Use of the multivariate approach to enhance the diagnostic accuracy of the treadmill stress test.

Twenty-one variables were analyzed in 142 male and 57 female individuals selected in a nonconsecutive fashion from a computerized list of patients with an angiogram and treadmill stress test. Patients were categorized as "normal" or "diseased" on the basis of angiographic results, significant coronary artery disease being defined as 70 percent or greater stenosis of one or more coronary arteries. Ten of the variables in males, and 14 of the 21 variables in females exhibited statistically significant differences between normal and diseased groups in univariate analysis of variance. Multivariate discriminant analysis, however, showed only three variable (duration, infarct by ECG, and ST depression in the immediate recovery period) to be significant in the final ranking for males; while five variables (infarct by history, presence of anginal pain during the test, ST resting changes in the normal individual, infarct by ECG, and age) were significant discriminators in females. For males, the sensitivity was 84%, the specificity was 80%, and the predictive value was 89%. For females, these values were 85%, 94% and 90%, respectively; while the entire group they were 85%, 86%, and 89%. In the classification matrices, males showed a misclassification percentage of 17% while females showed a 9% misclassification rate. Most of the misclassifications occurred with posterior probabilities ranging from 0.5 (50%) to 0.7 (70%); very few patients with high posterior probabilities were misclassified. It is concluded that multivariate discriminant analysis is a reliable means of determining the probability of coronary artery disease in a highly select group of patients; i.e. a large enough number of patients could be correctly classified with a high degree of certainty.