STatistical Methods for OPtimal Basket Trial Designs fOR Precision Medicine – a General, Customizable TOolbox (STOP OR GO)

精准医学领域对新疗法的需求一直很高。新的发现表明，从药物开发开始就通过靶向异常基因或途径来整合分子图谱方面的知识。因此，迫切需要为早期临床阶段开发新的研究设计。一种相对较新的方法是所谓的“篮子”设计。篮子试验是基于根据分子图谱将几个子研究结合起来的想法。因此，各个子研究共享一个共同的设计。主端点通常是二进制的。其他设计选项可能会有所不同，例如每个亚研究的研究臂数。篮子试验旨在纳入临时修改，因此它们包括自适应设计元素。通过这样做，他们有机会提高药物开发的效率，并允许开发个性化的治疗方法。然而，篮子设计也伴随着重大的统计挑战，包括更困难的样本量规划、多个检验、中期分析、最佳无效性界限的选择、适当的通过/不进行标准的定义以及纳入不同研究对象之间的信息交换(“借用”)的规则。本建议的总体目标是发展最优篮子设计，其中包括独立或组合的适应性因素，使得·聚类/借用、·无效性在过渡性、·样本量重新计算中停止。为此，该提案分为三个工作包：1.在第一个工作包内，我们的目标是通过将已公布的篮子设计重新表述为有约束的优化问题来改进和扩展，这些问题应以分析的方式解决。该方法允许1)找到给定设计的最佳参数解，以及2)将不同的篮子设计相互比较。在第二个工作包中，我们的目标是开发新的适应性设计元素，这些元素通过全面的方法学考虑来证明是合理的，从而满足特定的性能要求。3.在最终的工作包中，我们将把开发的约束优化问题与新的适应性设计元素结合起来。随后，我们将在R cran上实现一个相应的经过验证的图形用户界面R包，它允许通过组合任意自适应设计元素来设计最优的可定制篮子试验。该项目的研究结果将有助于在精准医学中建立高效的新型篮子试验设计。开发的验证、免费、易用的软件解决方案为新设计在实践中的应用提供了支持。

The demand for new treatments in precision medicine remains constantly high. New findings suggest incorporating knowledge on molecular profiling by targeting aberrant genes or pathways right from the beginning of drug development. Therefore, the development of new study designs for early clinical phases is of urgent need. A relatively new approach are so-called “Basket” designs. Basket trials are based on the idea of combining several substudies according to molecular profiling. The individual substudies thereby share a common design. The primary endpoint is usually binary. Other design options might vary, for example the number of study arms per substudy. Basket trials are designed to incorporate interim modification so they include adaptive design elements. By this, they have the chance to increase the efficiency of drug development and to allow developing individualized treatment approaches. However, Basket designs also come along with major statistical challenges including a more difficult sample size planning, multiple testing, interim analyses, the choice of optimal futility boundaries, definition of adequate go/no go criteria and rules for incorporating information exchange (“borrowing”) between different substudies.The global objective of this proposal is to develop optimal Basket designs, which incorporate isolated or combined adaptive elements such that • Clustering/ borrowing, • Futility stopping at interim, • Sample size recalculation. To do so, the proposal divides into three work packages:1. Within the first work package, we aim to improve and extend by reformulating published Basket designs as optimization problems with constraints, which shall be solved analytically. This approach allows 1) to find the optimal parameter solution for a given design and 2) to compare different Basket designs against each other.2. Within the second work package, we aim to develop new adaptive design elements, which are justified by thorough methodologic considerations such that specific performance requirements are met. 3. In a final work package, we will combine the developed constrained optimization problems with the newly adaptive design elements. Subsequently, we will implement a corresponding validated R package with graphical user interface on R CRAN, which allows to design an optimal customizable Basket trial by combining arbitrary adaptive design elements. The results of this project will help to establish efficient novel Basket trials designs in precision medicine. Application of the new designs in practice is supported by the developed validated, freely-available and easy to use software solution.