Plenary Lecture

Receiver Operating Characteristic (ROC) Curve: A Tool for Describing and Comparing Continuous Diagnostic Tests

Professor Ivana Horova
Masaryk University
Dept. of Math. and Statist.
Jan/a?ckovo n/am. 2a
602 00 Brno
Czech Republic

Abstract: The ROC methodology has been developed in 1950’s. It is derived from signal detection theory where it is used to determine if an electronic receiver is able to satisfactory distinguish between signal and noise. Recently there has been an increased use of ROC curves for assessing the effectiveness of continuous diagnostic markers in distinguishing between healthy and diseased individuals. The most common parametric methods to estimate the ROC curve are based on bi-normal or bi-logistic model. But problems can occur if the distributional assumptions are not satisfied. Non-parametric methods do not have any distributional assumptions and are an ideal alternative for ROC curve analysis. In this lecture two different approaches to nonparametric estimates via kernel methods are presented. The first method is based on kernel estimates of cumulative distribution functions and the second one uses the fact that, in statistical terms, ROC curve is the non-null distribution function of the P-value. We conduct a simulation study to compare ROC curves obtained by proposed methods to data sets one of which fulfils the assumptions for bi-normal model. In conclusion, these methods are applied to medical data.