Beyond treatment selection: Interim hypothesis addition and deletion in confirmatory seamless designs

Inferentially seamless Phase 2/3 designs can accelerate drug development by combining learning and confirmatory objectives within a single protocol while preserving strict Type I control. Standard multiplicity frameworks for such trials typically assume that the complete set of efficacy hypotheses is fixed at trial initiation. In practice, however, emerging internal or external evidence can motivate the introduction of new confirmatory hypotheses at an interim analysis (IA), for example by adding a new dose that was not initially planned for confirmatory testing. This creates a tension between operational flexibility and rigorous Type I error control, a statistical and regulatory requirement. We extend the conditional error rate (CER) framework to permit hypothesis addition and deletion at an IA while maintaining strong familywise error rate (FWER) control under a closed testing principle. We also derive the conditional covariance structures and reference distributions needed when intersection hypotheses involve a newly introduced component. We compare our CER procedure to (i) an inverse-normal stagewise p-value combination approach and (ii) a straightforward split-α comparator that allocates separate one-sided error to “old” versus newly introduced hypotheses. We evaluate operating characteristics in a two-stage seamless setting with interim best-dose selection and an added dose at the IA. Simulations under the global null and under partial-null configurations support the strong FWER control of the implementation, while sensitivity analyses examine misspecification of the working reference distribution. Across alternative scenarios, the CER procedure can provide meaningful power gains, especially when the newly introduced dose is promising.