All Times EDT
Keywords: multi-regional clinical trial, outlier detection, influence diagnosis, leave-one-out cross-validation, bootstrap
Due to the globalization of drug development, multiregional clinical trials (MRCTs) have been increasingly adopted in clinical evaluations. In MRCTs, the primary objective is to demonstrate the efficacy of new drugs in all participating regions, but heterogeneity of various relevant factors across these regions can cause inconsistency of treatment effects. In particular, outlying regions with extreme profiles can influence the overall conclusions of these studies. In this article, we propose quantitative methods to detect these outlying regions and to assess their influences in MRCTs. The approaches are as follows: 1) a method using a dfbeta-like measure, a studentized residual obtained by a leave-one-out cross validation (LOOCV) scheme; 2) a model-based significance testing method using a mean-shifted model; 3) a method using a relative change measure for the variance estimate of the overall effect estimator; and 4) a method using a relative change measure for the heterogeneity variance estimate in a random-effects model. In addition, we propose to apply bootstrapping methods to assess the statistical variability of their influential measures. We illustrate the effectiveness of these proposed methods via applications to two MRCTs, the RECORD and RENAAL studies.