The purpose of this investigation was to document the relative importance of three clinical and three radionuclide variables for prediction of cardiac death in a consecutive group of patients evaluated for coronary artery disease. During a 6 1/2-year period, beginning in January 1978, 2,042 consecutive patients underwent radionuclide angiocardiography, with a clinical diagnosis of coronary artery disease, at Duke University Medical Center. A subgroup of 318 patients who underwent surgical myocardial revascularization near the time of initial study were excluded from later analysis. Clinical follow-up information was complete in a group of 1,663 patients who did not undergo interventional therapy. The 141 cardiac deaths in these 1,663 patients were the study end point. Cox proportional hazards models analyzed the prognostic information contained in three clinical variables (pain type, age, and sex) and three radionuclide angiocardiogram variables (exercise ejection fraction, resting end-diastolic volume, and change in heart rate with exercise). One-variable models confirmed the prognostic importance of each of these six variables. A multivariable model in which all six variables were used showed clinical variables to contain only 5% and the radionuclide variables 95% of the prognostic information. The exercise ejection fraction was the single most important variable, which alone contained 85% of the total information in the model. Curves relating probability of no cardiac death to the exercise ejection fraction identified a value of 0.50 as an inflection point. Patients with exercise ejection fractions below 0.50 demonstrate a probability of cardiac death that increases as the ejection fraction decreases.(ABSTRACT TRUNCATED AT 250 WORDS)