Robust Borrowing from Historical Data with Meta-Analytic-Predictive Mixture Priors

• Activity Number: 591
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Beat Neuenschwander*+ and Sandro Gsteiger and Satrajit Roychoudhury and Heinz Schmidli
• Companies: Novartis Pharma AG and Novartis Pharma and Novartis Pharmaceuticals Corporation and Novartis Pharma
• Keywords: clinical trials ; meta-analysis ; meta-analytic-predictive (MAP) ; mixture priors

Abstract

Contextual evidence beyond the actual study data is increasingly used in clinical trials. The motivations are manifold: smaller trials, a decrease of cost and trial duration, and ethical consideration. However, in drug development the systematic use of historical data is still rather rare, which is mainly due to practical and methodological concerns. Major issues are the potential of prior-data conflict, and the ramifications of frequentist operating characteristics. We discuss these issues for the class of meta-analytic-predictive (MAP) priors, which arise naturally in a meta-analytic framework of the historical and actual trial data. We show that MAP priors 1) can conveniently be approximated by simple mixture priors, 2) provide good robustness and operating characteristics, and 3) are easy to implement and to interpret. A few examples from the many phase I and II clinical trials where we currently use the MAP approach will be discussed.