Methods for Covariate Adjustment in Combing Multiple Markers

• Activity Number: 561
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Soyoung Kim*+ and Ying Huang
• Companies: Fred Hutchinson Cancer Research Center and Fred Hutchinson Cancer Research Center
• Keywords: Area under the curve ; Biomarker combination ; Classification ; Covariate adjustment ; Receiver operating characteristic analysis

Abstract

In biomedical studies, receiver operating characteristic analysis (ROC) is widely used to find the optimal combination of biomarkers. The covariate-adjusted ROC curve has been proposed as a way to tease out the covariate effect when classification performance purely due to a marker is of interest. It is a common way to average covariate-specific ROC curves of the marker across different covariate level. However, no research has been done to find marker combinations that optimize this curve. In this article, we propose to maximize a nonparametric estimate of the area under the covariate-adjusted ROC curve, called covariate-adjusted AUC. We also evaluate the efficiency of covariate adjustment in finding marker combinations. Simulation studies are conducted to evaluate the efficiency of incorporating covariate adjustment by varying the association between covariate and markers and the association between covariates and the disease outcome.