIVE SIGNAL DETECTION 65

2.1 Different risks for NSAID-induced anaphylaxis 67

2.2 Signalling possible drug-drug interactions in a spontaneous reporting

2.3 Detecting drug-drug interactions using a database for spontaneous

2.4 Association between terbinafine and arthralgia, fever and urticaria:

3.1 Determinants of signal selection in a spontaneous reporting

3.2 Practical application of quantitative signal detection in the

General discussion 155

Addendum to Chapter 1.2 183

Addendum to Chapter 2.4 187

List of publications 193

Curriculum vitae 197

The need for pharmacovigilance

These centres usually rely on the method of ‘spontaneous reporting’ to monitor

Meyboom defined a signal as a set of data constituting a hypothesis that is

Methodology of spontaneous reporting

Quantitative signal detection

Methods applied in analytical epidemiology, such as the case control design,

(concomitant) medication and the suspected ADR(s). Among these covariates

Signal Signal Signal

Hypotheses may be formulated in the literature based on observations in medical

Objectives and outline of this thesis

Before 1960, virtually no systematic surveillance of drugs for unexpected adverse

Patient data Association between drug and ADR

Figure 2. The relationship between patient, ADR and drug

dysplasia in 3,980 diethylstilbestrol-exposed young women. Experience of the National

METHODOLOGICAL ASPECTS OF QUANTITATIVE SIGNAL

A COMPARISON OF MEASURES OF DISPROPORTIONALITY

EP van Puijenbroek1, A Bate2,3, HGM Leufkens4, M Lindquist2, R Orre5 and

SUBMITTED FOR PUBLICATION

Objective A continuous systematic review of all reported combinations of drugs

reported suspected ADR and the suspected medication, can be calculated by

Reports with the event Reports with other events

Reports with the suspected drug a b

ROR-1.96se>1 a³2 1.00 0.66 0.41 1.00 94.8

PRR-1.96se>1 a³2 1.00 0.70 0.44 1.00 94.8

Yule’s Q - a³2 1.00 0.60 0.38 1.00 99.5

Chi square a³2 1.00 0.71 0.46 1.00 100.0

Poisson a³2 1.00 0.53 0.34 1.00 100.0

The prior assumption for the Bayesian derivation of the IC is of independence

Choice of a measure of disproportionality

Table 2. Conditions, advantages and disadvantages of different measures of

ROR-1.96se 1 Cells a,b,c and -Easy applicable -Odds ratio and

PRR-1.96se 1 Cells a and c -Easy interpretation Standard error cannot

Chi square Always applicable Difficult to interpret

Yule’s Q-1.96se 0 cells a,b,c and -Standard error

Poisson Only for rare Correction for Only p-value

IC-2sd 0 none -Always applicable Relatively non

The ROR is a transparent measure, easily interpretable which can be easily