A Note on Sequential Monitoring of Select-the-Winner Design

Abstract

Consider a trial comparing two treatments or doses A and B with a control C. Based on a unblinded interim look, a winner W between A and B will be chosen, and future patients will be randomized to W and C and compared at the end of a study. The naïve test statistic Z under this setting follows an approximate normal distribution, as shown by Lan et al. (2006) and Shun et al. (2008). Results of these two articles apply only to the fixed sample size design. With simple modifications, this manuscript extends the previous works to the group sequential setting.