Algorithm to predict triple-vessel/left main coronary artery disease in patients without myocardial infarction. An international cross validation.

: Logistic regression was applied to the clinical, risk factor, and exercise data of consecutive angiographic referrals without prior myocardial infarction to determine an algorithm predicting the probability of triple-vessel/left main coronary artery disease. These data were obtained from a total of 1,074 such subjects from patient populations at four centers (Cleveland Clinic Foundation, Cleveland, Ohio; Hungarian Institute of Cardiology, Budapest, Hungary; the university hospitals, Zurich and Basel, Switzerland; and the Veterans Administration Medical Center, Long Beach, Calif.) and used to derive four separate probability algorithms. Each algorithm is based on patient data from study samples at three of the four centers and consists of 272 logistic functions, which are related to linear combinations of 13 variables (age, sex, type of chest pain, systolic blood pressure, resting electrocardiogram, serum cholesterol, fasting blood sugar, achieved exercise work load, achieved heart rate, exercise-induced angina and hypotension, heart rate-adjusted resting ST depression, and exercise ST slope). The four algorithms were cross validated by testing them on the populations not involved in their derivation. The resulting probabilities in the four test groups were then compared with the angiographic findings of triple-vessel/left main coronary artery disease. The discriminatory power of all the algorithms was fair to good (area under receiver operating characteristic curve, 0.68, 0.75, 0.82, 0.85) in the test groups. The algorithm did not significantly underestimate or overestimate disease probability except in one center (Long Beach).(ABSTRACT TRUNCATED AT 250 WORDS)