Abstract:
|
Including historical data can increase the power of a current clinical trial and reduce sample size requirements. The power prior is a Bayesian approach for including historical data in the analysis, but with the likelihood of the historical data downweighted by a weight parameter, to account for differences between studies. The weight parameter can be estimated from the data, such as in the modified power prior (MPP), where greater similarity of historical and current data leads to more borrowing of historical data. In this talk an overview is given of recent developments of the power prior. It is shown how the MPP can be extended to the case of multiple historical studies, and how robust components can be added to protect against prior-data conflict. Simulation results are presented to compare the MPP with alternative methods in terms of frequentist characteristics. It is found that including historical data using the MPP can increase the power with almost nominal control of the type I error rate. It is also discussed what conditions the historical data should satisfy, and what the remaining barriers are to acceptance of the power prior in a regulatory setting.
|