Inference of ROC Curves in the Presence of Verification Bias for Multiphase Studies

• Activity Number: 605
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Hua He*+ and Wan Tang
• Companies: University of Rochester and University of Rochester
• Keywords: diagnostic test ; verification bias ; ROC analysis ; multiphase ; AUC

Abstract

In diagnostic studies, it is common that only part of the subjects taking the test go further with gold standard test, and hence some of the subjects may have the true disease status unobserved. Simply ignoring those subjects without verified disease status may produce biased estimates in assessing the accuracy of a test (verification bias). There are many studies on the situations that subjects are directly selected for disease verification based on the test result and other patient characteristics which are always observed. In multiphase diagnostic studies, patients may be first selected for some further well behaved tests based on which gold standard test may be administered on selected patients for verification of exact disease status. In this paper, we develop methods to correct the verification bias in inference of ROC curves and the area under the ROC curves (AUC).