Meta-Analysis Using Dirichlet Process and Ordered Power Priors with Applications

• Activity Number: 484
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ram Tiwari*+ and Margaret Gamalo
• Companies: FDA and FDA/CDER/OB
• Keywords: power prior ; Dirichlet distribution

Abstract

Determining the non-inferiority margin (NI) for comparing an experimental treatment with an active comparator is based on carefully selected well-controlled historical clinical trials. With this approach, information on the effect of the active control from other sources including observational studies and early phase trials is usually ignored due to the need to maintain active control effect across trials. This leads to conservative estimates of the margin that translates to larger sample size requirements in the design and subsequent frequentist analysis, longer trial durations, and higher drug development costs. We provide Bayesian methodological approaches to determine NI margins that can utilize all relevant historical data through a novel Order Restricted Dirichlet Meta-analysis (ORDM) that puts ordered powers on the amount of information a set of data contributes in a completely data-driven way. The advantages of this method over more traditional frequentist are illustrated through several examples.