Incorporating Historical Data in Bayesian Phase I Trial Design: The Caucasian-to-Asian Toxicity Tolerability Problem.

Following phase I dose-finding oncology trials completed in Western countries, Asian investigators often conduct further phase I trials to determine the maximum tolerated dose for Asian patients. This may be due to concerns about possible differences in treatment tolerability between Caucasian and Asian patient groups. Our proposed approach aims to appropriately borrow strength from a previous Caucasian trial to improve the maximum tolerated dose determination in an Asian population of patients. We design an Asian phase I trial using the Bayesian continual reassessment method. First we analyze toxicity data from a Caucasian trial to derive the prior distributions for a subsequent Asian trial. Then, we calibrate the informativeness of the prior distributions according to prior effective sample size defined by Morita et al. Extensive simulation studies demonstrate favourable operating characteristics of the proposed method, compared with two methods based on power and noninformative priors, respectively.