Abstract
The incorporation of historical data (HD) into a clinical trial analysis can improve the precision and efficiency of treatment evaluation if the HD are exchangeable with clinical trial data. Evaluating the exchangeability of these two data sets is challenging, however, as an incorrect assessment of exchangeability yields invalid inference on the treatment effect that may produce bias and inflate the Type I error rate. To address this practical problem, we propose an adaptive fused group bridge penalty to evaluate the comparability of parameters between HD and clinical trial data and make inferences on the treatment effect. The proposed penalty has oracle properties, including consistency for identifying the underlying model and the asymptotic normality of the estimators. Simulation studies show that the proposed method controls the Type I error rate better and has higher power than competing methods under both exchangeable and non-exchangeable settings. We apply the proposed method by reanalyzing a Phase III trial while also leveraging a corresponding HD set.