Successful pharmaceutical drug development requires finding correct doses that provide an optimum balance between efficacy and toxicity. Competing responses to dose such as efficacy and toxicity often will increase with dose, and it is important to identify a range of doses to provide an acceptable efficacy response (minimum effective dose) while not causing unacceptable intolerance or toxicity (maximum tolerated dose). How this should be done is not self-evident. Relating efficacy to dose conditionally on possible toxicity may be problematic because whether toxicity occurs will not be known when a dose for a patient needs to be chosen. Copula models provide an appealing approach for incorporating an efficacy-toxicity association when the functional forms of the efficacy and toxicity dose-response models are known but may be less appealing in practice when the functional forms of the dose-response models and the particular copula association model are unknown. This paper explores the use of the BMA-Mod Bayesian model averaging framework that accommodates efficacy and toxicity responses to provide a statistically valid, distributionally flexible, and operationally practical model-agnostic strategy for predicting efficacy and toxicity outcomes both in terms of expected responses and in terms of predictions for individual patients. The performance of the approach is evaluated via simulation when efficacy and toxicity outcomes are considered marginally, when they are associated via gaussian and Archimedean copulas, and when they are expressed in terms of clinically meaningful categories. In all cases, the BMA-Mod strategy identified consistent ranges of acceptable doses.

翻译：成功的药物研发需要找到能够在疗效与毒性之间达到最佳平衡的正确剂量。疗效和毒性等对剂量的竞争性反应通常会随剂量增加而增强，因此确定既能达到可接受疗效（最小有效剂量）又不引起不可耐受的毒性（最大耐受剂量）的剂量范围至关重要。如何实现这一目标并非显而易见。在可能发生毒性的条件下将疗效与剂量关联可能存在困难，因为当需要为患者选择剂量时，是否会发生毒性尚不可知。当疗效和毒性剂量反应模型的函数形式已知时，Copula模型为纳入疗效-毒性关联提供了有吸引力的方法；但在实际应用中，当剂量反应模型的函数形式及特定Copula关联模型未知时，该方法可能缺乏吸引力。本文探讨了BMA-Mod贝叶斯模型平均框架的应用，该框架整合了疗效和毒性反应，提供了一种统计有效、分布灵活且操作实用的模型无关策略，用于预测疗效和毒性结局——既包括期望反应值，也包括个体患者的预测值。通过模拟评估了该方法的性能，评估场景包括：疗效和毒性结局分别独立考虑、通过高斯Copula和阿基米德Copula建立关联、以及以临床有意义的类别形式表达时。在所有场景中，BMA-Mod策略均识别出了一致的可接受剂量范围。