Simulation optimization for Bayesian multi-arm multi-stage clinical trial with binary endpoints

Multi-arm multi-stage designs, in which multiple active treatments are compared to a control and accumulated information from interim data are used to add or remove arms from the trial, may reduce development costs and shorten the drug development timeline. As such, this adaptive update is a natural complement to Bayesian methodology in which the prior clinical belief is sequentially updated using the observed probability of success. Simulation is often required for planning clinical trials to accommodate the complexity of the design and to optimize key design characteristics. This paper addresses two key limiting factors in simulations, namely the computational burden and the time needed to obtain results. We first introduce a generic process for simulating Bayesian multi-arm multi-stage designs with binary endpoints. Then, to address the computational burden and time, we optimize the method for calculating the posterior probability and posterior predictive probability of success.

多臂多阶段试验设计（Multi-arm multi-stage designs）指将多种活性治疗方案与对照组进行比较，并利用中期数据累积的信息增减试验组别的试验设计类型，该设计可降低研发成本、缩短药物研发周期。据此，这种适应性更新与贝叶斯方法（Bayesian methodology）天然适配：后者可通过观测到的成功概率，依次更新临床先验认知。为适配该设计的复杂性并优化关键设计特征，临床试验规划通常需要开展模拟研究。本文针对模拟研究中的两大核心瓶颈展开研究，即计算负荷与结果获取所需耗时。我们首先针对带有二分类终点（binary endpoints）的贝叶斯多臂多阶段试验设计，介绍一套通用的模拟流程。随后，为解决计算负荷与耗时问题，我们对计算后验概率（posterior probability）与后验预测成功概率（posterior predictive probability）的方法进行了优化。