Combining Information Across Studies for Clinical Trials: What We Have Learned at FDA's Center for Devices and Radiological Health

• Activity Number: 552
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Laura Thompson*+
• Companies: FDA/CDRH
• Keywords: hierarchical model ; fda ; medical devices ; prior studies ; Bayesian

Abstract

I will discuss the Center's experience reviewing applications that propose to combine results across prior and current studies via a Bayesian hierarchical model. Some considerations that CDRH uses in the evaluation of whether a hierarchical model is appropriate include the assessment of exchangeability, calculation of the prior probability of the study claim, sensitivity of hyperparameters of the random effects distribution, and the decision on whether to borrow among treatment differences or among treatment means. I will also discuss model checking and sensitivity analysis. I will focus on both parametric and nonparametric distributions, and indicate where a nonparametric approach is appropriate. I will briefly discuss other examples where hierarchical models can and have been used in medical device studies, such as for multiple comparisons, diagnostic assessment, and meta-analysis.