A Bayesian Adaptive Umbrella Trial Design with Robust Information Borrowing for Screening Multiple Combination Therapies

In immuno-oncology, developing combination therapies to overcome resistance to single agent or induce synergistic effects has become a new focus. To accelerate the screening process to identify promising combinations based on objective response rates, we propose a Bayesian adaptive Umbrella Trial design to simultaneously evaluate combinations of an investigational compound with different backbones, where information borrowing across combinations is allowed to increase trial efficiency. A robust borrowing approach is developed to strike a balance between borrowing and not borrowing by accounting for different configurations of homogeneity of treatment effects using Bayesian model averaging. Unlike existing methods that use the response rates to measure the degree of homogeneity by assuming all arms share a common control rate, an advantage of our approach is that it uses relative treatment effects to determine the degree of homogeneity by adjusting for different control effects across combinations. In the proposed design, Bayesian adaptive interim analyses are implemented to drop futile combinations and graduate early efficacious combinations. Simulation studies demonstrate that the proposed design with robust information borrowing outperforms some existing approaches. It improves power when treatment effects are homogeneous and maintains reasonable arm-wise Type I error rates when heterogeneity is present across combinations.

在肿瘤免疫治疗领域，开发可克服单药耐药或产生协同效应的联合疗法已成为新的研究热点。为加速基于客观缓解率筛选有前景的联合疗法，我们提出一种贝叶斯自适应伞式试验（Umbrella Trial）设计，可同时评估试验性化合物与不同基础治疗方案的联合方案，并允许跨联合方案进行信息借用以提升试验效率。我们开发了一种稳健信息借用方法，通过贝叶斯模型平均考量治疗效应同质性的不同配置情况，在信息借用与不借用之间实现平衡。与现有方法假设所有试验臂共享同一对照率、通过缓解率衡量同质性程度不同，本方法的优势在于，可通过调整不同联合方案间的对照效应，利用相对治疗效应来确定同质性程度。在所提出的试验设计中，通过贝叶斯自适应中期分析剔除无获益的联合方案，并对早期显示疗效的联合方案进行提前升级。模拟研究表明，采用稳健信息借用的所提设计优于部分现有方法：当治疗效应同质时，该设计可提升检验效能；而当联合方案间存在异质性时，可维持合理的试验臂一类错误率。