2026/07/27 by Paul Horton, Yajun Mei
Mathematics · Decision Sciences · #Statistical Methods in Clinical Trials #Advanced Causal Inference Techniques #Optimal Experimental Design Methods
paper · pdf · doi:10.1002/sim.70683
Clinical trials are inherently complex and become more challenging when selecting one treatment from many while also evaluating the efficacy. In this paper, we develop a pragmatic design that balances sample efficiency and practicality through a two-stage sequential approach. The first stage involves adaptive arm selection and early termination, while the second stage optimizes sample allocation. Inspired by the Kiefer-Weiss problem, the proposed method provides an efficient solution to an optimization framework that minimizes a weighted sum of the expected number of patients, probability of selecting a bad arm, and the probability of a global Type II error subject to constraints on the Family-Wise Error Rate and maximum sample size. The framework employs stochastic optimization to find the design parameters for the global minimum. We compare the performance against fixed and multi-arm multi-stage designs over a range of test conditions with varying error costs and treatment-response models. Our method improves the cost and practicality of a clinical trial for medical treatments or pharmaceutical development.