Improved interpretation of a continuous variable in diagnostic testing: probabilistic analysis of scintigraphic rest and exercise left ventricular ejection fractions for coronary disease detection.

The interpretation of a diagnostic modality requires consideration of the entire distribution of potential responses in both the normal and diseases population to extract the maximum available information from the test. Accordingly, we developed a probabilistic format for interpretation of scintigraphic left ventricular ejection fraction (LVEF) in the diagnosis of coronary artery disease (CAD). This format is distinguished (1) by accurate assessment of CAD likelihood based upon age, sex, and symptom classification using a logistical regression model (mean r = 0.9991, mean SEE = 0.008) and (2) by consideration of LVEF as a continuous probabilistic variable which obeys beta frequency distribution. This format allows assessment of the probability of CAD in an individual patient, according to Bayes' theorem of conditional probability of disease, given combinations of LVEF at rest and/or with exercise, and thereby is designed to provide more information than conventional categorical classification of "normal" or "abnormal" test observations.