|
Abstract:
|
Many clinical trials can be usefully accelerated by incorporation of information from historical controls, offering the potential for a shortened trial that exposes its participants to less risk. However, these advantages must be balanced against the need to maintain acceptable frequentist properties, such as Type I error and power. In this paper we investigate the use of power priors (Ibrahim and Chen, 2000) that provide a simple one-parameter framework for controlling the amount of borrowing of strength from past data. We investigate how the optimal amount of borrowing changes for different designs, as well as the sensitivity of our approach to hyperparameter specification when the trial sponsor seeks to borrow strength from only one historical study, where potential inter-study variability cannot be ascertained. We illustrate with data from a Mayo clinic colorectal cancer trial.
|
