Bayesian analysis of data from radionuclide angiocardiograms for diagnosis of coronary artery disease.

A continuous Bayesian model was developed by fitting a beta-function to the frequency distributions of resting ejection fraction, exercise ejection fraction, and change in end-systolic volume during exercise measured by radionuclide angiocardiography in a group of 249 men with coronary artery disease (CAD) and in a group of 56 men without disease. This model, then prospectively applied to 250 men with chest pain, did not increase the overall accuracy of the test but did increase the diagnostic content for individual patients. The diagnostic efficacy of the continuous Bayesian model was compared with previously determined optimal discrete criteria of a positive or negative test. Patients with CAD showed a maximum and mean increase in probability of disease of +0.58 and +0.11, respectively, by the continuous Bayesian model and +0.14 and +0.05, respectively, by discrete criteria. Men without significant disease showed a maximum and mean decrease in probability of disease of -0.73 and -0.38, respectively, by Bayesian analysis and -0.36 and -0.27, respectively, by optimal discrete criteria. Moreover, all 29 patients who died during a 35 month interval after study had a probability of CAD of 0.95 or greater by the continuous Bayesian model. These findings indicate that Bayesian analysis of radionuclide angiocardiographic test results with continuous distributions of left ventricular function measurement enhances the diagnostic and prognostic information for individual patients with symptoms suggestive of CAD.