Online Program

Evaluation of Diagnostic Tests for Low-Prevalence Diseases with Leveraging Real-World Data (302457)

*Wei-chen Chen, FDA/CDRH 
Heng Li, FDA/CDRH 
Nelson Lu, FDA/CDRH 
Changhong Song, FDA/CDRH 
Ram Tiwari, Bristol Myers Squibb 
Chenguang Wang, Johns Hopkins University 
Yunling Xu, FDA/CDRH 
Lilly Yue, US Food and Drug Administration 

Keywords: composite likelihood, power prior, in vitro diagnostics, sensitivity, specificity

Low prevalence diseases can pose challenges in diagnostic device evaluations. For example, the clinical validation of diagnostic tests for low prevalence diseases may require a large and lengthy clinical study through evaluating a large number of subjects to observe adequate number of positive cases. In this talk, we will discuss recent developments of propensity score-integrated approaches that help to accelerate such diagnostic clinical studies by leveraging the real-world data. The approaches include a Frequentist’s method based on the composite likelihood (PSCL) and a Bayesian method based on the power prior (PSPP) that allow down-weighting of the real-world data. We then introduce a statistical procedure based on propensity score matching and PSCL/PSPP to ensure that the real-world data being leveraged are similar to prospectively screened subjects in the diagnostic device evaluations. An illustrative example will be presented for the proposed procedure.