Borrowing Strength from Historical Control Data in Bayesian Trials

• Activity Number: 37
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xue Lin*+ and Boguang Zhen
• Companies: U.S. Food and Drug Administration and U.S. Food and Drug Administration
• Keywords: Bayesian design ; historical control

Abstract

In designing a Bayesian trial, information may be borrowed from historical control data to reduce the sample size in the current control arm, if the historical control data sets are determined to be exchangeable with the concurrent control arm. Hierarchical models are often used to control the amount of information borrowed, through the control of the study-wise variability. We discuss situations where multiple historical control data sets are available and show under different scenarios the frequentist's operational characteristics. We illustrate that the distribution of the historical control data sets relative to the concurrent controls may have huge impact on the study results and its interpretation. The traditional strength borrowing methodology may by augmented by taking this into account. We present some preliminary results in this direction. Results for simulated data will be presented.