Computer-assisted diagnosis in the noninvasive evaluation of patients with suspected coronary artery disease

Abstract

A microcomputer program called CADENZA, which employs Bayes' theorem to analyze and report the results of various clinical descriptors and noninvasive tests relative to the diagnosis of coronary artery disease, was evaluated in 1,097 consecutive patients without previous myocardial infarction. With this program, each patient was characterized by a probability for coronary artery disease, based on Framingham risk factor analysis, symptom characterization, electrocardiographic stress testing, cardiokymography, cardiac fluoroscopy, thallium perfusion scintigraphy and technetium equilibrium-gated blood pool scintigraphy. A total of 11,808 probability estimates derived from various combinations of the available observations were analyzed: 2,180 in 170 patients undergoing coronary angiography and 9,628 in 969 patients who completed a 1 year follow-up for coronary events. The predicted probability of disease correlated linearly with observed angiographic prevalence in the 170 patients who subsequently had coronary angiography (prevalence = [0.001 +/- 0.011] + [0.966 +/- 0.019] X probability). The difference between probability and prevalence averaged 3.1%, and the magnitude of this correlation was not affected by the type or amount of data analyzed. The prevalence of multivessel disease in these patients increased as a monotonic function of disease probability. Below a probability of 25%, single vessel disease was slightly more common than multivessel disease. Above a probability of 75%, multivessel disease predominated. In the 969 patients followed up for 1 year from the date of testing, the incidence of cardiac death and nonfatal infarction increased as a cubic function of disease probability (from approximately 0 to 8% per year for each). Above a probability of 90%, however, the standard deviation for predicting these events was wide. These data indicate that Bayes' theorem in general--and CADENZA in particular--is an accurate, clinically applicable means for quantifying the prevalence of angiographic coronary artery disease, the risk of multivessel disease and the incidence of morbid coronary events in the year after testing.