Adaptive Designs for Interim Dose Selection

Abstract

We consider an adaptive design for selecting multiple winner doses at interim in a dose ranging study, evaluating several doses in comparison to placebo. At the end of Stage 1 (interim), we consider four types of decision rules for selecting winner doses to be further evaluated in Stage 2. The final comparison of winner doses with placebo is performed on the data collected from both stages. Four multiple comparison approaches were used for analyses: Dunnett’s step down, Hochberg, Holm, and Bonferroni. The operational characteristics of the decision rules were evaluated by simulation, including the impact on the Type I error rate and power by sample size ratio between two stages, and the relationship between the primary endpoint and the surrogate endpoint on which the interim decision is based. A guideline is provided to determine the Stage 1 sample size and to set a threshold for winner selection.