A Semiparametric Bayesian Approach to Borrow Information from Historical Control Data in Two Arm Clinical Trials

Historical information is always relevant for designing clinical trials. The incorporation of historical information in the new trial can be very benefitial. Some of these benefits include reduction of effective sample size, a significant increase in the statistical power, reduction of cost and ethical hazard. However, if current and historical trils conflict, borrowing information can give misleading results. In this project a semiparametric Bayesian method based on Dirichlet Process prior is introduced to borrow relevant information from historical control data. The scale parameter of the DP prior plays a crucial role by controlling the depndencies between the historical and current trials. The performances of the proposed method is further compared with other competing methods in simulated data sets