Resuscitation (2008) 78, 265274

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/resuscitation

CLINICAL PAPER

Shock outcome prediction before and after CPR: A

comparative study of manual and automated active
compressiondecompression CPR
M.S. Box a,, J.N. Watson b, P.S. Addison b, G.R. Clegg c, C.E. Robertson c

a
South Western Ambulance Service NHS Trust, 42 Porchester Road, Bournemouth BH8 8LE, UK
b
CardioDigital Ltd., Elvingston Science Centre, Edinburgh, Scotland, UK
c
Department of Accident and Emergency Medicine, The Royal Inrmary of Edinburgh, Scotland, UK

Received 18 September 2007; received in revised form 2 March 2008; accepted 14 March 2008

KEYWORDS Summary We report on a study designed to compare the relative efcacy of man-
Electrocardiography; ual CPR (M-CPR) and automated mechanical CPR (ACD-CPR) provided by an active
Cardiopulmonary compressiondecompression (ACD) device. The ECG signals of out-of-hospital cardiac arrest
resuscitation (CPR); patients of cardiac aetiology were analysed just prior to, and immediately after, cardiopul-
Active compression/ monary resuscitation (CPR) to assess the likelihood of successful debrillation at these time
decompression-CPR; points. The cardioversion outcome prediction (COP) measure previously developed by our group
Cardiac arrest; was used to quantify the probability of return of spontaneous circulation (ROSC) after counter-
Shock outcome shock and was used as a measure of the efcacy of CPR. An initial validation study using COP to
prediction predict shock outcome from the patient data set resulted in a performance of 60% specicity
achieved at 100% sensitivity on a blind test of the data. This is comparable with previous studies
and provided condence in the robustness of the technique across hardware platforms. Signif-
icantly, the COP marker also displayed an ability to stratify according to outcomes: asystole,
ventricular brillation (VF), pulseless electrical activity (PEA), normal sinus rhythm (NSR). We
then used the validated COP marker to analyse the ECG data record just prior to and immedi-
ately after the chest compression segments. This was initially performed for 87 CPR segments
where VF was both the pre- and post-CPR waveform. An increase in the mean COP values was
found for both CPR types. A signed rank sum test found the increase due to manual CPR not
to be signicant (p > 0.05) whereas the automated CPR was found to be signicant (p < 0.05).
This increase was larger for the automated CPR (1.26, p = 0.024) than for the manual CPR
(0.99, p = 0.124). These results indicate that the application of CPR does indeed provide ben-
ecial preparation of the heart prior to debrillation therapy whether manual or automated
CPR is applied. The COP marker shows promise as a denitive, quantitative determinant of the

A Spanish translated version of the summary of this article appears as Appendix in the nal online version at

doi:10.1016/j.resuscitation.2008.03.225.
Corresponding author. Tel.: +44 1202789317.

E-mail address: martyn.box@ntlworld.com (M.S. Box).

0300-9572/$ see front matter 2008 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.resuscitation.2008.03.225
266 M.S. Box et al.

immediate positive effect of both types of CPR regardless of the details of use. In work of a
more exploratory nature we then used the validated COP marker to analyse the ECG pre- and
post-CPR for all rhythm types (212 traces). We show a signicant increase in the COP measure
(p < 0.001 in both cases) as indicated by a shift in the median COP marker distribution values.
This increase was more pronounced for automated ACD-CPR than for manual CPR. However, a
detailed statistical analysis carried out between the groups adjusted for pre-CPR value showed no
signicant difference between the two methods of CPR (p = 0.20). Similarly, adjusting for length
of CPR showed no signicant difference between the groups. Secondary, subgroup analysis of the
ECG according to the length of time for which CPR was performed showed that both types of CPR
led to an increase in the likelihood of successful debrillation after increasing durations of CPR,
however results were less reliable after longer periods of continuous CPR.
2008 Elsevier Ireland Ltd. All rights reserved.

Introduction
Experimental13 and clinical4 studies have indicated that
administering cardiopulmonary resuscitation (CPR) prior to
shock therapy can increase the likelihood of successful deb-
rillation for prolonged ventricular brillation (VF). These
results are consistent with clinical studies5,6 which indi-
cate that pre-shock CPR can improve the rates of return
of spontaneous circulation (ROSC) and survival to hospital
discharge when emergency medical services (EMS) response
times exceeded 45 min. There is recognition that, while
debrillation is the only effective means of reverting the
heart in cardiac arrest to normal sinus rhythm (NSR), max-
imising the quantity and quality of CPR has a substantial
bearing on the efcacy of debrillation therapy.
CPR performed effectively improves cardiac arrest sur-
vival. However, it is known that manual chest compressions
do not always achieve the recommended performance lev-
els in terms of rate, depth and hands off time.7,8 In
addition, it has been shown that trained paramedics can pro-
vide shallower and slower compressions over time without
noticing.9,10 To this end a number of automated mechan-
ical devices have been developed to provide consistent,
good quality CPR to the patient throughout the resuscitation
procedure.11 These devices all provide active compression
of the chest at a set rate and compression depth. The
LUCAS devices used in the study reported here also provide
active decompression of the chest.12 Such active compres-
sion/decompression CPR (ACD-CPR) devices are designed to
enhance the decrease in intrathoracic pressure during the
decompression phase thus improving venous blood return to
the heart.
Two recent studies of another mechanical CPR device
have produced conicting results. Both studies involved the
use of a load distributed band (LDB) CPR device. These
generate active compressions through a band placed round
the chest. However, these machines do not produce active
decompression. In an 1838 patient randomised multicentre
trial of patients experiencing out-of-hospital (OOH) cardiac
arrest, Hallstrom et al.13 found that a LDB-CPR device was
associated with worse neurological outcomes and a trend
towards worse survival than manual CPR (M-CPR). Survival
to hospital discharge was 9.9% for the manual CPR group
and 5.8% for the LDB-CPR group (p = 0.06), and cerebral
performance category of 1 or 2 at hospital discharge was
recorded at 7.5% of patients who underwent manual CPR
compared with 3.1% of the LDB-CPR group. These ndings
led to the early termination of the trial at the rst planned
interim monitoring point. Another recent study by Ong et
al.14 showed signicant improvement in outcomes for LDB-
CPR when compared with manual CPR: ROSC (34.5% for
LDB-CPR versus 20.2% for manual CPR), survival to hospi-
tal admission (30.9% for LDB-CPR R versus 11.1% for manual
CPR) and survival to hospital discharge (9.7% LDB-CPR for
versus 2.9% for manual CPR). However, they found no sig-
nicant difference in cerebral performance category and
overall performance category for these patients.
The recent deployment of ACD-CPR devices throughout
the South Western Ambulance NHS Trust provided an oppor-
tunity to compare the relative effectiveness of manual and
automated CPR on OOH cardiac arrest patients. ECG data
records downloaded from the Medtronic LIFEPAK 12 debril-
lators used throughout the Trust were used in this respect.
The current non-randomised observational clinical study was
designed to quantify changes in the ECG caused by episodes
of CPR compressions. To do this a shock outcome prediction
metric was computed from the ECG immediately before and
just after episodes of CPR compressions. The cardioversion
outcome prediction (COP) measure, previously developed
by our group, was employed for this task. The COP marker
pairs (pre- and post-CPR) were computed both for manual
and ACD-CPR and the results compared. The COP marker was
rst validated for use on this data set using the shock seg-
ments of ECG trace contained within the OOH data record.
The COP marker was then used to quantify the pre and post-
CPR ECG signals, for both manual and automated mechanical
ADC-CPR.
Evaluation of the cardioversion outcome
predictor (COP) marker for use in CPR study
Prior to the analysis of the CPR episodes described in the
next section, the COP algorithm was rst tested for its
efcacy as a useful predictor of shock outcome for this
particular database using the shock segments of the ECG
signals.
The study data
The study was conducted by the South Western Ambulance
Service NHS Trust. ECG data was acquired over an 8-month
period from November 2005 to June 2006 inclusive. The data
was downloaded from Medtronic LIFEPAK 12 debrillators
Shock outcome prediction before and after CPR
Table 1 Outcomes of the 141 shocks included in the
analysis
Rhythm Occurrence (N)
Asystole 23
Ventricular brillation 49
Pulseless electrical activity 59
Normal sinus rhythm 10
used in the study. ECG signals were recorded from a conve-
nience sample of 54 patients during out-of-hospital (OOH)
cardiac arrest of cardiac aetiology during the study period.
The study was approved by the Dorset Ethics Committee
(REC Number 06/Q2201/94). All patient data were stored
in accordance with the Trust policy with regard to the Data
Protection Act.
Of 169 shocks recorded through the Generic Paddles
(general use paddles as dened by the manufacturer) of the
debrillator, 28 were excluded. These exclusions consisted
of: 4 traces containing a shock of a non-VF rhythm; 8 traces
where less than 4 s of useful pre-shock signal was available
to analyse, and 16 traces each of which contained less than
12 s of post-shock trace required to accurately conrm the
outcome rhythm. The breakdown of post-shock rhythms of
the remaining 141 shocks included in the study is detailed
in Table 1.
The cardioversion outcome predictor (COP) marker
The main CPR study (described fully in CPR study) was con-
ducted using 4 s segments of ECG signal just prior to and
immediately after each CPR episode occurring in the patient
OOH records. Each pre- and post-CPR ECG analysis seg-
ment was quantied using the COP shock outcome prediction
measure previously developed by our group.15 Such shock
outcome prediction measures are derived by interrogating
the ECG to provide a metric predictive of the success of
debrillation based on the signal characteristics. This quan-
titative measure of the signal is indicative of the state of
the myocardium. Many shock outcome prediction measures
have been suggested including the COP marker developed by
our group which has shown itself to be consistently better
than other popular methods at correctly identifying shock
outcomes.16,17
The COP marker is based on wavelet transform signal pro-
cessing technology. Over recent years, wavelet transform
analysis has increasingly shown itself particular valuable for
analysing problematic signals across a wide variety of areas
in science, engineering and medicine.18 It has been partic-
ularly useful in analysing a variety of biosignals including,
most notably, the ECG.19 Because of its ability to eluci-
date simultaneously local spectral and temporal information
from a signal, the wavelet transform-based method is par-
ticularly useful for the analysis of transient, aperiodic and
other non-stationary signal features. It overcomes a signi-
cant limitation of the more widely used Fourier transform,
which contains only globally averaged information and so has
the potential to obscure transient or location specic fea-
tures within the signal. Another key advantage of wavelet
techniques is the variety of wavelet functions available thus
267
allowing the most appropriate to be chosen for the sig-
nal under investigation. This is in contrast to Fourier-based
analysis, which is restricted to one feature morphology: the
sinusoid.
We have previously reported the identication of
coherent structure in VF signals using our wavelet transform-
based methods.2022 Using such techniques we have
quantied timefrequency characteristics of the VF signal
and used them as reliable markers for shock outcome pre-
diction, using data collected during episodes of OOH cardiac
arrest.1517
The wavelet transform of a signal x(t) is dened as
  
+
1 tb
T (a, b) = x(t) dt (1)
a a
where *(t) is the complex conjugate of the wavelet func-
tion (t), a is the dilation parameter of the wavelet and b is
the location parameter of the wavelet. Temporal behaviour
of local signal features can be quantied from the wavelet
transform time-scale representation of the signal. One such
measure of this temporal behaviour can be derived from an
entropy-like intermittency measure computed over one or
more of the scalogram scales: the COP marker, dened as

ln |T (a , b)| db
WEa =  (2)
|T (a , b)| db

where |T(a , b)| are the wavelet transform modulus values
at a selected scale a . All modulus values across the selected
scale a are used in this analysis. Four seconds of ECG trace
immediately prior to the shock was used in the wavelet
transform analysis employing a complex Morlet wavelet with
characteristic frequency of 5.5 rad/s.23 Scale a used in the
analysis corresponds to a characteristic central frequency of
the wavelet function of 45 Hz.
Results of the COP validation
The COP metric was computed for each of the 141 pre-shock
traces in the data set, all of which exhibited VF. Figure 1
contains the box plots of the COP marker value distribu-
tions obtained for each post-shock rhythms (as classied in
Table 1). Interestingly, the groupings of COP values calcu-
lated on the pre-shock VF ECG exhibit a distinct monotonic
trend, from (arguably) the poorest outcome to the opti-
mal outcome: asystole, VF, PEA, and NSR. This provides
condence that the marker is capable of resolving a sub-
tle spectrum of information indicative of the state of the
myocardium.
Further grouping the outcome classications into those
corresponding to a return of spontaneous circulation
(ROSC = NSR) and those which do not (non-ROSC = asystole,
VF, PEA) we can test the efcacy of the marker using a lin-
ear decision threshold across the two distributions to obtain
that systems test sensitivity and specicity. Here, the sys-
tem sensitivity is dened as the proportion of patients that
are successfully debrillated (i.e. achieve ROSC) which are
correctly identied. The system specicity is therefore the
proportion of patients that do not respond to debrillation
(failures) that are correctly identied. Unless these suc-
cess and failure classes as completely differentiable there is
always a trade-off in the sensitivity and specicity achiev-
268
Figure 1 Box plots of the COP marker values computed for
each of the outcome classications: asystole, VF, PEA, and NSR.
(The line of 100% sensitivity between ROSC and non-ROSC is
shown dotted on the plot. Mean distribution values are asys-
tole = 9.05, VF = 7.51, PEA = 5.12, and NSR = 2.96.).
able by a system. This trade-off is described by plotting
sensitivity/specicity pairs in a receiver operator character-
istic (ROC) curve. ROC curves are used widely in the medical
literature to assess diagnostic test performance. These ROC
curves plot the specicity values against their associated
sensitivities as the decision boundary is moved in parameter
space. The area under the ROC curve summarizes diagnos-
tic performance. A unit area represents a perfect diagnostic
test whereas 0.5 represents a worthless test.
Figure 2 contains the ROC curve for the COP calculated
for the Dorset data set. The Dorset data set contains only
10 ROSC outcomes: too few to allow for a cross-validation
study of statistical signicance (Watson et al., 2006). (Hence
there are no error bars on the resulting ROC plot.) The COP
metric was therefore tested blind on the whole data set. The
computed ROC curve indicates a very similar performance
to that reported previously for other data sets (cf. Figure 6
Figure 2 ROC curve where sensitivity is a measure of the
successful classication of ROSC and specicity the successful
classication of non-ROSC.
M.S. Box et al.
of Ref. 16 and Figure 2(a) of Ref. 17), with an area of 0.86
and a specicity of 60% being achieved as the curve passes
between the 90% and 100% sensitivity point. This conrms
the validity of the COP marker for this data set and pro-
vides condence for its use in the main CPR study described
in the following section. As the COP marker was originally
developed using a different data set acquired on another
manufacturers debrillator,16 the results also highlight the
robustness of the COP measure itself across platforms.
CPR study
The study data
The ECG database described in the previous section was
used. Sections of CPR trace suitable for analysis were
identied and isolated by considering the ECG record in
conjunction with the corresponding impedance trace, also
downloaded from the debrillator devices. Of the iden-
tiable CPR segments a number were excluded from the
analysis due to following reasons: traces with less than 4 s
of trace pre- or post-CPR available for analysis (8 traces);
traces from patients with pacemakers tted (2 patients: 13
traces); traces starting <12 s from a previous shock mak-
ing clear identication of the rhythm difcult (27 traces);
traces with clear breathing ventilation artefact on the ECG
(1 trace). In total 212 trace segments containing CPR were
extracted for study: 114 corresponding to manual CPR and
98 to automated CPR. Of these 212 traces, 87 corresponded
to CPR segments where VF was both the pre- and post-CPR
rhythm (49 corresponding to manual CPR and 38 to auto-
mated CPR).
Figure 3 shows an example plot of a manual CPR episode
taken from the ECG database (patient 8 episode 5). The ECG
shown in the upper plot shows a VF waveform which contains
a CPR episode from 276 to 303 s. The lower plot contain-
ing the associated impedance trace shows quite clearly the
period where chest compressions were applied. This allowed
for ease of selection of the ECG segments for analysis. By
comparing the COP marker value for 4 s of hands-free trace
pre- and post-CPR (shown in the gure), the effectiveness of
the two types of CPR were quantied using the COP marker.
Results of the CPR analysis
VF only
All 87 traces for which VF was the pre- and post-CPR rhythm
were analysed. COP values were computed for the VF sig-
nal immediately before and immediately after CPR. The box
plots for the results are shown in Figure 4a and b. Compar-
ing the two gures we observe a distinct increase in the
mean COP values for both CPR types. However, quantify-
ing this distributional change in the pre- and post-CPR data
is problematic due to the non-parametric nature of the dis-
tributions themselves which exhibit signicant skew directly
attributable to the logarithmic non-linearity inherent within
the COP marker (as dened by equation (2)). Hence we
report results simply in terms of median values, m, quartile-
1 and quartile-3 values q1 , q3 , and signicance, p. Boxplots
of the distributions are also shown to provide a further visual
overview of data distributions. Employing the median as the
Shock outcome prediction before and after CPR
Figure 3 Example of an ECG and associated impedance trace (patient 8 episode 5: manual CPR). The 4 s regions selected for COP
analysis are indicated schematically on the plot.
statistical index of central tendency of the distributions pro-
vides robustness against outliers, while q1 and q3 provide
an indication of the distributional spread either side of the
median value. The signicance of the differences in the pre-
and post-CPR COP metrics are quantied by the Wilcoxon
signed rank sum test p values. This test is a non-parametric
Figure 4 Box plots of COP distributions according to CPR type
for VF (a) pre-CPR and (b) post-CPR.
269
alternative to the paired Students t-test for the case of two
related samples or repeated measurements on a single sam-
ple. As such it does not make assumptions concerning the
underlying distribution of the measurements.
Table 2 contains a summary of the results of the COP
analysis on the CPR segments. We can see that for both CPR
types the post-CPR COP median value is greater than the
pre-CPR value indicating a general increase in the shock-
ability of the patients due to CPR. The signicance of the
differences in the pre- and post-CPR COP metrics are indi-
cated in the table by the Wilcoxon signed rank sum test p
values. Only the automated CPR shows a signicant result
(p < 0.05), although it should be noted that the data sets are
small.
All rhythm types
In work of a more exploratory nature, we extended the anal-
ysis to include all 212 CPR segment signals; i.e. asystole, VF
and PEA were all included. The initial (i.e. pre-CPR) distribu-
tion of COP scores for these data sets are shown in Figure 5
differentiated by the CPR type. Both distributions are similar
in form with comparable median and mean values. However,
Figure 5 The spectrum of pre-CPR COP values for the manual
and LUCAS CPR showing similar distributions.
270 M.S. Box et al.

0.024
p
of medians
Difference

1.26
Post-CPR COP value

0.1249.13 (5.4/11.5) 7.9 (5.4/11.5)

for presentation purposes.
Pre-CPR COP value
LUCAS

105
of medians
Difference

COP results given as median values with associate q1 and q3 values (in parenthesis). These are divided by
0.99
Post-CPR COP value

9.37 (7.1/10.3) 8.38 (6.7/9.4)

The inuence of CPR on the COP marker valueVF only (87 trace segments)

Figure 6 Box plots of COP distributions according to CPR

types: (a) pre-CPR and (b) post-CPR.

the skewed nature of the pre-CPR distributions is obvious

Pre-CPR COP value

from the gure. Figure 6 contains the pre- and post-CPR

COP distributions displayed as box plots according to CPR-
type. Note that both forms of CPR have led to a signicant
Standard

reduction in the large negative outlying COP values. These

are marked below the bottom whiskers in the plots by the
crosses and correspond mainly to asystole (11 out of 14 of
the manual CPR and 4 out of 7 of the automated CPR).
As a reference, the minimum value of the pre-shock COP
values (Figure 1) of 17.8 105 is plotted as a dotted line
Number of CPR segments (N)

across the gures (note that Figure 1 contains only VF data

which the paramedics considered suitable for shocking). The
corresponding differences between the paired pre- and post-
CPR COP values are shown in the boxplots of Figure 7. The
Lucas

increases in the median of the COP marker due to the applied

38

CPR are 0.97 105 and 1.90 105 , respectively for manual
CPR length

and automated CPR.

Standard

Table 3 contains a summary of the results of the COP

analysis on the CPR segments. A further subdivision of the
49

analysis with respect to the length of time that CPR was

applied (<1, 12, 23, and >3 min) is also contained in the
Time (s)
Table 2

table and shown in Figure 8. We can see that in all cases

the post-CPR COP median value is greater than the pre-
All

CPR value. The signicance of the differences in the pre-

Shock outcome prediction before and after CPR 271

0.0004
0.029
0.028
0.048
0.720
Pre-CPR COP value Post-CPR COP value Difference of mediansp

0.0001 9.9 (6.3/13.7) 8.0 (5.3/12.0) 1.90

0.0009 8.94 (5.3/15.3)7.31 (4.4/12.4)1.63
0.031 8.47 (5.5/12.7)7.88 (4.7/11.8)0.59
0.690 9.69 (7.8/12.6)8.2 (6.7/10.2) 1.49
11.50 (6.9/13.8)0.20
COP results given as median values with associate q1 and q3 values (in parenthesis). These are divided by 105 for presentation purposes.
0.065 11.70 (8.4/16.2)
LUCAS

Pre-CPR COP value Post-CPR COP valueDifference of mediansp

The Inuence of CPR on the COP marker valueall rhythms (212 trace segments)

(6.2/9.9) 0.97
(6.3/10.0)1.04
(6.5/9.7) 1.31
(8.8/12.4)0.08
(4.3/7.8) 2.27

Figure 7 Change in COP marker values according to CPR type.

(a) Original box plots. (b) Zoom into central region of (a) with
zero difference value marked as dotted line.

and post-CPR COP metrics are indicated in the table by

the Wilcoxon signed rank sum test p values. It is noticeable
8.67
8.96
8.21
9.29
(5.5/9.9)5.36

that the signicance deteriorates for longer CPR periods. We

(6.9/12.3)
(6.9/13.7)
(7.2/12.2)
(6.9/11.5)

speculate that this could be due to a combination of factors

including fewer data points available per group, the associ-
ated dissimilarity of pre-CPR input data distributions making
comparison of relative changes difcult, and the fact that
Number of CPR segments (N)Standard

many of the longer periods of manual CPR records included a

9.64
10.00
9.52
9.37
7.63

number hands-off periods during which CPR ceased while

patients were ventilated.
Figure 9a plots the COP values before and after CPR for
the two CPR groups. Note the tight cluster of points at the
top right, plus a few further down and to the left. A log
transformation of the data (Figure 9b) spreads out the tight
cluster where we see the majority of the points are near
a diagonal line for both groups indicating similar changes
StandardLucas

from pre- to post-regardless whether manual or automated.

98
30
35
18
15
Time (s) CPR length

However, there is a noticeable difference between manual

and automated CPR when the pre-CPR COP value is less than
114
48
60119 39
120179 17
10

around 2 106 corresponding to post-CPR values which are

much higher in the manual group. This is conrmed using
Table 3

Wilcoxon rank sum tests which show non-signicant results

059

180

overall (p = 0.85) for both pre-CPR and post-CPR, but gives

All

p = 0.0061 for post-CPR for those with pre-CPR values less

272
Figure 8 Change in COP marker values according to CPR type and CPR period. (Left to right: 01, 12, 23, and >3 min.) Note
that the plots have been cropped and some outliers now lie outwith the plotted regions.
than 2 106 . Analysis of covariance was then performed
on the data adjusting for the pre-CPR value to compare the
performance of the two groups. (This was performed on the
log scaled data as it was more normally distributed.) This
Figure 9 Post-CPR versus pre-CPR COP values for auto-
mated and manual devices. (a) Original COP values. (b) Log
transformed COP values. (Manual CPR: lled diamonds and auto-
mated CPR: empty square.).
M.S. Box et al.
was done only for those with pre-CPR greater than 2 106
as the other data were clearly different. This gives p = 0.20
showing no clear statistical evidence that manual and auto-
mated differ in post-CPR value after adjusting for pre-CPR
value. Analysis of the length of CPR produced p = 0.22 show-
ing no signicance either.
There are 13 manual CPR data points with pre-CPR val-
ues less than 2 106 . These correspond to four separate
patients. No obvious reason for their erroneous behaviour
can be identied. However, it is interesting that the man-
ual CPR has provided this marked positive effect in so many
traces. A larger scale study would allow the signicance of
this to be probed in further detail.
Concluding remarks
The COP measure
The study utilised the COP shock outcome prediction mea-
sure to quantify changes in the ECG, due to the application
of CPR. The measure, developed for VF signals, is indica-
tive in an improvement of the state of the myocardium
in preparation for a debrillation shock. The COP measure
was validated for use in the study by rst analysing the
shock segments of the ECG data record. This resulted in a
performance of the measure comparable with previous stud-
ies with 60% specicity achieved at 100% sensitivity on a
blind test of the data. This result is very similar to those
from other studies by our group and provides condence in
the robustness of the technique across hardware platforms.
Remarkably, also, the COP marker demonstrated an ability
to stratify according to outcomes: asystole, VF, PEA, and
NSR.
CPR analysis
Subsequent to the testing of COP, the measure was com-
puted just prior to and immediately after CPR segments
of the ECG data record. This was initially performed for
those 87 CPR segments where VF was both the pre- and
post-CPR waveform. An increase in the mean COP values for
both CPR types was found to demonstrate that CPR leads
Shock outcome prediction before and after CPR
to an increase in the likelihood of successful debrillation.
This result indicates that the application of CPR does indeed
provide benecial preparation of the heart prior to debril-
lation therapy whether manual or automated CPR is applied.
This is in accordance with the results from the work of oth-
ers. A signed rank sum test found the increase due to manual
CPR not to be signicant (p > 0.05) whereas the automated
CPR was found to be signicant (p < 0.05). This increase was
larger for the automated CPR (1.26, p = 0.024) than for the
manual CPR (0.99, p = 0.124), however, it is recommended
that a substantially larger dataset is used to conrm these
ndings with signicance.
We then explored a more general the use of the COP
marker by applying the method to all 212 CPR segments con-
taining all rhythm types pre- and post-CPR. The original COP
distributions for both data sets (manual and automatic CPR)
were very similar and the resulting levels of signicance for
the change in distributions due to the effect of CPR were
high (p < 0.001 in both cases). This increase is more pro-
nounced for automated ACD-CPR than for manual CPR. A
further analysis of this data involving stratication of the
input data into length of CPR (<1, 12, 23, and >3 min)
all indicated that both types of CPR led to an increase in
the COP value. These increases were quite variable between
CPR types and for longer periods a deterioration in the sig-
nicance of the results (i.e. p > 0.05) occurred. The results
of the analysis within each group indicates that CPR has an
measurable effect on the waveform regardless of rhythm.
However, these increases cannot simply be equated with an
increase in the shockability of the patients due to CPR as
a proportion of the pre- and post-CPR rhythms were PEA or
asystole and are hence not shockable. An analysis carried
out between the groups adjusted for pre-CPR value showed
no signicant difference between the two methods of CPR
(p = 0.20). Similarly, adjusting for length of CPR showed no
signicant difference between the groups (p = 0.22). How-
ever, a number of low value pre-CPR COP measures did
experience a dramatic increase of signicance. This is a
subgroup worthy of further study.
Discussion
Lewis and Niemann11 have suggested the details of use
for the apparently contradictory results of the Hallstrom13
and Ong14 studies of (load distributed) ADC-CPR devices.
Including almost certainly the time to deployment and
inuence of deployment on time to debrillation when
appropriate. Axelsson et al.24 cite the delay between
collapse and the start of the intervention as a reason
for the similar outcomes for manual versus a LUCAS ADC-
CPR device with manual chest compressions performed
until the mechanical chest compression device was set
up.
In the study we present here we detail a method for direct
characterisation (using the COP marker) of the effect of CPR
on the shockability of the patient by considering the ECG
exhibiting VF immediately before and after CPR. This was
done by comparing the computed COP values for VF signals
pre- and post-CPR. By doing so, the time scales (and over-
all procedure) followed between CPR and countershock are
removed from the analysis thus a localised analysis of the
273
efcacy of the CPR alone is performed. The method reported
here therefore provides a denitive, quantitative determi-
nant of the immediate positive effect of both types of CPR
regardless of the details of use.
In addition to the analysis of the pre- and post-CPR VF
rhythms, our work also showed that the COP marker was
able to: (1) stratify according to shock outcomes on the
pre-shock VF waveform and (2) demonstrate a signicant
measurable increase through the actions of both types of
CPR when all rhythm types were analysed. The marker has
therefore shown further potential as a measure of the state
of the myocardium. If COP proves to be a useful measure
for obtaining a metric of the myocardial state for all rhythm
types considered, then it has promise in a wide range of
applications, specically CPR efcacy as a quality measure
and feedback to user.
The work described leads directly to the main aim of the
research effort of our group to develop a suite of tools for
the complete analysis of the ECG signal during OOH car-
diac arrest including: shock outcome prediction; pre- and
post-CPR ECG-analysis to determine effectiveness; analy-
sis during CPR to detect the ongoing efcacy of CPR and,
signicantly, to detect changes in the underlying rhythm dur-
ing CPR25 the latter leading directly to debrillation during
CPR once a shockable rhythm is realised. It is on this latter
area we now concentrate our investigations.
Conict of interest
With regards to the above research Martyn Box states he has
no conict of interest.
Acknowledgements
The authors acknowledge the support of the Wellcome Trust
(Grant number 069078/Z/02/Z) and the Joint Royal Col-
lage Ambulance Liaison Committee (JRCALC) in this work.
The authors also acknowledge the assistance of Dr. Rob
Elton in the statistical analysis. CardioDigital Ltd. is a Well-
come Trust supported University Spin Off Company set up to
research biomedical signal processing applications, includ-
ing the analysis of ventricular brillation. James Watson and
Paul Addison are co-founders and company directors of Car-
dioDigital. Gareth Clegg and Colin Robertson are members
of the CardioDigital Medical Advisory Panel.
References
1. Achleitner U, Wenzel V, Strohmenger HU, et al. The bene-
cial effect of basic life support on ventricular brillation
mean frequency and coronary perfusion pressure. Resuscitation
2001;51:1518.
2. Berg RA, Hilwig RW, Kern KB, Ewy GA. Precountershock car-
diopulmonary resuscitation improves ventricular brillation
median frequency and myocardial readiness for success-
ful debrillation from prolonged ventricular brillation:
a randomised, controlled swine study. Ann Emerg Med
2002;40:56371.
3. Kolarova J, Ayoub IM, Yi Z, Gazmuri RJ. Optimal timing for
electrical debrillation after prolonged untreated ventricular
brillation. Crit Care Med 2003;31:20228.
274
4. Eftestol T, Wik L, Sunde K, Steen PA. Effects of CPR on predic-
tors of VF debrillation success during out-of-hospital cardiac
arrest. Circulation 2004;110:105.
5. Cobb LA, Fahrenbruch CE, Walsh TR, et al. Inuence of car-
diopulmonary resuscitation prior to debrillation in patients
with out-of-hospital ventricular brillation. J Am Med Assoc
1999;281(13):11828.
6. Wik L, Hansen TB, Fylling F, et al. Delaying debrillation to give
basic cardiopulmonary resuscitation to patients with out-of-
hospital ventricular brillationa randomized trial. J Am Med
Assoc 2003;289(11):138995.
7. Abella BS, Alvarado JP, Myklebust H, et al. Quality of cardiopul-
monary resuscitation during in-hospital cardiac arrest. JAMA
2005;293:30510.
8. Wik L, Kramer-Johansen J, Myklebust H, et al. Quality of
cardiopulmonary resuscitation during out-of-hospital cardiac
arrest. JAMA 2005;293:299304.
9. Ochoa FJ, Ramalle-Gomara E, Lisa V, Saralegui I. The effect of
rescuer fatigue on the quality of chest compressions. Resusci-
tation 1998;37(June (3)):14952.
10. Hightower D. Decay in quality of closed-chest compressions over
time. Ann Emerg Med 1995;26:3003.
11. Lewis RJ, Niemann JT. Manual vs. device-assisted CPR. JAMA
2006;295(22):26614.
12. Steen S, Liao QM, Pierre L, et al. Evaluation of LUCAS, a
new device for automatic mechanical compression and active
decompression resuscitation. Resuscitation 2002;55(3):28599.
13. Hallstrom A, Rea TD, Sayre MR, et al. Manual chest compression
vs. use of an automated chest compression device during resus-
citation following out-of-hospital cardiac arrest: a randomized
trial. JAMA 2006;295:26208.
14. Ong MEH, Ornato JP, Edwards DP, et al. Use of an automated,
load distributing band chest compression device for out-of-
hospital cardiac arrest resuscitation. JAMA 2006;295:262937.
15. Watson JN, Uchaipichat N, Addison PS, et al. Improved
prediction of debrillation success for out-of-hospital VF car-
diac arrest using wavelet transform methods. Resuscitation
2004;63(3):26975.
M.S. Box et al.
16. Watson JN, Addison PS, Clegg GR, Steen PA, Robertson CE.
Practical issues in the evaluation of methods for the predic-
tion of shock outcome success in out-of-hospital cardiac arrest
patients. Resuscitation 2006;68:519.
17. Watson JN, Addison PS, Clegg GR, Steen PA, Robertson
CE. Wavelet transform-based prediction of the likelihood of
successful debrillation for patients exhibiting ventricular b-
rillation. Meas Sci Technol 2005;16:L16.
18. Addison PS. The illustrated wavelet transform handbook:
introductory theory and applications in science, engineering,
medicine and nance. Bristol, UK: Institute of Physics Publish-
ing; 2000.
19. Addison PS. Wavelet analysis of the ECG: a review. Physiol Meas
2005;26:R15599.
20. Addison PS, Watson JN, Clegg GR, Holzer M, Sterz F, Robertson
CE. A novel wavelet based analysis reveals hidden struc-
ture in ventricular brillation. IEEE Eng Med Biol 2000;19(4):
38392.
21. Watson JN, Addison PS, Clegg GR, Holzer M, Sterz F, Robertson
CE. Evaluation of arrhythmic ECG signals using a novel wavelet
transform method. Resuscitation 2000;43(2):1217.
22. Addison PS, Watson JN, Clegg GR, Steen PA, Robertson CE.
Finding coordinated atrial activity during ventricular brilla-
tion using wavelet decomposition. IEEE Eng Med Biol 2002;21:
5865.
23. Addison PS, Watson JN, Feng T. Low-oscillation complex
wavelets. J Sound Vibration 2002;254(4):73362.
24. Axelsson C, Nestin J, Svensson L, Axelsson AB, Herlitz J. Clini-
cal consequences of mechanical chest compression in the EMS
system for treatment of out-of-hospital cardiac arresta pilot
study. Resuscitation 2006;71:4755.
25. Watson JN, Addison PS, Uchaipichat N, Clegg GR, Steen PA,
Robertson CE. Extracting heart rhythm information during
CPR using wavelet transform methods. In: MEDSIP: advances
in medical signal and information processing, conference.
2006.