ORACLE II – Optimal Rules for Adaptive Designs with reCalculation of sampLE size
Final Report Abstract
Adaptive two-stage designs are an increasingly popular method for choosing the sample size of a clinical trial. With their possibility to adapt the sample size at a planned interim analysis, they allow more flexible sample sizes that fit to the specific needs of a particular trial. The DFG-funded project ORACLE aimed at developing tools for scoring adaptive two-stage designs and for computing designs that are optimizing defined scoring criteria. Within this project, a new performance score could be developed. This score does not only take the expectation of the sample size into account as classical recalculation rules but also considers its variability. Additionally, the expectation and the variability of the conditional power within the recalculation area are incorporated into the score. This new performance score allows a sensitive rating of adaptive two-stage designs and helps to choose a suitable adaptation rule in a clinical trial. The work on scoring adaptive designs resulted in a smoothing as well as resampling technique that makes adaptive designs more suitable in practice. Both approaches reduce the variability of the sample size and, therefore, make the adaptive design more acceptable for a practitioner since it is less uncertain how much sample size will be needed. Furthermore, especially the resampling addresses the randomness of the observed interim effect size. Mathematical and numerical techniques were used to compute optimal adaptive two-stage designs. For a given scoring criterion and a set of constraints as type I error rate, type II error rate, conditional power, etc., the adaptive two-stage design that is solving the resulting optimization problem can be derived. Due to their optimality, those designs are an exciting option for practical application. In particular, those optimal designs are entirely pre-specified before the trial starts and are thus less prone to arbitrary decisions. This makes them also recommendable from a regulatory point of view. In the freely available R-package adoptr, the optimization approach to adaptive two-stage designs is implemented. This package can be used to define and solve the user’s individual optimization problem and the resulting designs can be used for clinical trials. In summary, the project ORACLE achieved all project goals. A broad understatement of how to score an adaptive design could be developed and resulted in the deployment of a new performance score. The idea of optimizing a given scoring criterion under desirable constraints lead to the approach to determine an optimal adaptive design for a trial-specific optimization problem.
Publications
- (2019). A variational approach to optimal two-stage designs. Statistics in Medicine 38(21):4159–4171
Pilz M, Kunzmann, K, Herrmann C, Rauch G, Kieser M
(See online at https://doi.org/10.1002/sim.8291) - (2020). A new conditional performance score for the evaluation of adaptive group sequential designs with sample size recalculation. Statistics in Medicine 39(15):2067-2100
Herrmann C, Pilz M, Kieser M, Rauch G
(See online at https://doi.org/10.1002/sim.8534) - (2020). A note on the shape of sample size functions of optimal adaptive two-sage designs. Communications in Statistics – Theory and Methods
Pilz M, Kilian S, Kieser M
(See online at https://doi.org/10.1080/03610926.2020.1776875) - (2020). Comments on ’Adaptive sample size modification in clinical trials: start small then ask for more? Statistics in Medicine 39(1):97–98
Pilz M, Kieser M, Kunzmann K
(See online at https://doi.org/10.1002/sim.8427)
