Goldilocks Design (GD) utilizes predictive probability to adaptively select a trial’s sample size based on accumulating data. In order to control type I error at a desired level for a subset of the null space, extensive simulations at the study design stage are required to choose critical values, which is a challenge for this type of Bayesian adaptive design to be used for confirmatory trials. In this article, we propose a Modified Goldilocks Design (MGD) where type I error is analytically controlled over the entire null space. We do so by applying the conditional invariance principle and a combination test approach on $p$-values that are obtained from independent cohorts of subjects. Simulation studies show that despite analytic control of type I error rate, the proposed MGD has similar power when compared with the original GD. We further apply it to an example trial with time-to-event endpoint in oncology.

Goldilocks Design (GD) 利用预测概率根据累积数据自适应地选择试验的样本量。为了将零空间子集的 I 类错误控制在所需水平，需要在研究设计阶段进行大量模拟以选择临界值，这对于将此类贝叶斯自适应设计用于验证性试验是一个挑战. 在本文中，我们提出了一种改进的金发姑娘设计 (MGD)，其中 I 型错误在整个零空间上进行分析控制。我们通过应用条件不变性原理和组合测试方法来做到这一点$磷$-从独立的受试者队列中获得的值。仿真研究表明，尽管对 I 类错误率进行了分析控制，但与原始 GD 相比，所提出的 MGD 具有相似的功效。我们进一步将其应用于肿瘤学中具有时间到事件终点的示例试验。