Fitting ROC Curves When the Non-Disease Sample Is Contaminated

• Activity Number: 336
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Shiju Zhang*+
• Companies: St. Cloud State University
• Keywords: Empirical Likelihood ; Mixture ; Receiver Operating Characteristic Curve

Abstract

The accuracy of a diagnostic test can be assessed by estimating its receiver operating characteristic (ROC) curve based on disease and non-disease samples. While the disease sample is accertained, the non-disease sample can be contaminated with values from the disease population whose probability density function (pdf) is f(x). We propose an approach that treat the non-disease sample as one from a mixtrue distribution. Specifically, we assume that the disease sample is from f(x) and the non-disease sample is from cf(x)+(1-c)g(x), where g(x) is the pdf of the non-disease population and c measures the contamination rate. This biased sampling problem then can be handled using the empirical likelihood method. Simulation is used to show how different the wrong ROC curve can be from the true ROC curve.