Nonparametric comparison of two tests of cardiac function on the same patient population using the entire ROC curve

Typically a diagnostic algorithm based on a cardiac-function test uses a threshold on a parameter to separate normal from diseased patients. This threshold may be determined through discriminant analysis or through a fixed choice of sensitivity or specificity. A more global approach is to plot sensitivities versus specificities as the threshold varies over its range. Such a plot is the receiver operating characteristic (ROC) curve. The area under the curve provides a nonparametric measure of the ability of a test to separate the two populations. A methodology is presented for comparing two tests on the same patient population using the differences in ROC areas. The method is applied to nuclear ventriculography parameters in the same patients, a normal group of 40 and a group of 24 with coronary disease and visible apical dyskinesis. The variables of interest are the fast filling fraction (FFF), ejection fraction (EF), and phase (PH). In terms of ROC area, there is evidence that FFF and PH differ, and some evidence that EF and PH differ, but none that FFF and EF differ.<<ETX>>