An enhanced exact permutation rank-based inferential seamless phase 2/3 design

Seamless phase 2/3 adaptive designs integrate dose optimization and confirmatory analysis into a single trial, accelerating drug development. A phase 3 dose may be selected based on phase 2 short-term biomarkers results and carried forward in phase 3 study. However, type I error inflation due to dose selection and unknown correlations between short-term biomarkers and long-term efficacy endpoints poses significant challenges. Existing multiplicity adjustment methods, such as the Sidak adjustment and the rank-based Dunnett test, often remain conservative, leading to power loss. This paper proposes a permutation rank-based method that preserves the underlying correlation structure between biomarkers and efficacy endpoints without explicit estimation and enhances efficiency over existing approaches. Simulations demonstrate that the permutation rank-based method robustly controls family-wise error rates at the nominal level, while achieving uniformly higher power than the existing methods. A case study in oncology, using progression-free survival as the primary endpoint, further validates the method's ability to maintain higher power and control type I error.