European Heart Journal (1999) 20, 1459–1464
Article No. euhj.1998.1529, available online at http://www.idealibrary.com on
Epidemiological classification of acute myocardial
infarction: time for a change?
P. Porela*, H. Helenius†, K. Pulkki‡ and L.-M. Voipio-Pulkki*
*Departments of Medicine, †Biostatistics, ‡Clinical Chemistry, University of Turku, Turku, Finland
Aims The classification of an acute ischaemic cardiac event
is traditionally based on cardiac enzymes, electrocardiogra-
phy (ECG) and clinical symptoms. The impact of new
specific cardiac markers on the diagnostic classification of
suspected acute myocardial infarction remains poorly
studied. We therefore set out to compare the diagnostic and
prognostic information provided by the MONICA code
and a patient classification based on the maximal level of
creatine kinase MB isoenzyme. The significance of typical
pain and various ECG algorithms were separately analysed.
Methods and Results The study population consisted of
311 consecutive patients who were evaluated for suspected
acute myocardial infarction in a regional referral hospital.
Patients were retrospectively classified according to the
MONICA criteria, by a simplified code combining symp-
toms and creatine kinase MB, and solely using the maximal
creatine kinase MB concentration. Total mortality was
followed for 1 and 5 years. The creatine kinase MB based
classification was shown to be the strongest predictor of
Introduction
The classification of acute myocardial infarction is the
basis of all myocardial infarction studies. Nevertheless,
the criteria of acute myocardial infarction vary
from study to study. The WHO MONICA study
classification[1] uses conventional enzyme activities, the
Minnesota ECG code[2] and clinical symptoms (typical
pain >20 min). The Minnesota ECG code is complicated
and carries significant inter-observer variability[3]. The
new cardiac injury markers[4–6] have dramatically
changed the options available to rule in and rule out
acute myocardial infarction in the emergency room.
However, their use in the general epidemiological classi-
fication of suspected acute myocardial infarction
patients is poorly characterized.
Because the new markers also contain prognostic
information[7–9], we set out to study how the basis of the
Revision submitted 8 January 1999 and accepted 15 January 1999
Correspondence: Pekka Porela, MD, Department of Medicine,
University of Turku, FIN-20520 Turku, Finland.
0195-668X/99/201459+06 $18.00/0
mortality (OR=2·8–3·7, p<0·001) for outcome both at 1
and 5 years. Typical pain and a positive Minnesota ECG
had no prognostic relevance. However, an analysis algor-
ithm of the admission ECG was predictive of 1- and 5-year
survival.
Conclusions The epidemiological classification of sus-
pected acute myocardial infarction could be based solely on
a specific cardiac marker, such as creatine kinase MB mass.
This approach contains prognostic information and is
accurate enough for the structured diagnosis of acute
myocardial infarction. Other outcome predictors could be
used to identify patient subgroups and assess therapy.
(Eur Heart J 1999; 20: 1459–1464)
 1999 The European Society of Cardiology
Key Words: Myocardial, infarction, classification,
MONICA, creatine kinase MB.
See page 1446 for the Editorial comment on this article
classification could be changed to incorporate such
information. At the same time, it is essential to analyse
what happens to the number of cases in each category if
the basis of the classification changes, particularly in
centres that have participated in long-term epidemio-
logical studies. Because the ECG is used to select
patients for thrombolytic therapy, we specially investi-
gated the role of the admission ECG as a prognostic
marker.
Methods
Subjects and design
The study population consisted of 311 consecutive
patients (204 men, 107 women) aged mean (SD) 67 (13)
years who came to the only primary hospital in our
district for the evaluation of a suspected acute ischaemic
event. Patients with trauma and recent surgical proce-
dures were excluded. At arrival 45/311 of the total
population received thrombolysis. 97/311 (31%) patients
had a previously documented myocardial infarction. A
 1999 The European Society of Cardiology
1460 P. Porela et al.
Table 1 Criteria of the new combination code, designed
to mimic MONICA classification
CK-MB Typical symptom Atypical symptom
>40 ìg . l
1 Definite AMI Definite AMI
20–40 ìg . l
1 Definite AMI Possible AMI
<20 ìg . l
1 Possible AMI No AMI
12-lead ECG and a serum sample for markers of myo-
cardial injury were obtained immediately upon admis-
sion and then every 8 h for 48 h and thereafter daily for
3 days. Two patients were excluded because of techni-
cally incomplete data. Thus, the final study population
consisted of 309 patients. All available information was
used for the purpose of our study. Two hundred and
seventy-one (88%) of the patients were hospitalized for
at least one day. Mortality information was collected
from the national register and was available in 305 out
of 309 patients (99%) at 1 and 5 years follow-up. The
total in-hospital mortality was 14%. The total 1 year
mortality was 24% and the 5 year mortality 41%.
Patient classification
We studied three different classifications: the WHO
MONICA, a combination of clinical symptoms and
creatine kinase MB isoenzyme concentration data, and a
creatine kinase MB based classification. The cut-off
limits for the new combination code were interactively
chosen to reach maximal concordance with the
MONICA after the exclusion of ECG data (Table 1).
The creatine kinase MB classification was based on the
maximal value of creatine kinase MB during hospital-
ization regardless of the number of available samples.
The limits d5 ìg . l
1 and d10 ìg . l
1 were used as
cut-offs for possible and definite acute myocardial
infarction.
Cardiac injury markers
Serum creatine kinase isoenzyme MB mass fraction was
determined by Microparticle Enzyme Immunoassay
(MEIA) using the IMX CK-MB assay (Abbot Labora-
tories, Abbot Park, IL, U.S.A.). The upper reference
limit was 5 ìg . l
1. Coefficient of variation in our
laboratory was 4%.
Serum creatine kinase, lactate dehydrogenase and its
isoenzyme 1 activities were determined by using an
Hitachi 704 or 717 analyser (Hitachi Ltd, Tokyo, Japan)
as described previously[10]. The upper reference limit
used for CK was 201 IU . l
1, for lactate dehydrogenase
451 IU . l and for lactate dehydrogenase 1 131 IU . l
1.
These enzymes were defined as raised when more than
twice the upper reference limit was achieved, and as
borderline when the maximal value was elevated but
remained less than twice the upper reference range.
Eur Heart J, Vol. 20, issue 20, October 1999
ECG acquisition and analysis
The Marquette automated 12-lead system (Marquette
Electronics Inc., Milwaukee, WI, U.S.A.) was used
for data collection and storage. The data matrix of
Marquette Electronics was used for the durations and
amplitudes of the Q, R and S waves as well as for the
magnitudes of ST deviations. All available 12-lead
ECGs were manually coded according to the Minnesota
code in five categories: definite acute myocardial
infarction, possible acute myocardial infarction, non-
developing changes, no acute myocardial infarction and
uncodable ECG.
To categorize the admission ECGs, we first used the
simplified Selvester QRS score to detect already existing
Q wave infarctions[11]. The Selvester QRS score was
originally developed to assess myocardial injury size, so
that 1 point corresponds to a 3% loss of left ventricular
mass. When the admission Selvester QRS score was
<10, patients were classified based on the amount of ST
elevation (modified Aldrich score)[12]. The Aldrich score
was originally developed to predict the size of the injury
from the admission ECG. ST depressions (horizontal or
downsloping d1 mm) were recorded according to the
MPIP code[13]. The presence of left bundle branch block
or pacemaker rhythm was coded, as suggested by the
Marquette automated diagnosis. If none of the above
mentioned criteria were fulfilled, the classification of the
admission ECG was no diagnostic changes. The categor-
ization of the admission ECGs was fully automated
and computerized and was based on the commercially
available Marquette data matrix.
Statistical methods
Observer agreement of categorical assessment[14], a
method to assess the inter-observer variation of categ-
orical assessments, was used to study the concordance of
the different classifications. The level of the proportion
of agreement that signifies agreement is arbitrary[14].
Given the size of our population under consideration,
we chose 0·70 to indicate agreement between two codes.
The odds ratios (OR) and 95% confidence intervals
(95% CI) were calculated, using the SAS system for
Windows release 6.12/1996. P values less than 0·05 were
interpreted as statistically significant.
Results
Table 2 shows the distribution of patients into the
categories of definite, possible and no acute myocardial
infarction using the three different classifications. The
MONICA code classified almost half of the patients
presenting with suspected myocardial infarction as poss-
ible acute myocardial infarctions. After omitting manual
ECG Minnesota coding, we could, at best, achieve
the same diagnosis as that given by the MONICA in
 Classification of AMI 1461

Table 2 Number of diagnoses with different classifica- Prognostic significance of classifications

tions
We then studied mortality at 1 (Table 4A) and 5 years
Definite Possible No AMI (Table 4B). The odds ratio 1 was selected to represent
risk of death from all causes in the no acute myocardial
MONICA 99 135 75 infarction group. The creatine kinase MB classification
Combination code 110 114 85 had the strongest prognostic significance at both 1
CK-MB 144 35 130 (P<0·001) and 5 years (P<0·001, Table 4). All codes were
statistically associated with outcome, but most of this
CK-MB=creatine kinase MB. association was due to the fact that the possible acute
myocardial infarction group had good prognosis using
both the MONICA and the new combination code at 1
as well as 5 years (Tables 4A and B). In contrast, both
Table 3 The concordances between different classifica-
the definite and possible acute myocardial infarction
tions. The limit for statistically significant concordance
classes based on creatine kinase MB were independently
was 0·70
and statistically strongly associated with mortality at 1
MONICA Combination code CK-MB
year (OR=3·5–3·7) and at 5 years (OR=2·8–3·4).
To further explore the cause of the poor prognostic
significance of the MONICA classification, we studied
MONICA
definite — 0·73 0·64
separately the three basic variables (conventional en-
possible — 0·74 0·12 zyme activities, Minnesota ECG and symptoms) from
no AMI — 0·91 0·35 which the final MONICA classification is constructed.
Combination code Table 5 shows the variables that were statistically sig-
definite 0·73 — 0·76 nificantly associated with 1 and 5-year mortality. Odds
possible 0·74 — 0·14
ratio 1 was selected to signify association in the group
no AMI 0·91 — 0·36
CK-MB with normal findings. Of the three variables, only
definite 0·64 0·76 — enzymes (P=0·003 for 1 year and P<0·001 for 5 years)
possible 0·12 0·14 — and the Minnesota ECG code (P=0·005 for 1 year and
no AMI 0·35 0·36 — P<0·001 for 5 years) were associated with mortality.
However, only definitely raised (d2upper reference
CK-MB=creatine kinase MB. limit) enzymes and non-developing or uncodable ECGs
were predictive of 1 year mortality (Table 5). At 5 years,
borderline (elevated, but <2upper reference limit)
267/309 patients (86%). This required the use of rela- enzymes and possible acute myocardial infarction ECGs
tively high creatine kinase MB cut-off values (Table 1). also emerged as prognostically important variables.
An even more pronounced shift of cases from the Paradoxically, typical symptoms tended to be associated
possible to the definite and no acute myocardial infarc- with better prognosis, but this was of borderline
tion categories occurred when the maximal creatine statistical significance (P=0·085 and P=0·051).
kinase MB was used as the sole classification basis.
As shown in Table 2, with creatine kinase MB classi-
fication the number of patients with definite acute myo- Automated analysis of admission ECG
cardial infarction was the highest (144; 47%) and also
the number of no acute myocardial infarction cases (130; To investigate the possible prognostic role of the pre-
42%) was higher than with any other code. senting ECG patterns, we used an automated ECG
programme to classify the main findings in the admis-
sion ECG recordings (see Methods). Positive findings
were compared with the no diagnostic changes group,
Differences between classifications which was given an odds ratio of 1. Total mortality in
this group at 1 year was 13% and at 5 years 29%. As
Table 3 shows the concordance[14] among the three shown in Table 6, statistically significantly worse prog-
difference classifications. The concordance reached stat- nosis was associated with all other positive ECG find-
istically significant levels in the case of MONICA vs the ings except ST elevations (Table 6). Of the 59 patients
new combination code (0·73 for the definite, 0·74 for the presenting with computer-coded ST elevation, 25 (42%)
possible and 0·91 for the no acute myocardial infarction received thrombolysis at admission.
group) as the new code was generated to mimic
MONICA coding. The creatine kinase MB classification
did not reach statistically significant concordance with Discussion
any other classifications. In general, the greatest discrep-
ancies were present in the possible acute myocardial Our study demonstrates the fundamental differences in
infarction group. the behaviour of the MONICA code, developed for

Eur Heart J, Vol. 20, issue 20, October 1999

1462 P. Porela et al.

Table 4 Relative odds ratios for mortality at 1 year (panel A) and 5 years (panel B) using
the three different classifications. OR=odds ratio, CI=confidence interval. P-value <0·05
indicates that the classification as a whole was statistically significantly associated with
prognosis

P-value Died/n (%) OR 95% CI

A.
1 year
MONICA 0·033
Definite 31/97 (32) 1·3 0·7 to 2·5
Possible 22/133 (17) 0·6 0·3 to 1·1
No AMI 21/75 (28) 1
Combination code 0·002
Definite 35/107 (33) 1·2 0·7 to 2·3
Possible 15/113 (13) 0·4 0·2 to 0·8
No AMI 24/85 (28) 1
CK-MB class <0·001
Definite (d10 ìg . l
1) 46/140 (33) 3·5 1·9 to 6·5
Possible (d5 ìg . l
1) 12/35 (34) 3·7 1·6 to 8·9
No AMI 16/130 (12) 1
B.
5 years
MONICA 0·036
Definite 47/97 (48) 1·1 0·6 to 2·0
Possible 44/133 (33) 0·6 0·3 to 1·03
No AMI 34/75 (45) 1
Combination code 0·001
Definite 54/107 (50) 1·1 0·6 to 2·0
Possible 31/113 (27) 0·4 0·2 to 0·8
No AMI 40/85 (47) 1
CK-MB class <0·001
Definite (d10 ìg . l
1) 75/140 (54) 3·4 2·0 to 5·7
Possible (d5 ìg . l
1) 17/35 (49) 2·8 1·3 to 6·0
No AMI 33/130 (25) 1

CK-MB=creatine kinase MB.

epidemiological purposes, the tailored ECG scores and logical acute myocardial infarction classification. Bio-
specific cardiac markers. Our study was designed to chemical tissue damage was found to be the most
emphasize the prognostic significance of an epidemio- powerful in the detection of patients with adverse
outcome after an ischaemic event (Tables 4A and B).
Table 5 Significant predictors of 1 year (panel A) and 5 We chose the admission ECG to represent electrocar-
year (panel B) mortality among to the MONICA vari- diographic information because this data is consistently
ables. OR=relative odds ratio compared to the normal available and is used to guide therapeutic choices.
findings group, CI=confidence interval Regardless of the presence or absence of a classified
ischaemic event, the admission ECG score was found to
P-value Died/n (%) OR 95% CI be a strong predictor of long-term survival (Table 6).
Previous studies[7–9] have shown that with the use of
A.
the new specific cardiac injury markers the prognostic
Enzymes 0·003 classification of suspected acute myocardial infarction
Raised 29/102 (28) 2·1 1·1 to 4·1 patients can be enhanced, particularly in the unstable
Minnesota ECG 0·005 angina subgroups. Our approach was more epidemio-
Non-developing 34/99 (34) 6·1 1·7 to 21·3 logical in nature. In fact, we originally set out to study
Uncodable 11/30 (37) 6·8 1·7 to 28·6
Pain 0·085
the prognostic value of peak creatine kinase MB concen-
B. tration in the entire study population. The cut-off limit
Enzymes <0·001 of d5 ìg . l
1 and d10 ìg . l
1 bear no difference in
Raised 46/102 (45) 2·5 1·4 to 4·4 predictive power (Tables 4A and B), but are presented
Borderline 34/75 (45) 2·5 1·3 to 4·6 separately because 10 ìg . l
1 was the most accurate
Minnesota ECG <0·001
cut-off point for the MONICA definite acute myocardial
Possible 42/102 (41) 3·9 1·5 to 10·2
Non-developing 50/99 (51) 5·4 2·1 to 14·1 infarction group. The analysis of several clinical and
Uncodable 20/30 (67) 9·1 2·8 to 29·4 laboratory measures and their combinations (Tables
Pain 0·051 4–6) revealed no benefit over the simplistic approach
based on maximum creatine kinase MB alone. The

Eur Heart J, Vol. 20, issue 20, October 1999

 Classification of AMI 1463

Table 6 Prognostic significance of automated classifica-
tion of admission ECG, based on injury scores, for 1
(panel A) and 5 years (panel B) mortality. OR=relative
odds ratio compared with the no diagnostic ECG changes
group, CI=confidence interval
by new markers and maintain coding stability, but this
will probably not be possible in most centres. Therefore,
we suggest that the time has come to redefine the goals
and practices of epidemiological classification of acute
ischaemic syndromes.

P-value Died/n (%) OR 95% CI We wish to acknowledge The EVO Foundation of the Turku
University Central Hospital; Heart Foundation of South-
western Finland; Finnish Heart Foundation; Turku University
A. Foundation.
1 year <0·001
Q-wave 17/33 (52) 7·6 3·3 to 17·9
ST-elevation 13/59 (22) 1.9 0.9 to 4.2
ST-depression 15/48 (31) 3·2 1·4 to 6·9 References
LBBB or pacemaker 10/20 (50) 6·8 2·5 to 18·2
No diagnostic changes 19/145 (13) 1
[1] Gillum RF, Fortmann SP, Prineas RJ, Kottke TE. Inter-
B.
national diagnostic criteria for acute myocardial infarction
5 years <0·001
and acute stroke. Am Heart J 1984; 108: 150–8.
Q-wave 19/33 (58) 3·6 1·7 to 8·1
[2] Blackburn H, Keys A, Simonson E, Rautaharju PM, Punsar
ST-elevation 22/59 (37) 2·8 0·8 to 2·8
S. The electrocardiogram in population studies. A classifi-
ST-depression 26/48 (54) 3·1 1·6 to 6·1
cation system. Circulation 1960; 21: 1160–75.
LBBB or pacemaker 16/20 (80) 10·0 3·2 to 31·3
[3] Rautaharju PM, Warren JW, Jain U, Wolf HK, Nielsen CL.
No diagnostic changes 42/145 (29) 1
Cardiac infarction injury score: an electrocardiographic cod-
ing scheme for ischemic heart disease. Circulation 1981; 64:
249–56.
[4] Puleo P, Meyer D, Wathen C et al. Use of rapid assay of
validity of our approach is of course dependent on the subforms of creatine kinase MB to diagnose or rule out acute
myocardial infarction. N Engl J Med 1994; 331: 561–6.
ultimate purpose of the classification. We suggest that
[5] Hamm CW, Goldmann BU, Heeschen C, Kreymann G,
prognosis, represented by the 1 and 5 years mortalities in Berger J, Meinertz T. Emergency room triage of patients with
our study, should be the central goal of an epidemiologi- acute chest pain by means of rapid testing for cardiac troponin
cal classification of acute ischaemic syndromes. If this T or troponin I. N Engl J Med 1997; 337: 1648–53.
task is successfully fulfilled by biochemical data alone, [6] Winter R, Koster R, Sturk A, Sanders G. Value of myoglobin,
any further characterization of patient subpopulations Troponin T, and CK-MB mass in ruling out an acute myo-
cardial infarction in the emergency room. Circulation 1995;
could rely on other methods, such as ECG, which can be 92: 3401–7.
specifically adjusted to the particular needs in question. [7] Lindahl B, Venge P, Wallentin L and the FRISC study group.
In line with previous observations[15], the uncodable Relation between Troponin T and the risk of subsequent
ECG population also had the highest mortality in our cardiac events in unstable coronary artery disease. Circulation
study. On the other hand, previous studies have associ- 1996; 93: 1651–7.
[8] Hamm C, Ravkilde J, Gerhardt W et al. The prognostic value
ated specific ECG features with adverse outcome in the of serum troponin T in unstable angina. N Eng J Med 1992;
acute myocardial infarction population. The non-Q 327: 146–50.
wave acute myocardial infarction patients now have a [9] Ravkilde J, Nissen H, Horder M, Thygesen K. Independent
better short- as well as long-term prognosis than patients prognostic value of serum creatine kinase isoenzyme MB
who ultimately develop Q waves[16–22]. This evidently mass, cardiac troponin T and myosin light chain levels in
suspected acute myocardial infarction. J Am Coll Cardiol
means that the non-Q wave acute myocardial infarction 1995; 25: 574–81.
patients with initial ST elevation have a relatively good [10] Kairisto V, Hänninen K-P, Leino A et al. Generation of
prognosis after thrombolysis[16,17,21], but patients with reference values for cardiac enzymes from hospital admission
ST-depression or left bundle branch block do laboratory data. Eur J Clin Chem Clin Biochem 1994; 32:
worse[18,19,21]. Also the use and benefits of various 789–96.
treatments differ in the various ECG subgroups[22–24]. [11] Wagner G, Freye V, Palmeri S et al. Evaluation of a new QRS
scoring system for estimating myocardial infarct size. I.
Thus, incorporation of ECG data in patient characteri- specificity and observer agreement. Circulation 1982; 65:
zation algorithms appears justified, although we do not 342–7.
think that such information is needed to detect the [12] Aldrich H, Wagner N, Boswick J et al. Use of initial ST-
presence of myocardial injury per se. This could now be segment deviation for prediction of final electrocardiographic
based on biochemical information alone. The workload size of acute myocardial infarcts. Am J Cardiol 1988; 61:
749–53.
of ECG classification can be minimized by automating [13] Greenberg H, Gillespie J, Edward M. A new electrocardio-
the ECG coding in epidemiological studies[25–27]. How- graphic classification for post-myocardial infarction clinical
ever, there is no consensus of optimal interpretation trials. Am J Cardiol 1987; 59: 1057–63.
algorithms for that purpose[28]. [14] Grant JM. The fetal heart rate trace is normal isn’t it?
In this study we provide an example of how difficult it Observer agreement of categorical assessments. Lancet 1991;
337: 215–8.
can be to monitor long-term epidemiological trends in
[15] Lim LLY, Kinlay S, Fisher JD, Dobson AJ, Heller RF. Can
today’s clinical environment. In principle, there are two ECG changes predict the long-term outcome in patients
ways to go. By comparing various classification algor- admitted to hospital for suspected acute myocardial
ithms it could be possible to analyse the changes induced infarction?. Cardiology 1997; 88: 460–7.

Eur Heart J, Vol. 20, issue 20, October 1999

1464 P. Porela et al.
[16] Barbagelata A, Califf RM, Sgarbossa EB et al. Thrombolysis
and Q wave versus non-Q wave first acute myocardial infarc-
tion: A GUSTO-I substudy. J Am Coll Cardiol 1997; 29:
770–7.
[17] Goodman SG, Langer A, Ross AM et al. Non-Q-wave versus
Q-wave myocardial infarction after thrombolytic therapy.
Circulation 1998; 97: 444–50.
[18] Lee HS, Cross SJ, Rawles JM, Jennings KP. Patients
with suspected myocardial infarction who present with ST
depression. Lancet 1993; 342: 1204–7.
[19] Belotti G, Rochitte CE, de Albuquerque CP et al. Usefulness
of ST-segment depression in non-infarct-related electrocardio-
graphic leads in predicting prognosis after thrombolytic
therapy for acute myocardial infarction. Am J Cardiol 1997;
97: 1323–8.
[20] Cannon CP, McCabe CH, Stone PH et al. The electrocardio-
gram predicts one-year outcome of patients with unstable
angina and non-Q wave infarction: results of TIMI III registry
ECG ancillary study. J Am Coll Cardiol 1997; 30: 133–40.
[21] Langer A, Goodman SG, Topol EJ et al. Late assessment of
thrombolytic efficacy (LATE) study: prognosis in patients
with non-Q wave myocardial infarction. J Am Coll Cardiol
1996; 27: 1327–32.
[22] Aguirre FV, Younis LT, Chaitman BR et al. Early and 1-year
clinical outcome of patients’ evolving non-Q-wave versus
Eur Heart J, Vol. 20, issue 20, October 1999
Q-wave myocardial infarction after thrombolysis. Results
from TIMI II Study. Circulation 1995; 91: 2541–48.
[23] White HD, Van de Werf FJJ. Thrombolysis for acute myo-
cardial infarction. Circulation 1998; 97: 1632–46.
[24] Fibrinolytic Therapy Trialists’ (FTT) Colloborative group.
Indications for fibrinolytic therapy in suspected acute myocar-
dial infarction: collaborative overview of early mortality and
major morbidity results from all randomised trials of more
than 1000 patients. Lancet 1994; 343: 311–22.
[25] Rautaharju PM, MacInnis PJ, Warren JW, Wolf HK, Rykers
PM, Calhoun HP. Methodology of ECG interpretation in the
Dalhousie program; NOVACODE ECG classification proce-
dures for clinical trials and population health surveys. Method
Inform Med 1990; 29: 362–74.
[26] Michaelis J, Lippold R, Gluck E, Schindler; Scheidt E.
Automated serial ECG-analysis within an epidemiological
study. J Electrocardiol 1987; 20 (Suppl): 34–6.
[27] Porela P, Hänninen K-P, Vuorenmaa T et al. Computer-
assisted electrocardiography in structured diagnosis of acute
myocardial infarction. Scand Cardiovasc J 1999; 33: 89–96.
[28] Macfarlene PW. A brief history of computer-assisted electro-
cardiography. Methods Inf Med 1990; 29: 272–81.