The Use of Historical Information in Clinical Trials

• Activity Number: 208
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Scott M. Berry and Kert Viele*+
• Companies: Berry Consultants
• Keywords: Bayesian Analysis ; Hierarchical Models ; Historical Data ; Clinical Trials ; Dynamic Borrowing

Abstract

Historical data has been used in different ways in drug development. Usually it is all or nothing-- meaning a single arm trial may be used for comparison to historical rates, or randomization is done in the trial and no borrowing is used. Rarely is historical data explicitly modeled with the current clinical trial to create a combined analysis. Typically a clinical team or regulatory agency is concerned with the similarity of the historical data to the new trial -- and if they are confident in the similarity borrow full information, but if they are uncertain they borrow nothing. Hierarchical models allow for a flexible amount of borrowing -- dynamic borrowing -- in which the amount of historical information borrowed depends in the similarity of the observations. Thus the amount of borrowing depends on the empirically observed similarity. In this paper we compare different borrowing techniques, characterize the amount of borrowing, and the characterization of dynamic borrowing, and demonstrate the use of these techniques in clinical trials.