Bivariate Test for testing the average areas under correlated receiver operating characteristic curves

Abstract

Methodology developed for comparing correlated ROC curves are mainly based on nonparametric methods. These nonparametric methods have several disadvantages. In this paper the authors propose an asymptotic bivariate test for comparing pairs of AUCs for independent data based on the Dorfman and Alf maximum likelihood approach. The properties of the test are examined by using simulation studies. The method is illustrated on an example of angiogram results from Sri Lanka. The test applied to the example found that there was a significant difference in the predictive power of three different cut-offs examined