Article
Utility-based optimization of one-stage basket trials
Search Medline for
Authors
-
Lukas Sauer
- Institute of Medical Biometry, University of Heidelberg, Heidelberg, Germany
-
Alexander Ritz
- Institute of Mathematics, TU Clausthal, Clausthal-Zellerfeld, Germany
-
Meinhard Kieser
- Institute of Medical Biometry, University of Heidelberg, Heidelberg, Germany
| Published: | September 6, 2024 |
|---|
Outline
Text
Introduction: In phase II trials of oncological antibody therapies, a common research setting is testing the same therapy in several patient cohorts, e.g. different tumor localizations with the same genetic trait targeted by the antibody. Such a research setting can be implemented using so-called basket trial designs. These are clinical trial designs that unite separate sub-cohorts within a single clinical trial – a concept that is not exclusive to oncological antibody research. Besides obvious advantages in terms of organization and trial authorization, basket trial designs come with statistical benefits as well. They enable “borrowing” information between substrata if their responses to the treatment are similar. This feature leverages small sample sizes in the single strata and aims to increase power for detecting responsive strata while keeping the resulting type-I error inflation moderate. In recent years, a multitude of Bayesian and frequentist techniques have been suggested for implementing borrowing in basket trial designs, see [1] for an overview. These designs may include several numerical parameters for fine-tuning up to which degree information is borrowed.
Methods: The optimal choice of these tuning parameters for basket trial designs is subject to current research: Ideally, it should offer a compromise between power maximization and type-I error control in different substrata across a range of possible response scenarios. As analytical constrained optimization using Lagrange multipliers is not feasible, we suggested an approach using numerical optimization algorithms to optimize utility functions. We apply this approach to a Bayesian basket trial design suggested by Fujikawa et al. [2]. In a simulation study, we investigate the performance of different algorithms and utility functions.
Results and conclusion: The simulation study’s results are currently still pending, but will be available at the time of GMDS 2024 conference. Preliminary results in pilot simulation showed good convergence. Our approach is not restricted to Fujikawa’s design and may hence inform the planning of other types of basket trials as well as further statistical research on basket trial designs.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Pohl M, Krisam J, Kieser M. Categories, components, and techniques in a modular construction of basket trials for application and further research. Biometrical Journal. 2021;63(6):1159–84.
- 2.
- Fujikawa K, Teramukai S, Yokota I, Daimon T. A Bayesian basket trial design that borrows information across strata based on the similarity between the posterior distributions of the response probability. Biometrical Journal.2020;62(2):330–8.
