The Role of Bayesian Statistics in Confirmatory Drug Development Trials

• Activity Number: 554
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): David Ohlssen*+
• Companies: Novartis Pharmaceuticals Corporation
• Keywords: Bayesian analysis ; missing data ; sensitivity analysis ; phase III trials ; historical data ; futility rules

Abstract

In phase III, due to regulatory requirements on type one error control, there is a perception that Bayesian statistics has a limited role. Here, by considering a number of examples, we question this view. Firstly, as the vast majority of phase III trials employ single imputations to handling missing data, it is unclear that the desired type I error control is achieved. To improve, a principled approach based on modeling and sensitivity analysis is required. As Bayesian approaches are well suited to this, they perhaps should become an important partner in phase III statistical analysis. Secondly, we consider an example where a Bayesian futility rule was applied in a phase III study. Finally, more controversial designs will be discussed, where historical data is used to form a prior for one arm. When the control treatment is well established, this type of design could make sense.