Statistical Methods for PPV Confidence Interval Estimation for Clinical Bridging Studies When PPV Is Near or at 100%

• Activity Number: 237 - Statistical Challenges in Evaluating Molecular Diagnostic Assays Geared Toward Companion Diagnostics Devices
• Type: Topic-Contributed
• Date/Time: Wednesday, August 11, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Meijuan Li and Lei Yang* and Chang Xu and Cui Guo
• Companies: Foundation medicine and Foundation medicine and Foundation medicine and Foundation medicine
• Keywords: PPV; Confidence interval; Clinical bridging studies

Abstract

Positive predictive value (PPV) defined as the conditional probability of clinical trial assay (CTA) being positive given Companion diagnostic device (CDx) being positive is a key performance parameter for evaluating the clinical validity utility of a companion diagnostic test in clinical bridging studies. When bridging study patients are enrolled based on CTA assay results, Binomial-based confidence intervals (CI) may are not appropriate for PPV CI estimation. Bootstrap CIs which are not restricted by the Binomial assumption may be used for PPV CI estimation only when PPV is not 100%. Bootstrap CI is not valid when PPV is 100% and becomes a single value of [1, 1]. In this paper, we proposed a risk ratio-based method for constructing CI for PPV. By simulation we illustrated that the coverage probability of the proposed CI is close to the nominal value even when PPV is close to 100%. We demonstrated its application using a real case bridging study example. Moreover, because there is a lack of R package for PPV CI calculation, we will develop a publicly available R package to implement the proposed approach and some other existing methods.