Discriminant function analysis using thallium-201 scintiscans and exercise stress test variables to predict the presence and extent of coronary artery disease

Abstract

The ability to predict the presence and extent (number of affected vessels) of coronary artery disease objectively from an exercise treadmill test and thallium-201 myocardial perfusion scintiscans was evaluated using linear discriminant function analysis. Exercise and redistribution scans in the 30 ° left anterior oblique view were characterized by their two dimensional Fourier transforms. The analysis was performed in 141 persons, including 110 patients with coronary artery disease (70 percent or greater stenosis of luminal diameter) and 31 control subjects. There were 43 patients with single vessel and 67 patients with multivessel disease.Input to the discriminant analysis included age, sex, 18 variables from the exercise treadmill test and 36 Fourier frequency coefficients from each scan (exercise and redistribution). Two analyses were performed. In the first, a discriminant function was constructed to detect the presence of coronary artery disease. Seven input variables were chosen: maximal exercise pressure-rate product, sex, anginal pain, change in S-T segment slope with exercise, two Fourier coefficients from the exercise scan and one Fourier coefficient from the redistribution scan. The function correctly classified 103 of 110 patients with coronary artery disease and 27 of 31 control subjects; the estimated sensitivity of the technique is 94 percent, with 87 percent specificity and an overall accuracy of 92 percent for the detection of coronary artery disease. The discriminant function was significantly more sensitive and accurate than qualitative scan interpretation (p < 0.05).In the second analysis, two discriminant functions were developed to predict the extent of disease. In addition to the preceding variables, the duration of exercise, change in systolic blood pressure during the last two stages of exercise, and another Fourier coefficient from the exercise scan were chosen. Using two functions, the method detected the presence of coronary artery disease in 104 of 110 patients (95 percent sensitivity) and correctly classified 28 of 31 control subjects (90 percent specificity). Multivessel disease was correctly predicted in 53 of 67 patients (79 percent accuracy); another 11 patients with multivessel disease were predicted to have single vessel disease. Twenty-nine (67 percent) of 43 patients with single vessel disease were correctly classified; of the remaining 14 patients, 12 were predicted to have multivessel disease and 2 were judged to have normal vessels.The data demonstrate that numerical functions derived from the exercise treadmill test variables and Fourier coefficients of thallium-201 scans can be used to detect the presence of coronary artery disease with 92 percent accuracy. The predictive accuracy for the number of diseased vessels was 78 percent (110 of 141 patients classified correctly).