VOL 56, No 4, OCTOBER 1977
CIRCULATION
552
tients
3. Guiney TE, Rubenstein JJ, Sanders CA, Mundth ED: Functional evaluation of coronary bypass surgery by exercise testing and oxygen consumption. Circulation 47 & 48 (suppl III): 111-141, 1973
4. Dodek A, Kassebaum DG, Griswold HE: Stress electrocardiography in
the evaluation of aortocoronary bypass surgery. Am Heart J 86: 292,
1973
5. Balcon R, Honey M, Rickards AF, Sturridge MF, Walsh W, Wilkinson
RK, Wright JEC: Evaluation by exercise testing and atrial pacing of results of aortocoronary bypass surgery. Br Heart J 36: 841, 1974
6. Siegel W, Lim JS, Proudfit WL, Sheldon WC, Loop FD: The spectrum
of exercise test and angiographic correlations in myocardial revascularization surgery. Circulation 51 & 52 (suppl 1): I-156, 1975
7. Merrill AJ, Thomas C, Schrechter E, Cline R, Armstrong R, Stanford
W: Coronary bypass surgery: Value of maximal exercise testing in assessment of results. Circulation 51 & 52 (suppl I): 1-173, 1975
8. Bruce RA, Hornsten TR: Exercise stress testing in evaluation of patients
with ischemic heart disease. Prog Cardiovasc Dis 11: 371, 1969
9. Balke B, Grillo G, Konecci E, Luft U: Work capacity after blood donation. J Appl Physiol 7: 231, 1954
10. Sheffield LT, Roitman D, Reeves TS: Submaximal exercise testing. J
South Carolina Med Assoc 65 (suppl I): 18, 1969
11. Sones FM Jr, Shirey ED: Cine coronary arteriography. Mod Conc Cardiovasc Dis 31: 735, 1962
12. Judkins MP: Selective coronary arteriography. Radiology 89: 815, 1967
13. Barry WH, Pfeifer JF, Lipton MJ, Tilkian AG, Hultgren HN: Effects of
coronary artery bypass grafting on resting and exercise hemodynamics in
patients with stable angina pectoris: A prospective randomized study.
Am J Cardiol 37: 823, 1976
14. Martin CM, McConahay DR: Maximal treadmill exercise electrocardiography. Correlations with coronary arteriography and cardiac hemodynamics. Circulation 46: 956, 1972
in this subset is very small in each series, and there
well be differences in the extent of the CAD in the
patient groups (13 of 20 such patients in our study had only
single-vessel CAD). The other studies parallel our own in
showing a low incidence of false positive tests (14-33%) but
an excessive incidence of false negative tests (55-73%). It is
apparent that treadmill stress testing provides an imperfect
means of evaluating the success of myocardial revascularization and that the high incidence of false negative
exercise tests limits the usefulness of such a normal test in
the individual postoperative patient.
may
Acknowledgment
The authors gratefully acknowledge the secretarial assistance of Mrs. Joyce
Neumann.
References
Lapin ES, Murray JA, Bruce RA, Winterscheid L: Changes in maximal
exercise performance in the evaluation of saphenous vein bypass surgery.
Circulation 47: 1164, 1973
2. Bartel AG, Behar VS, Peter RH, Orgain ES, Kong Y: Exercise stress
testing in evaluation of aortocoronary bypass surgery: Report of 123 patients. Circulation 48: 141. 1973
I.
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
Estimation of the Probability of Exercise-induced
Ischemia by Quantitative ECG Analysis
MAARTEN L. SIMOONS, M.D.,
AND
PAUL G. HUGENHOLTZ, M.D.
SUMMARY In order to improve the value of exercise tests for the
detection of coronary artery disease (CAD) a system for on-line computer processing of the Frank lead exercise ECG was developed.
Data were analyzed from 95 patients with CAD and 129 ostensibly
healthy men. All subjects had a normal ECG at rest.
Visual ECG interpretation during exercise yielded a sensitivity of
50% and a specificity of 95%. A large number of QRS and ST
measurements were compared by discriminant function analysis in a
group of 86 normal subjects and 52 patients (designated training
group). Best results were obtained with a combination of two ST
amplitudes from lead X: sensitivity, 85%, specificity, 90%. This was
confirmed in a test group of 43 patients and 43 normal subjects.
The results of the discriminant function were expressed as the
likelihood ratio for an abnormal or normal ST segment at a given
heart rate, a figure which provides a quantitative assessment of the
degree of exercise-induced ischemia. This is a more realistic approach than classification into normal or abnormal since persons with
and without CAD fall along the same continuous spectrum.
GRADED EXERCISE TESTS on a bicycle ergometer or
treadmill have been advocated as reliable methods for
the early detection of subjects with coronary artery disease
(CAD). In recent studies a correlation has been shown
between the presence of ST depressions in the electrocardiogram during exercise (XECG) and the findings on selective cine-coronary arteriography in male patients with chest
pain.1 Furthermore it has been shown that asymptomatic
subjects with ST depressions during exercise have a considerably higher probability of developing manifest CAD
than those without such ECG changes.8-1 However, the ex-
ercise tests which
on a
From the Thoraxcenter, Erasmus University, Rotterdam, The
Netherlands.
Address for reprints: Maarten L. Simoons, M.D., Thoraxcenter, Erasmus
University, P.O. Box 1738, Rotterdam, The Netherlands.
Received March 25, 1977; revision accepted May 24, 1977.
are
currently in
use
have four important
shortcomings.
1) The interpretation of the tracings is often difficult
when excessive muscle noise, baseline drift or artifacts
occur, despite careful skin preparation and electrode
fixation.
2) Part of the test cannot be interpreted when the patient
terminates the test before sufficient stress of the cardiovascular system is provoked.
3) As a result of these and other factors, the fraction
"false negative" tests reported in all studies is considerable. The sensitivity of the ECG during symptomlimited exercise tests for prediction of obstructive
CAD ranges from 52% to 85% at a specificity level
between 83% and 100% (fig. 1).
4) According to current criteria, exercise ECGs are inter-
553
PROBABILITY OF EXERCISE-INDUCED ISCHEMIA/Simoons, Hugenholtz
TABLE 1. Selection Criteria of the Subjects in the Training
Set and in the Test Group
Ostensibly healthy men,
no complaints of chest pain
blood pressure <160/95
serum cholesterol <6.7 mM/L
Patient with chest pain and
79% or greater obstructions
in one or more coronary arteries
Patients with typical angina
pectoris during the exercise
test, which disappeared within 5
min after exercise
Patients with a nontransmural
myocardial infarction: typical
history, serial enzyme changes
and serial ECG changes
Training
Test
group
group
86
43
50
-
100
90
80
70
60
--I 701
->I
I7
.
,
-,,t
I
_ 90
S
E
32
9
11
30
S
T
I
70
0D
T
4
9
(i
Y
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
Details of the selection procedure of the normal subjects have been
published in a previous report.22 All subjects had a normal ECG at rest.
6.7 mM/L = 260 mg% cholesterol.
preted either as "normal" or as "abnormal." This
seems a rather arbitrary separation since in reality a
whole spectrum exists between subjects with normal
coronary arteries and patients with completely
occluded vessels. It is also likely that ischemia
develops gradually when the oxygen demand of the
myocardium increases during exercise.
Computer processing of the ECG signal during and after
exercise may help to overcome each of these four shortcomings. The signal quality can be improved by averaging of
a number of selected beats.'2`i6 Different types of ST
measurements can be taken'6-20 and can be compared with
values obtained in normal subjects at the same heart rate
during or after exercise.
The results of the present investigation indicate that in
this manner the fraction of false negative tests can be
reduced. Finally, a procedure is presented which permits
assessment of the degree of abnormality of the ECG at all
stages of the test.
Selection of Subjects and Methods
Exercise electrocardiograms were analyzed from 129 normal men and 95 male patients with coronary artery disease.
The selection criteria are presented in table 1. The records
from 86 normals and 52 patients were used as a training set
for development of criteria for computer assisted interpretation of XECGs, those from the remaining 43 normals and
I
I
I
I
Test set
Normal subjects
Patients
T
=
6
4
9
2
1
1
total number of subjects in each group.
50
(A).
43 patients were employed as a test set for verification of the
results from the training set.
All subjects had a normal QRS complex and ST segment
at rest according to the Minnesota code (21): Q waves indicative of myocardial infarction were absent and the ST-T
segment was within normal limits. Patients with P wave abnormalities and first degree atrioventricular (A-V) block
were disregarded. None of the subjects had Wolff-Parkinson-White syndrome (WPW). Patients receiving digitalis or
3-blocking drugs within two weeks prior to the exercise test
were not included. The subjects were selected without prior
knowledge of the XECG. The age distributions of the subjects are shown in table 2.
.All subjects underwent a symptom limited exercise test on
a bicycle ergometer with stepwise workload increments of 10
W/min or 30 W/3 min. Details on the procedure have been
published elsewhere.22 ST depressions were the reason to terminate the test in only nine patients. After the highest level
of exercise, cycling was continued during 6 min in order to
prevent discomfort caused by pooling of blood in the legs.
All tests at the Thoraxcenter were performed in the same
air-conditioned room. Temperature was kept between 16°
and 20°C. A physician and a technician were always present.
61-65
T
4
6
5
6
12
23
13
20
10
7
4
2
86
52
1
4
13
20
10
12
10
8
3
1
1
43
43
1
60
SPECIFICITY
(years)
5
®A
_
FIGURE 1. Sensitivity and specificity of visual interpretation of
the exercise ECG for prediction of obstructions in the coronary
arteries. The results are given of seven recent studies (references
1-7) employing symptom limited exercise tests on a bicycle
ergometer or on a treadmill (1-7), and of the present investigation
TABLE 2. Age Distribution of the Patients and Normal Subjects in the Two Sets of Data
51-55
56-60
46-50
21-25 26-30 31-35 36-40 41-45
Control set
Normal subjects
Patients
Q 80
O
N
554
CIRCULATION
A defibrillator, oxygen, anti-arrhythmic drugs and nitrates
were immediately available. Blood pressure was measured
with a cuff around the upper arm at rest and every second
minute during and after exercise.
The Frank lead ECG was recorded with chest electrodes
at the level of the fifth intercostal space in sitting position.
The F electrode was placed on the sacrum, the H electrode
on the neck. Much attention was paid to proper preparation
of the skin (rubbing with gauze soaked in alcohol) and fixation of the electrodes and wires. The ECG was continuously
monitored on a nonfading oscilloscope, and recorded on
paper with alternating low (10 mm/sec), normal (25
mm/sec) and high (50 mm/sec) paper speeds. The calibration was 1 mV/cm. In addition the ECG was continuously
recorded on a FM tape recorder.
Data Processing
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
The ECGs were processed with a PDP-8E computer
system. The three orthogonal leads were sampled during 20
seconds at rest, in supine and sitting position, every second
or third minute during exercise and the first four minutes of
the recovery period. The sample rate was 500 per second.
Single representative beats were computed by selective
averaging as reported earlier.'5 Waveform analysis was done
by the program and checked visually. A large number of
measurements were taken from the averaged ECGs including:
1) amplitudes and slopes at 10 msec intervals from 0 until
90 msec after the end of the QRS complex.7' 19-20
2) Time normalized amplitudes and slopes of the PQ,
QRS and ST segment.'6
3) Time-integrals of the positive and negative parts of the
ST segment.'8
An attempt was made to improve the separation of
patients and normal subjects by combinations of ECG
measurements instead of a single measurement. Such a combination is called a discriminant function. A discussion of
the mathematical aspects of these discriminant functions is
beyond the scope of this paper, but can be found in statistical
handbooks."
Linear discriminant function analysis was performed with
a program developed at the University Computer Center in
Utrecht, utilizing a CDC 7600 system. This program
generated discriminant functions with all possible combinations of at most four parameters. This limitation was
chosen because the series were relatively small. The results
of different algorithms of separation of normals and patients with CAD were expressed in terms of the sensitivity
(fraction of abnormal findings in the patients) and the
specificity (fraction of normal findings in the control group).
TABLE 3. Discriminant Function Analysis of ST Amplitudes at Fixed Intervals after the End of QRS
Measurements
Spec.
Sens.
ST50Y, ST70X
ST50Z, ST70Y, ST90X
ST308, ST501, ST70Y, ST90X
0.78
0.83
0.84
0.84
0.90
ST80X
0.87
0.88
0.90
Based on maximum exercise records.
ST80 = 80 msec after QRS etc; Spec. = specificity (fraction of normal
ECGs in the normal subjects); Sens.
sensitivity (fraction of abnormal
ECGs in the patients).
VOL 56, No 4, OCTOBER 1977
A.
mV
0 2-
x
a.
0.2-
n-A
uI
-0.2
:
B.
.
-
-
s~~~~~~~*
=
a
HEART RATE
100
200
FIGURE 2. Plot of the amplitudes at 60 msec after the end of QRS
in lead X (ST605) in 86 normal subjects (a) and in 52 patients with
CAD (b). The continuous lines represent the boundaries of the area
in which the measurements from the normal subjects were found.
The dotted line corresponds to a 0.1 m V ST depression.
Results
Since only subjects were selected who had a normal QRS
complex and ST segment at rest, it is not surprising that
even with multivariate analysis of the ECG measurements at
rest, a poor separation of the patients and normal subjects in
the training set was achieved. With a linear discriminant
function of four ST measurements the sensitivity and the
specificity were each 67%. This indicated that the ECGs at
rest in the patients were very similar to those in the normals.
The best separation of patients and normals at the highest
workload was obtained with ST amplitudes at a fixed interval after the end of the QRS complex. Representative results
are presented in table 3. Time-normalized ST amplitudes,
ST slopes, ST areas and ST Chebyshev coefficients did not
improve the separation of the two groups, nor did transformation to polar coordinates or measurements of the
differences between the rest and exercise ECG. Extensive
data on these different measurements will be published
elsewhere.'4
In a previous study it was demonstrated that many ST
measurements in normal subjects change gradually during
-exercise.22 An example of this pattern is shown in figure 2.
Two straight lines could be computed which enclose the
measurements in the normal subjects. A considerable improvement of the separation of patients and normals could
be obtained when the distances of the ST amplitudes toward
line II were employed instead of the actual amplitudes.
Representative results are presented in table 4.
The results of application of the derived criteria for interpretation of exercise ECGs to the independent test set are
presented in table 5. Visual interpretation of the ECGs acTABLE 4. Diseriminant Function Analysis of the Distance
from the ST Measurements to the Lower Limit of Meacsurements
in the Normal Subjects at the Same Heart Rate (fig. 2)
Measurement(s)
ST60x
ST20., ST80x
ST1IO, ST20, ST40, ST50X
=
For legend see table 3.
Spec.
Sens.
0.93
0.91
0.90
0.81
0.85
0.87
555
PROBABILITY OF EXERCISE-INDUCED ISCHEMIA/Simoons, Hugenholtz
TABLE 5. Comparison on Specificity and Sensitivity of Visual ECG Interpretation and of Computer ECG processing with ST Amplitudes Adjusted for Instantaneous Heart Rate
Training Group
spec.
Visual ECG interpretation
Heart rate dependent criteria
ST60
ST20X, ST80S
Test Group
spec.
sens.
sens.
81/86
0.94
26/52
0.50
41/43
0.95
22/43
0.51
80/86
0.93
0.91
44/52
42/52
0.81
0.85
38/43
40/43
0.93
0.88
30/43
36/43
0.70
0.84
78/86
The criteria were derived from a training group of 86 normals and 52 patients. The results were verified in an independent test
group of 43 patients and 43 normals.
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
cording to the standard criterion of a 0.1 mV horizontal ST
depression yielded a sensitivity of 50% and 51% in the two
groups while the specificities were also similar: 94% and
95%. Furthermore excellent reproducibility was obtained
with a linear combination of two heart rate adjusted ST
amplitudes from lead X: sensitivities 85% and 84%;
specificities 91% and 88% in the two groups.
Further analysis of the combined data from the training
group and the test group indicated that the difference
between the results obtained by the three criteria in table 5
was not due to the choice of the cut-off point for separation
of the two groups since the sensitivity of the combination of
ST20x and ST80x exceeded the sensitivity of ST601 at all
cut-off points which yielded a specificity between 70% and
90%. This is illustrated in figure 3. Both criteria for computer assisted interpretation of the XECG were superior to
visual ECG reading, as shown in the figure.
No significant differences were observed between the
XECGs in the three groups of patients selected according to
different criteria (table 1). The combination of ST20x and
ST80x at the highest workload produced an abnormal
50
S
E
60
80
70
<
S
_N
90
100
90
8
T
70
v
T
Y
CTrnra
S PE C F C T Y
FIGURE 3. Sensitivity and specificity of two measurements from
the ST segment. These data are from the ECGs at the highest heart
rate in 129 normal subjects and 95 patients with CAD. The sensitivity decreases when the specificity increases. At all specificities
between 70% and 90% the sensitivity of the combined ST20X and
ST80X amplitudes is better than the sensitivity of the ST60X amplitude. The closed triangle represents the results of visual interpretation of the XECG in the present study.
response in 34 out of 41 patients with documented coronary
artery obstructions, in 35 out of 41 patients with angina pectoris during the exercise test, and in 11 out of 13 with a nontransmural infarction.
Interpretation of the ECG at Different
Levels of Exercise
In the preceding paragraphs only the data obtained at the
highest workload reached by each individual were analyzed.
However, it is convenient to perform on-line analysis of the
ECG at all stages of the exercise test. This might lead to
detection of ECG abnormalities at an early stage of the test
before they can be distinguished by visual inspection of the
ECGs. It would then also become possible to terminate
those tests at a lower workload. Therefore the two best
criteria for classification of the ECG at the highest workload
were applied to all records at rest and during exercise from
the patients and the controls from both the training group
and the test group. The results in table 6 show some reduction in specificity for ST60. at lower heart rates and a large
reduction of the specificity of the combined ST20X and
ST80X amplitudes. This indicates that the threshold values
for separation of normal and abnormal ECGs should be
different at high and low heart rates.
In figure 4 the results of the discriminant function applied
to all records obtained during exercise have been plotted.
These discriminant scores of the patients and the normal
subjects at low heart rates overlap almost completely. This is
in accordance with the fact that only patients with a normal
ECG at rest were selected for these studies. At higher heart
rates the separation of the patients and the controls
gradually improved. However, at heart rates over 160
beats/min the measurements from the two groups
overlapped again. Thus the 15 patients who reached these
high heart rates during exercise could not be distiw-uished
from the normal subjects on basis of the XECG.
Five of these patients had documented obstructions in the
coronary arteries, three had had a previous infarction, and
TABLE 6. Comparison of Specificity and Sensitivity of Two
Criteria for Interpretation of Exercise ECGs at Different Workloads
Max.
Workload only
spec. sens.
All records
HR >100
spec. sens.
All records*
rest + exercise
spec. sens.
0.93 0.74
0.90 0.84
0.90 0.77
0.57 0.85
0.84 0.77
0.30 0.94
l
ST60
ST20, ST80
*129 normal subjects and 95 patients with coronary artery disease.
Note reduction of specificity when the records at early stages of the test
are included, in particular, for the discriminant function of ST20x and
ST80x.
HR = heart rate.
556
VOL 56, No 4, OCTOBER 1977
CIRCULATION
40
ST20
that the injury currents were cancelled. At present it is not
possible to prove or disprove one of these or other
mechanisms because no independent quantitative information on the degree and extent of ischemic myocardial areas
was available.
ST80X
If,
1
0
rX
)~~~~~~~
60-79
LI1
I
Estimation of the Probability of
Myocardial Ischemia During Exercise
80-99
[X
t
100 -119
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
FIGURE 4. Histograms of the combined ST20X and ST80X
amplitudes in different heart rate intervals. The continuous line indicates the patients and the dotted line the normal subjects. Note a
gradual improvement of the separation of the two groups when
heart rate increases. However, at heart rates over 160 beats/min the
measurements overlap completely.
seven experienced chest pain during the exercise test. It may
be argued that no myocardial ischemia occurred in these patients, in spite of the strenuous exercise. However, the chest
pain in seven of these suggests otherwise. Also it is possible
that small ischemic areas were present, with locations such
40-
When is is assumed that the measurements have normal
distributions, the ratio can be computed of the probability
that a given value is found in the patients and the probability
that it occurs in the controls. This quantity is called the
likelihood ratio.23 In figure 5 the likelihood ratio has been
plotted for different scores of the discriminant function of
ST20X and ST80X at heart rates between 140 and 159
beats/min. For example, at a discriminant score of -100 ,V
the likelihood ratio abnormal/normal equals 25, while the
ratio is 3.7 at a score of -50 1AV. In this particular example
the ratio equals one at a discriminant score of zero. The
likelihood ratio functions corresponding to the heart rate intervals in figure 4 have been plotted in figure 6. Those for the
heart rates from 60 to 140 beats/min are more or less
parallel (curves 1-4). The positions of these curves relative
to the horizontal axis correspond to the positions of the
histograms of the normal subjects along the same axis in
figure 4. The other two curves appeared to be less steep. In
particular the likelihood ratios at the highest heart rate interval are much closer to unity than the others (curve 6, 160
to 179 beats/min). This reflects the overlap of measurements
in this range of heart rates.
In order to test its applicability for on-line interpretation
of XECGs throughout the test, the likelihood procedure was
applied to all records in this study. A ratio abnormal/normal of 2.7 or more was found in at least one record in 17 of
the 129 normal subjects (13%) and in 70 of 95 patients
(74%). This likelihood ratio corresponded to a specificity of
95% at the different heart rate intervals. In only three normal subjects (2%) such a likelihood ratio was found twice or
more during the test, while this occurred in 57 patients (60%).
Thus in clinical practice the exercise tests can be terminated
after two observed likelihood ratios greater than or equal to
2.7 without increasing the fraction of false positive tests.
Discussion
-200
0
\
100-
200
0V
In the introduction four shortcomings of visual interpretation of exercise ECGs were given. The results of the present
~~LIKELIHOOD - RATIO
ABNORMAL/NORMAL
1=-HR 60 -79
2-=
3=
4
10-
1\
100]
ST
20x ST8Ox
=
5=
6
80 -99
100 -119
120 -139
140 -159
6 =- 160 - 179
1o-
1-
0.1-I
FIGURE 5. Histogram of the combined ST20X and ST80X
amplitudes at heart rates between 140 beats/min and plot of the
likelihood ratio function at different values of the combined ST
measurements.
-200
01
-100
LIKELIHOOD -RATIO ABNORMAL I NORMAL
ST 2, STBOX
FIGURE 6. Likelihood ratio functions at the different heart rate intervals in figure 4.
PROBABILITY OF EXERCISE-INDUCED ISCHEMIA/Simoons, Hugenholtz
investigation indicate that these can be avoided to a large extent by computer assisted ECG interpretation. Thus a considerable improvement of the diagnostic value of exercise
ECGs can be achieved.
Part of this gain by computer processing was due to noise
reduction and part to detection of small ECG changes in
patients who stopped the test at a low heart rate. In addition,
the results of the exercise ECG could be presented in a
manner which indicated the probability of exercise induced
myocardial ischemia at all stages throughout the test.
Selection of Patients and Normal Subjects
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
It is obvious that, in order to prevent bias, the XECG
itself should not be used as a criterion for the selection of
either patients or controls. In most recent studies of the correlation between CAD and the XECG,1-7 a narrowing of
70% or more in one of the major coronary arteries was
chosen as the independent selection criterion. In the present
study this criterion was maintained. In order to enlarge the
data base, patients with a documented myocardial infarction as well as patients who had angina pectoris during the
exercise test were also accepted. It was expected that this
would not influence the results significantly since in the study
of Bartel et al.4 87% (203 out of 235) of the subjects with
angina pectoris during the test had coronary artery narrowings exceeding 70%. Furthermore grouping all patients
together is supported by the observation that the results of
visual ECG reading as well as computer assisted interpretation in the present investigation appeared to be independent of the selection criteria.
It should be emphasized that the patients were selected on
clinical evidence (infarcts), on anatomical evidence
(angiography) or on basis of the history (angina pectoris
during the stress test). These criteria indicate the presence of
CAD. However, it does not necessarily follow that all
patients had coronary insufficiency during the test. Furthermore the occurrence of coronary insufficiency is often but
not .necessarily accompanied by detectable ST segment
changes. This is explained in part by our earlier observation
that the spatial orientation of the ST shift in patients with
CAD and a normal ECG at rest cannot be distinguished
from the orientation of the ST shift in normals.'5
In previous studies on the diagnostic power of XECGs,
patients without significant abnormalities in the coronary
arteriogram were used as controls.' This was not done in
the present study since it might have biased the results. At
the Thoraxcenter coronary arteriograms are made mainly in
those patients who are likely candidates for coronary bypass
surgery. All these patients have a history of severe chest pain
suggesting CAD. As a result of this selection procedure
significant narrowings of the major coronary arteries could
be demonstrated in all but 10% of the patients who underwent an exercise test and coronary arteriography during
the study period. Therefore, as a control group, subjects
were selected who had no signs of manifest CAD and in
whom the major risk factors for development of CAD were
absent. These men had no history of chest pain, a normal
ECG at rest, a normal blood pressure and normal serum
cholesterol levels. Despite these precautions it still remains
possible that some subjects in the control group had abnormal coronary arteries. In fact, seven of these had a 0.1 mV
557
or greater ST depression during exercise. However, this factor is of no great influence when a specificity of 90-95% is
chosen for the diagnostic criteria. Preliminary data on comparison of the reference population in the present study with
subjects without significant obstructions in the coronary
arteriogram studied at another hospital indicate that the
XECGs in these groups are similar.26
Comparison with Other Studies
Visual interpretation of the exercise ECG yielded a sensitivity of 50% and a specificity of 95%. These results are in
good agreement with those reported by Ascoop et al.3 In
other investigations higher sensitivities were found. This
difference is probably related to patient selection. For example in two studies' I patients with slight repolarization abnormalities at rest were included while other authors2' 4 included patients with abnormal QRS complexes. In two
reports6 7 very high sensitivities were found (85% and 83%).
These may be due to the small numbers of subjects studied
by these authors (57 and 50 patients respectively). The
largest series4 produced a sensitivity of 65% and a specificity
of 91% in 367 patients with both normal and abnormal
ECGs at rest. The strong influences of the patient selection
are further illustrated by the results of coronary arteriography in asymptomatic subjects with an "abnormal" XECG
since in about half of these the coronary arteries were completely normal.27, 28
"Best" ECG Measurements during Exercise
The best separation between patients with CAD and normal subjects was obtained by ST measurements from lead
X. Addition of measurements from the other two leads did
not separate the two groups any further. These results are in
agreement with other reports.20' 25 Therefore a single bipolar
lead like CM, or CB, may be sufficient for detection of myocardial ischemia during exercise in patients with a normal
ECG at rest.20 However, in patients with abnormal ECGs at
rest, multiple leads remain necessary since the ST changes
during exercise in these patients occur in all possible directions2" and may therefore be missed by a single lead. The
best parameter for prediction of CAD appeared to be a
linear combination of two ST amplitudes. Addition of other
parameters like ST slopes or ST time integrals did not improve the results. This may seem contradictory to the results
obtained by McHenry et al.'7 and Ascoop at al.'0'
However, when one realizes that a combination of the ST
amplitude and ST slope as proposed by these authors is
mathematically the same as a linear combination of two
amplitudes from the same lead, these findings are not surprising.
In the present investigation the corrected orthogonal
Frank lead system was used. The same type of criteria may
be applied to conventional leads when the changes of the
measurements from these leads during exercise have been established.
The Probability of an Abnormal ECG Response during Exercise
Exercise tests are frequently performed in order to find
objective evidence of CAD in patients who present with
atypical chest pain. The results of the test are usually
558
CIRCULATION
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
reported as "positive" when ST depressions are found, or
"negative" when the ECG stays within normal limits. In
reality, however, a considerable number of tracings are
neither clearly "normal" nor "abnormal." Therefore a
procedure was developed to estimate the probability of
exercise-induced myocardial ischemia for one individual patient. The likelihood ratio was chosen since it can readily be
understood by clinicians. This ratio is the quotient of the
probabilities that a given measurement is found in the
patients or in the normal reference population. Thus the
likelihood ratio is a measure of the degree of abnormality of
a given set of measurements. This ratio has been computed
with the assumption that the measurements have normal distributions. This is certainly not true for all data (see fig. 4).
However, such a finding does not necessarily mean that the
results obtained for data not normally distributed are incorrect, as pointed out by Cornfield.29 In the present series,
for example, a likelihood ratio of 2.7 corresponded to a
specificity close to 95% at all stages of the test even though
the ST parameters at different heart rates were not always
normally distributed (fig. 4).
The fraction of false positive results increased when
criteria for classification of the ECG which had been
developed from maximum exercise records were applied to
earlier stages of the test (table 6). This could be corrected by
computation of likelihood ratio functions at different heart
rate intervals. Therefore the developed criteria may be
applied to all records throughout a symptom limited exercise test, as well as to records obtained during a submaximal
test, provided that similar workload increments are used.
Conclusions
Detailed computer analysis of the ECG during exercise in
the present series resulted in an increased sensitivity relative
to standard visual interpretation for detection of CAD, especially when heart rate was taken into account, while the
specificity could be kept at the same level. These results in a
training population of 86 normal men and 52 patients with
CAD could be reproduced in a test population of 43 patients
and 43 normals. While these results are highly encouraging,
verification of the developed criteria in larger series is mandatory before they can be used routinely.30 In particular
further studies are needed of the ECG changes during exercise in male patients with abnormal ECGs at rest and also in
women, since these were not included in the present in-
vestigation.
The results of the ECG analysis could be expressed on a
continuous scale with a likelihood ratio procedure. With this
method the probability of myocardial ischemia during the
exercise test can be estimated instantaneously. Tests which
are performed for diagnostic reasons may now be terminated as soon as a certain confidence level is reached. This
should reduce the risks of the test, as well as the discomfort
of the patient and the time involved in exercise testing.
Furthermore this approach may help to circumvent the contradictions discussed in recent editorials3l-33 since the likelihood ratio can easily be combined with the prevalence of
coronary artery disease in different settings in order to obtain the true probability of obstructive coronary artery disease in a given subject.29
It is concluded that quantitative analysis of exercise ECGs
VOL 56, No 4, OCTOBER 1977
with a computer system will enhance the value of exercise
tests as a noninvasive method for the detection of coronary
artery disease and for the assessment of the effect of treatment in patients with this illness.
Acknowledgment
The authors thank Els Leuftink, Therese Krebbers, Ed Smallenburg, Joop
de Wit, Annelies van de Velde, Muriel Wildschut, Astrid Melker and Maud
van Nierop for technical and secretarial assistance.
References
1. Ascoop CA, Simoons ML, Egmond WG, Bruschke AVG: Exercise test,
history, and serum lipid levels in patients with chest pain and normal electrocardiogram at rest; comparison to findings at coronary arteriography.
Am Heart J 82: 609, 1971
2. Martin CM, McConahay DR: Maximal treadmill exercise electrocardiography; Correlations with coronary arteriography and cardiac
hemodynamics. Circulation 46: 956, 1972
3. Ascoop CA, Distelbrink CA, Mattart AVJ: De waarde van elektro- en
vectorcardiografie tijdens inspanning voor de diagnose van ischemische
hartziekten. Ned Tijdschr Geneesk 117: 146, 1973
4. Bartel AG, Behar VS, Peter RH, Orgain ES, Kong Y: Graded exercise
stress tests in angiographically documented coronary artery disease. Circulation 49: 348, 1974
5. Piessens J, Van Mieghem W, Kesteloot H, DeGeest H: Diagnostic value
of clinical history, exercise testing and atrial pacing in patients with chest
pain. Am J Cardiol 33: 351, 1974
6. Linhart JW, Turnoff HB: Maximum treadmill exercise test in patients
with abnormal control electrocardiograms. Circulation 49: 667, 1974
7. Rios JC, Hurwitz LE: Electrocardiographic responses to atrial pacing
and multistage treadmill exercise testing; Correlation with coronary
arteriography. Am J Cardiol 34: 661, 1974
8. Kattus AA, Jorgensen CR, Worden RE, Alvaro AB: ST-segment depression with near-maximal exercise in detection of preclinical coronary heart
disease. Circulation 44: 585, 1971
9. Aronow WS: Thirty-month follow-up of maximal treadmill stress test
and double Master's test in normal subjects. Circulation 47: 287, 1973
10. Froelicher FV, Thomas MM, Pillow C, Lancaster MC: Epidemiologic
study of asymptomatic men screened by maximum treadmill testing for
latent coronary artery disease. Am J Cardiol 34: 770, 1974
11. Ellestad MH, Wan MKC: Predictive implication of stress testing;
Follow-up of 2700 subjects after maximum treadmill stress testing. Circulation 51: 363, 1975
12. Brody DA: The laboratory computer in electrocardiography. In Clinical
Electrocardiography and Computers, edited by Cacerres CA, Dreifus LS.
New York, Academic Press, 1970, p 401
13. Wolf HK, Macinnis PJ, Stock S, Helppi RK, Rautaharju PM: Computer
analysis of rest and exercise electrocardiograms. Comput Biomed Res 5:
329, 1972
14. Van Bemmel JH, Hengeveld SJ: Clustering algorithm for QRS and ST-T
waveform typing. Comput Biomed Res 6: 442, 1973
15. Simoons ML, Boom HBK, Smallenburg E: On-line processing of
orthogonal exercise electrocardiograms. Comput Biomed Res 8: 105,
1975
16. Blomqvist CG: The Frank lead exercise electrocardiogram; quantitative
study based on averaging technic and digital computer analysis. Acta
Med Scand 178 (suppl 440): 1965
17. McHenry PL, Stowe DE, Lancaster MC: Computer quantitation of the
ST-segment response during maximal treadmill exercise; clinical correlation. Circulation 38: 691, 1968
18. Sheffield LT, Holt JH, Lester FM, Conroy DV, Reeves TJ: On-line
analysis of the exercise electrocardiogram. Circulation 40: 935, 1969
19. Hornsten TR, Bruce RA: Computed ST forces of Frank and bipolar exercise electrocardiograms. Am Heart J 78: 346, 1969
20. Ascoop CA, Distelbrink CA, De Lang P, Van Bemmel JH: Quantitative
comparison of exercise vectorcardiograms and findings at selective coronary arteriography. J Electrocardiol 7: 9, 1974
21. Blackburn H, Keys A, Simonson E, Rautaharju P, Punsar S: The electrocardiogram in population studies: A classification system. Circulation
21: 1160, 1960
22. Simoons ML, Hugenholtz PG: Gradual changes of ECG waveform during and after exercise in normal subjects. Circulation 52: 563, 1975
23. Afifi AA, Azen SP: Statistical Analysis: A Computer Oriented Approach. New York and London, Academic Press, 1972
24. Simoons ML: Classification of exercise electrocardiograms in patients
with a normal ECG at rest. Comput Biomed Res 1977 (in press)
25. Simoons ML, Van Den Brand M, Hugenholtz PG: Quantitative analysis
of exercise electrocardiograms and left ventricular angiograms in patients
with abnormal QRS complexes at rest. Circulation 55: 55, 1977
26. Simoons ML, Ascoop CA, Block P. Distelbrink CA, DeLang P, Vinke
ISCHEMIC ST DISPLACEMENT/Vincent, Abildskov, Burgess
R: Computer processing of exercise electrocardiograms. In Trends in
Computer-Processed ECGs, edited by Van Bemmel JH, Willems JC.
Amsterdam, North Holland Publ Cie, 1977, pp 383-406
27. Froelicher Jr VF, Yanowitz FG, Thompson AJ, Lancaster MC: The correlation of coronary angiography and the electrocardiographic response
to maximal treadmill testing in 76 asymptomatic men. Circulation 48:
597, 1973
28. Erikssen J, Enge I, Furfang K, Storstein 0: False positive diagnostic tests
and coronary arteriographic findings in 105 presumably healthy males.
Circulation 54: 371, 1976
29. Cornfield J: Statistical classification methods. In Computer Diagnosis
559
Methods, edited by Jacques J. Springfield, Ill, CC Thomas, 1972
30. Pipberger HV, Schneiderman MA, Klingeman JD: The love-at-first-sight
effect in research. Circulation 38: 822, 1968
31. Redwood DR, Borer JS, Epstein SE: Whither the ST segment during exercise. Circulation 54: 703, 1976
32. Sheffield LT, Reeves TJ, Blackburn H, Ellestad MH, Froelicher VF,
Roitman D, Kansal S: The exercise test in perspective. Circulation 55:
681, 1977
33. McHenry P: The actual prevalence of false positive ST segment
responses to exercise in clinically normal subjects remains undefined. Circulation 55: 683, 1977
Mechanisms of Ischemic ST-Segment Displacement
Evaluation by Direct Current Recordings
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
G. MICHAEL VINCENT, M.D., J. A. ABILDSKOV, M.D.,
AND MARY Jo BURGESS, M.D.
SUMMARY The electrophysiologic basis of ischemic ST-segment
displacement was investigated in 40 open chest dogs. Epicardial and
subendocardial electrograms were recorded with direct current
coupled amplifiers during partial and complete coronary artery occlusion. The time course and magnitude of DC potential changes, and
the effects on the DC potentials of heart rate and subendocardial
ischemia were investigated.
TQ segment depression, representing loss of resting membrane
potential, was found to be the consistent and most specific mechanism
of "ST displacement" due to ischemia. True ST-segment displacement, due to alterations of phase 2 of the transmembrane action
potential, occurred less frequently, and was not specific for ischemia.
The DC potential changes were similar in both subendocardial and
epicardial tissue. Increased heart rate by atrial pacing increased the
magnitude of TQ segment depression and produced subendocardial
ischemia during partial coronary flow reduction.
The findings have importance in the clinical interpretation of STsegment displacement. Potential limitations in the use of precordial
ST-segment mapping to estimate infarct size or severity are
described. The results help explain the variable and conflicting findings of previous studies investigating the mechanisms of ST displacement, and may help explain the poor correlation between rapid atrial
pacing-induced ST displacement and coronary artery disease defined
by angiography. Electrocardiographic interpretation based on
physiologic data such as those presented in this study should improve
the usefulness and accuracy of the electrocardiographic examination.
ST-SEGMENT DISPLACEMENT, one of the most useful
signs of acute ischemic heart disease, has at least two major
possible physiologic mechanisms: localized loss of resting
membrane potential and alteration of the transmembrane
action potential waveform or time of onset of the action
potentials.' Action potential waveform changes resulting in
ST displacement include alterations in the duration,
amplitude, and slope of phase 2. The time of onset of the action potential may be altered by conduction abnormalities
which result in delayed activation and therefore delayed
repolarization. Previous studies investigating the relative
frequency anid magnitude of these two mechanisms have
produced conflicting results."'' These variable results, plus
the increased interest in the use of ST-segment displacement
brought about by precordial ST-segment mapping, and body
surface isopotential mapping, indicate that further elucidation of the underlying physiologic mechanisms is needed.
In the standard electrocardiogram, using capacitor
coupled (A-C) amplifiers, both TQ segment and true STsegment displacement appear as "ST-segment displacement," and the two cannot be differentiated. The technique of recording cardiac potentials using direct current
coupled amplifiers does, however, allow the identification of
these two mechanisms individually. Using this technique,
loss of resting membrane potential is manifest by a depression of the TQ segment (baseline), while action potential
waveform or timing changes are manifest by a shift of the
true ST segment.' 6 These two separate mechanisms are illustrated in figure 1. Previous studies using this technique
have reported variable results, some indicating only TQ segment depression,9- " and others reporting both mechanisms
to be present, but with variable relative importance.` This
study describes the time course and magnitude of changes in
direct current recorded epicardial and endocardial electrograms from ischemic cardiac tissue in the experimental
animal. Additional new information is reported on the effect
of heart rate on the DC recorded electrogram. Through the
use of coronary artery flow probes to document the
magnitude of coronary artery flow reduction, the effects of
partial flow reduction and subendocardial ischemia are
described.
Methods
The mechanism of ST displacement produced by com-
From the Cardiovascular Research and Training Institute and Cardiology
Divisions, University of Utah School of Medicine and LDS Hospital, Salt
Lake City, Utah.
Supported in part by Program Project Grant HL 13480 from the National
Institutes of Health, Award 73-710 from the American Heart Association,
and a Research Award from the Utah Heart Association.
Address for reprints: G. Michael Vincent, M.D., Department of Medicine,
LDS Hospital, Salt Lake City, Utah 84143.
Received June 7, 1976; revision accepted May 27, 1977.
Estimation of the probability of exercise-induced ischemia by quantitative ECG analysis.
M L Simoons and P G Hugenholtz
Downloaded from http://circ.ahajournals.org/ by guest on July 28, 2017
Circulation. 1977;56:552-559
doi: 10.1161/01.CIR.56.4.552
Circulation is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231
Copyright © 1977 American Heart Association, Inc. All rights reserved.
Print ISSN: 0009-7322. Online ISSN: 1524-4539
The online version of this article, along with updated information and services, is located on
the World Wide Web at:
http://circ.ahajournals.org/content/56/4/552
Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally
published in Circulation can be obtained via RightsLink, a service of the Copyright Clearance Center, not the
Editorial Office. Once the online version of the published article for which permission is being requested is
located, click Request Permissions in the middle column of the Web page under Services. Further
information about this process is available in the Permissions and Rights Question and Answer document.
Reprints: Information about reprints can be found online at:
http://www.lww.com/reprints
Subscriptions: Information about subscribing to Circulation is online at:
http://circ.ahajournals.org//subscriptions/