General Multistage Gatekeeping Procedures for Identifying Beneficial Drug Combinations in Factorial Trials with Isotonic Gains

Abstract

Recently, Dmitrienko, Tamhane, and Wiens (2008) proposed a general multistage gatekeeping procedure (GMGP) as a novel procedure for testing multiple hypotheses grouped into ordered families. This article discusses an application of the GMGP to combination drug efficacy trial with the goal of identifying all effective, superior or simultaneously effective and superior combinations. A general framework to formulate and solve these problems is introduced. Under the assumption of isotonic gains the hypotheses are grouped into ordered families, where a family consists of all hypotheses corresponding to noncomparable combinations. The GMGP with the truncated Holm component is then used to test the hypotheses in a stepwise manner. The main advantage of this approach is the strong control of the overall error rate. Moreover, the GMGP can be applied to designs of relatively large dimensions. In general, the performance of the procedure depends not only on parameters but also on a prespecified value (values) of a truncation fraction. As simulations indicate, if GMGP uses a single truncation fraction, then the correct specification of the fraction is especially important if all combinations are beneficial: in these settings conventional Holm (1979) can dominate the GMGP in terms of power.