BOP2试验设计方法的先验敏感性分析研究

Prior Sensitivity Analysis in BOP2 Design

目的BOP2设计(贝叶斯最优II期设计)是在一个统一框架下可以处理复杂终点临床试验的设计方法,因其良好的统计性能、易于实践等优势,已在临床试验中得到广泛应用。和一般贝叶斯方法一样,模型未知参数的先验分布设置十分关键。本文旨在研究BOP2设计对先验选取的敏感性以及先验选择的一般规律。方法通过计算机模拟研究比较BOP2设计在不同无信息先验、乐观先验和保守先验下的统计表现。结果基于模拟结果,发现部分无信息先验以及保守先验,在不同场景下BOP2设计均有良好的统计性能,而乐观先验易引起一类错误率膨胀,仅当乐观先验与实际疗效相一致时,其统计性能良好。结论保守先验下BOP2设计的表现最稳健。若研究者对试验药物疗效持有相当积极乐观的态度,可以谨慎地选择乐观先验。

Objective Bayesian optimal phase II(BOP2) design, which is capable of handling simple(e.g.,binary) or even complicated(e.g.,ordinal, nested, and co-primary) endpoints under a unified framework, has been widely applied to clinical trials due to its good performance and flexibility.Since the setup of prior distribution for unknown parameters is critical within Bayesian approach, we herein aim to investigate the sensitivity of BOP2 design to prior selection, and to further summarize the general rules for selecting priors.Methods Simulation was performed to compare the statistical performance of BOP2 design using different priors.Results We found that both non-informative and conservative priors represented good statistical performance under different scenarios.Type I error tended to be inflated with optimistic prior, under which BOP2 however exhibited better performance(i.e.,more power and unlikely to incorrectly terminate the trial early) only if the therapy has favorable curative effect.Conclusion Conservative prior could be the most robust.If the researcher was quite optimistic about the efficacy of the trial drug, an optimistic prior would be recommended with caution.