Statistical Design and Considerations of a Phase 3 Basket Trial for Simultaneous Investigation of Multiple Tumor Types in One Study
Statistical Design and Considerations of a Phase 3 Basket Trial for Simultaneous Investigation of Multiple Tumor Types in One Study
复制标题
DOI：
10.1080/19466315.2016.1193044
复制
发表时间：
2016-01-01
影响因子：
1.8
通讯作者：
Beckman, Robert A.
中科院分区：
文献类型：
--
作者：
Chen, Cong;Li, Xiaoyun (Nicole);Beckman, Robert A.
常见癌症的多种分子亚型的发现引发了对生物标志物的研究，这些生物标志物可能预测多种组织学中实验性治疗的治疗效果。然而，组织学中推定的预测生物标志物的流行率通常较低，这使得在传统的基于组织学的验证性试验中招募足够数量的患者具有挑战性。另一种方法是在跨多个组织学的篮子试验中研究具有共同生物标志物特征的患者。这项研究设计之前曾被用来探索具有潜在变革作用的实验疗法。我们提出了广泛适用于任何有效疗法的 3 期篮子试验的总体设计概念。该试验的设计具有科学性和统计学上的严谨性，以便根据单项研究的结果批准多种肿瘤适应症的实验性治疗。考虑到适应症选择的难度，基本思路是中期分析时剔除不活跃的适应症，最终分析时汇集活跃的适应症。篮子设计的一个关键统计问题是修剪后汇总分析的 I 型错误控制。虽然剪枝可能被视为倾向于夸大 I 型错误的择优挑选，但它也与绑定无效性分析具有相似性，如果所有指示都被剪枝，则往往会缩小 I 型错误。修剪的净影响很复杂。使用不同的端点进行修剪和池化使问题进一步复杂化。本文将提供剪枝后三种样本量调整策略下一般篮子设计概念的第一类误差控制的统计细节。还提供功效和样本量计算。与未经修剪的简单设计进行比较。本文的补充材料可在线获取。
The discovery of numerous molecular subtypes of common cancers leads to the investigation of biomarkers potentially predictive of treatment effect of an experimental treatment in multiple histologies. However, the prevalence of a putative predictive biomarker within a histology is often low, which makes it challenging to enroll adequate number of patients in a conventional histology-based confirmatory trial. An alternative approach is to study patients with a common biomarker signature in a basket trial across multiple histologies. This study design has previously been used to explore experimental therapies with potentially transformative effects. We present a general design concept of a Phase 3 basket trial broadly applicable to any effective therapy. The trial is designed with scientific and statistical rigor to enable the approval of an experimental treatment in multiple tumor indications based on the outcome from a single study. Given the difficulty in indication selection, the basic idea is to prune the inactive indications at an interim analysis and pool the active indications in the final analysis. A critical statistical issue of the basket design is Type I error control for the pooled analysis after pruning. While pruning may be seen as cherry-picking which tends to inflate the Type I error, it also shares similarity with a binding futility analysis which tends to deflate the Type I error if all indications are pruned. The net impact of pruning is complicated. The use of different endpoints for pruning and pooling further complicates the issue. This paper will provide statistical details on Type I error control for the general basket design concept under three sample size adjustment strategies after pruning. Power and sample size calculations are also provided. Comparisons are made to a straightforward design without pruning. Supplementary materials for this article are available online.