ROC CURVES AND THE AREAS UNDER THEM FOR DICHOTOMIZED TESTS - EMPIRICAL-FINDINGS FOR LOGISTICALLY AND NORMALLY DISTRIBUTED DIAGNOSTIC-TEST RESULTS

Many measures, including sensitivity and specificity, predictive value s, and likelihood ratios, are available for the assessment of diagnost ic tests. A drawback of the use of these measures is that continuous t est results are often dichotomized, with consequent loss of informatio n. Receiver operating characteristic (ROC) curves do not depend on dis crimination thresholds, and therefore the area under the ROC curve (AU C) is one of the preferred measures. Although quantitative test result s are often presented dichotomized, it would be convenient still to be able to estimate the ROC curve and the AUC. The authors present equat ions for such estimates when only one pair of a true- and a false-posi tive rate is given, for inherently logistically and normally distribut ed data. Illustrative empirical data are provided for both distributio ns. In contradiction to earlier reports, the authors also show that di fferential disease verification may skew the ROC curve. The ROC curve is thus not invariant to selection bias.