Sequential Testing of the Empirical Positive and Negative Predictive Value Curves with Unknown Prevalence

• Activity Number: 684
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Joseph S. Koopmeiners*+ and Ziding Feng
• Companies: University of Minnesota and Fred Hutchinson Cancer Research Center
• Keywords: Biomarkers ; PPV and NPV Curves ; Group Sequential

Abstract

Validation of a novel diagnostic biomarker is an expensive and time consuming process. One approach to conserving time and resources is to utilize group sequential designs when evaluating the performance of a novel biomarker. Koopmeiners and Feng (2011) derive the asymptotic properties of the sequential empirical positive and negative predictive value curves under case-control sampling and illustrate how their results can be used to apply existing group sequential methods to the evaluation of a novel biomarker. A limitation to their approach is that the prevalence can't be estimated from a case-control study and must be assumed known. We extend their work to the case where the prevalence is estimated in the context of a cohort study and derive asymptotic results for the sequential empirical positive and negative predictive value curves under cohort sampling. These results allow us to apply existing group sequential methods to cohort studies that evaluate the performance of a novel biomarker. We evaluate the operating characteristics of our design by simulation and illustrate its application in the context of a study to evaluate the performance of a novel prostate cancer biomarker.