A “Backward” Bayesian Method for Determination of Criteria for Making Go/No-Go Decisions in the Early Phases

We introduce a “backward” Bayesian method to assist sponsors formulating early phase Go/No-Go criteria based on the ultimate efficacy or safety target, which is usually clearer for Phase 3. Derived from the definition of success for Phase 3, involving prior information and cost of later phases, this work presents the quantitative relationships among the following factors: previous and current study results, study designs (e.g., sample size, duration, or dose), true effect, target probability of success (PoS), expected financial loss, expected probability of terminating a potentially successful asset. An example is given to demonstrate how to accomplish these objectives for an exponential model describing the trajectory of weight loss. The expected loss and the probability of terminating a valuable compound are plotted against a range of criteria. The sponsors can then optimize the Go/No-Go criteria based on their tolerance for their objectives. This method can also be generalized to other nonlinear models. A byproduct of this work is to highlight the naivety of conventional gut feeling approaches in early internal decision-making process by explicitly identifying the necessary, albeit elaborate, information, and assumptions. Supplementary materials for this article are available online.

本研究提出一种“逆向贝叶斯（backward Bayesian）”方法，以辅助申办方基于最终疗效或安全性目标制定早期阶段的放行/止步决策标准——此类目标在Ⅲ期临床试验（Phase 3）中通常更为明确。本研究基于Ⅲ期临床试验的成功定义，结合先验信息与后续研发阶段的成本，阐明了以下要素间的定量关系：既往与当前研究结果、试验设计（如样本量、研究周期或给药剂量）、真实效应、目标成功概率（probability of success, PoS）、预期经济损失、终止潜在成功研发资产的预期概率。本研究以描述体重变化轨迹的指数模型为例，演示了如何达成上述目标；将预期损失与终止有价值候选化合物的预期概率，与一系列决策标准进行关联绘图。申办方可据此基于自身对研发目标的耐受度，优化放行/止步决策标准。该方法亦可推广至其他非线性模型。本研究的另一附带成果在于，通过明确识别虽复杂但必要的信息与假设，揭示了早期内部决策中传统经验直觉方法的局限性。本文的补充材料可在线获取。