Bayesian analysis using Fourier transforms of thallium-201 scintiscans to predict the presence of coronary artery disease.

Bayes' theorem of conditional probability was applied to the diagnosis of coronary artery disease (CAD) using thallium-201 scintigraphy as the testing procedure. Thallium-201 scintiscans were evaluated with a discriminant function previously developed using the amplitude coefficients of the Fourier transforms of the scans. The technique was applied prospectively to a population of 100 patients undergoing diagnostic coronary arteriography and thallium-201 scintigraphy, including 83 patients with CAD (70% or greater stenosis of luminal diameter) and 17 control subjects. A pretest probability of CAD was determined for each patient from the patient's age, sex and anginal symptoms. The pretest probability was combined with the patient's discriminant score to determine a posttest probability for CAD. For patients with CAD, the mean posttest probability was 0.85. Moreover, 57 of 83 patients (69%) had posttest probabilities exceeding 90%, including 40 patients (48%) with posttest probabilities exceeding 99%. For control subjects, the mean posttest probability was 0.19, with 11 of 17 (65%) having a posttest probability of less than 10%. Overall, 68 subjects had a posttest probability either less than 10% or more than 90% of which 63 were correctly classified (93%). Using a 50% posttest probability as a cutoff for classification, the technique has an 89% sensitivity, an 82% specificity and an overall accuracy of 88%. Therefore, this method objectively distinguishes patients with CAD from control subjects and provides a measure of the certainty of diagnosis. In addition, the discriminant function avoids the problem of inter- and intraobserver variability in visually interpreting thallium-201 scans.