ARTICLE ABSTRACT
Evaluating a novel treatment in a randomized controlled trial requires comparison against existing therapies. If several existing therapies of similar benefit exist, the identification of a single control regimen may be difficult. For this situation, we propose a reverse selection design which, in its simplest form, includes a single experimental treatment arm and two control arms. Rather than carrying both control arms through the entire trial, the control arms are compared at an early interim analysis, ideally while accrual is ongoing. At this time, the worst-performing control arm is dropped and randomization continues to the remaining arms. At the end of the study, we compare the treatment to the remaining control arm. When no head-to-head comparison of the extant therapies is available or feasible, this design requires a smaller sample size than a traditional three-arm design or two sequential trials in which the extant therapies are compared and the better treatment is used in a subsequent trial as the control arm. This is because the final comparison is only between two arms and because the early interim analysis occurs prior to the end of accrual—yet with enough information such that any substantially better control arm will be selected. We evaluate the operating characteristics of a reverse selection design via simulation and show that it reduces the required sample size needed to compare the treatment against the best control, controls type I error, and likely selects the right control arm to use in the final analysis.
