Some Recent Advances on Statistical Approaches for Planning Multi-regional Clinical Trials

Abstract

Many multi-regional clinical trials are faced with possible heterogeneity in treatment effect among regions and consequently the interpretation of the trial results is quite challenging. Regional heterogeneity can be caused by the differences in intrinsic factors, extrinsic factors or trial/data quality among regions. An apparent regional difference in treatment effect can be caused by a play of chance or sampling variability. In another aspect, regional heterogeneity may have substantial ramifications on sample size estimation for planning a multi-regional trial, as stipulated in Hung et al. (Pharm Stat 9:173–178, 2010), Lan and Pinheiro (Stat Biosci 14:235–244, 2012) and Quan et al. (Stat Med 33:2191–2205, 2014). Analyses of multi-regional trials are commonly based on a fixed-effect model assuming that the true treatment effects in all regions are equal in magnitude which may or may not be practical. Random-effect modelling has been considered as an alternative to deal with regional heterogeneity. In this research work we shall discuss the statistical implications of the statistical modelling approaches on the type I error probability and statistical power associated with testing the global treatment effect. We also develop formulae to assess the probability that some regions may falsely show an apparent negative trend by chance or sampling variability. As another alternative, a fixed-effect model that accounts for acceptable regional heterogeneity will be introduced.