|
Abstract:
|
Investigators often conduct clinical trials for evaluating experimental interventions: treatment and control. The use of control can impose recruitment and retention challenges in addition to ethical and logistic challenges. A plausible remedy could be borrowing historical data to reduce the number of control subjects in CTs. However, borrowing strength from studies differing substantially from the concurrent trial can result in biased and inflated false discovery outcomes. The systematic use of dynamic borrowing can moderate borrowing based on the similarity between historical and concurrent studies and enhance the evaluation of comparative effectiveness of an intervention. Although plenty of literature discussed effective uses of Bayesian dynamic borrowing, most of them show use of aggregate-level data. Here, we focused on borrowing IPD and applied this to time-to-event endpoints. We conducted a comparative review and evaluation of IPD borrowing and the prior selection. We also offer simulations and examples of data analysis to compare different IPD borrowing approaches and the effect of priors using survival curve, bias, mse, fdr, power, and coverage probability.
|
