Internally Adaptive Designs for Parallel Group Trials

Abstract

This paper discusses the necessity of developing designs for clinical studies that make possible an adaptation of the design parameters to findings at planned interim analyses. The adaptation should 1) provide robustness against wrong initial assumptions and 2) reduce the number of patients needed in general without sacrificing basic scientific and statistical principles. Such designs will be called internally adaptive. As an example, a two-stage design is proposed that, in contrast to usual two-stage group sequential designs, makes possible the adaptation of sample size for the final analysis to the observed treatment difference at the interim analysis. A three-stage design that combines this procedure with an interim estimate of variability proves to have a power that is robust against initial underestimation of variability and to reduce expected sample size substantially relative to an adaptive procedure proposed by Wittes and Brittain (1).