Nomenclature. Nomenclature and criteria for
diagnosis of ischemic heart disease. Circula-
tion 1979; 59:607-609] was used for the
diagnosis of infarction. Twelve-lead electro-
cardiograms were recorded on admission and
repeated every 24h for four days. The devel-
opment of Q waves with a duration of 0.04s
was considered diagnostic of an acute trans-
mural infarct; S-T and T wave changes with
evolution were taken to indicate a nontransmu-
ral infarction.
In this investigation, subjects were assigned to
the diagnostic classes according to a widely accepted
set of clinical and laboratory criteria. Although
these criteria do not qualify as a gold standard, as
there is still no universally acknowledged perfect
diagnostic method for the clinical diagnosis of
myocardial infarction (Lee and Goldman 1986), they
constitute a generally accurate and reproducible
diagnostic tool. However, the results from the very
laboratory studies under study are included among
the criteria used for reference classification. There-
fore, some degree of incorporation bias is certain to
affect the performance estimates. Still, the absence
of a gold standard method of classification necessi-
tates the use of this reference method. It must be
kept in mind, however, that the findings of the
evaluation are biased toward overestimating the
diagnostic performance of the enzyme markers.
Study population
The third component of a performance evalua-
tion report is a definition of the clinical setting in
which the study is to be used. This means that the
report must indicate the relevant medical history,
signs and symptoms, and preliminary clinical and
laboratory study results in patients in whom the
study under review is to be used. The definition,
while specific, should be general enough to include
the range of patient presentations actually seen in
practice. The clinical setting for use of the labora-
tory studies evaluated by Werner
et al.
was:
Our study was designed to arrive at a protocol
for the use of enzymes in the diagnosis of
myocardial infarct ... under actual clinical
circumstances ...
The "actual clinical circumstances" referred to
can be deduced from the make-up of the study
population; it consisted entirely of patients admitted
to a Cardiac Care Unit. The setting, therefore, is
inpatient evaluation of acute myocardial infarction.
The findings of the evaluation may not apply to the
use of the diagnostic studies in different clinical
settings such as work-up of acute chest pain in the
emergency room (Lee and Goldman 1986). The
extent to which they apply to the diagnosis of
perioperative myocardial infarction depends upon the
representation of such patients among the study
population. The wording of the reports suggests that
none of the subjects were perioperative. The clinical
setting is further defined by the authors in terms of
the interval between the occurrence of the infarct
and the time of admission and diagnostic testing of
the patients:
Patients for whom the aggregate data
suggested that infarction occurred within 24
hour before hospitalization were classified as
"early admission," patients in whom infarction
occurred within 24-48 hours before hospitali-
zation as "intermediate admissions," and
patients in whom infarction occurred within
48-72 hours before hospitalization as "late
admissions." Enzyme values were separated
into four groups: those obtained from
uninfarcted patients, and those obtained from
infarcted patient on the first, second, and third
day after infarction.
After defining the clinical setting, the subjects
who participated in the evaluation are described.
Obviously, the study subjects should be a sample of
individuals from the stipulated clinical setting. This
condition is sometimes not met, though. For
instance, patients with completely unrelated disor-
ders or even healthy persons may be used to deter-
mine the specificity of a diagnostic study for which
the sensitivity has been ascertained in a population
limited to patients known to have the diagnosis, such
as patients attending a specialty clinic. In such an
evaluation, specificity will be overestimated. The
subjects should also constitute a representative
sample. Thus the spectrum of clinical variability
within the study population should be as broad as
possible (Ransohoff and Feinstein 1978). Patients
with mild and early forms of a disorder or those
falling into more favorable prognostic classes should
be included in the evaluation, as should patients with
unusual clinical features. The population should not
Evaluating Classification Studies
4-3
