Diagnostic value of computerized exercise testing in men without previous myocardial infarction. A multivariate, compartmental and probabilistic approach.

Abstract

The value of exercise testing for the diagnosis of coronary artery disease is disputed but very few studies have taken advantage of all recent improvements, namely computer averaging of the ECG signals, multivariate analysis of the data, a compartmental diagnostic approach and probabilistic interpretation of the results. These methods were tested in a group of 387 men who had a computer-assisted multistage maximal exercise test; none had a history of myocardial infarction. In 284 symptomatic patients, the diagnosis was made by arteriography; 103 ostensibly healthy men were also included. The computer-averaged ECG signals (X, Y, Z) recorded at maximal exercise, maximal heart rate, blood pressure and workload, and the onset of angina pectoris during exercise were submitted to a multivariate stepwise discriminant analysis. The pretest likelihood for CAD was calculated from age and history; the post-test likelihood was calculated from Bayes' theorem and the average information content of several diagnostic methods was assessed in categorical and compartmental models. By multivariate analysis, 5 variables collected at maximal exercise were selected, namely the heart-rate, the ST60 segment level, the onset of angina during the test, the workload and the slope of the ST segment in lead X. The average information content of the analysis using 5 variables was 44% in a categorical model versus 55% in a compartmental model (P less than 0.001). For comparison, the information content of the analysis using the ST60 segment level alone was only 16% in the categorical model and 27% in the compartmental model. The clinical value of these diagnostic methods (categorical versus compartmental, univariate versus multivariate) was assessed by a probabilistic classification of the patients. The classification provided by the analysis of the ST60 segment changes was barely better than that one provided by the simple history. The probabilistic use of a multivariate and compartmental analysis of the data led to a significantly better and more accurate classification of the patients (83% of correct classification).