A Dunnett–Bonferroni‐based parallel gatekeeping procedure for dose–response clinical trials with multiple endpoints

Abstract

This paper proposes a Dunnett-Bonferroni-based parallel gatekeeping procedure in a setting of dose-response clinical trials with multiple endpoints. It follows the Dunnett-based parallel gatekeeping strategy of Dmitrienko et al. (Pharm. Stat. 2006; 5:19-28), but differs in the calculation of the critical values. The implementation of the Dunnett-based parallel gatekeeping procedure relies on assumptions difficult to justify in typical clinical trials, namely (a) that the joint distribution of the test statistics from different endpoints can be approximated by a multivariate-t distribution and (b) that the true correlation between multiple endpoints can be well estimated using observed data. The proposed Dunnett-Bonferroni-based parallel gatekeeping procedure relaxes the preceding assumptions by splitting type I error rate among families using the Bonferroni inequality. While it is potentially less powerful than a Dunnett-based procedure when both procedures are applicable, the power loss is very minimal. Our proposed method avoids assumptions that might be challenged by regulatory agencies and does so with virtually no cost. Moreover, in most cases this method is easier to implement compared with the Dunnett-based procedure.