Robust ROC Analysis Using Auxiliary Variables in the Presence of Missing Biomarker Values

• Activity Number: 396
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Xiaoxi Zhang*+ and Qi Long
• Companies: Pfizer Inc. and Emory University
• Keywords: Area Under the Curve ; Biomarker ; Missing at Random ; Not Missing at Random ; Receiver Operating Characteristic Curve ; Sensitivity Analysis

Abstract

The Area Under the Receive Operating Characteristics (ROC) curve (AUC) is a widely used summary measure in ROC analysis. The estimate of AUC in observational studies of diagnostic tests is often complicated by missing values of biomarkers, which potentially leads to bias. We develop robust statistical methods to estimate the AUC using auxiliary variables. In the case of missing at random (MAR), our estimators is consistent if one correctly specifies, conditional on auxiliary variables, either the model for the probability of a missing biomarker or the model for the biomarker values. In the case of not missing at random (NMAR), we propose a sensitivity analysis to assess the impact of assumptions on missing mechanism. The finite sample behavior of the proposed estimators is evaluated in simulation studies and they are further illustrated using data from a psychiatric study.