Article
adoptr – an R-package for optimal adaptive two-stage designs
Search Medline for
Authors
-
Maximilian Pilz
- Universität Heidelberg, Heidelberg, Germany
-
Kevin Kunzmann
- University of Cambridge, Cambridge, United Kingdom
-
Carolin Herrmann
- Charité - Universitätsmedizin Berlin, Berlin, Germany
-
Geraldine Rauch
- Charité - Universitätsmedizin Berlin, Berlin, Germany
-
Meinhard Kieser
- Universität Heidelberg, Heidelberg, Germany
| Published: | September 6, 2019 |
|---|
Outline
Text
When planning a clinical trial, it is an interesting option to choose a flexible sample size rule. The advantages of such procedures, for example with respect to ethical and financial aspects, are well known. A valid choice of a flexible trial design is an adaptive two-stage design with one interim analysis.
Classical design choices for flexible clinical trials rely on combination tests while sample size recalculation rules are mainly based on conditional power considerations. However, it may be an interesting option to apply the two-stage design that is optimizing a specified objective criterion as, e.g., expected sample size under a valid set of constraints with respect to the type one error rate and power while imposing boundaries on the conditional power or continuation probabilities.
The R-package adoptr allows to compute such optimal designs for the common case of (approximately) normally distributed outcomes. The full design can be determined such that it is optimal where the objective criterion and possible constraints can be chosen freely by the user. In addition to the first-stage sample size and the critical values for early stopping, adoptr computes flexible sample sizes and critical values for the second stage in dependence on the stage-one outcome. While some common scoring criteria are pre-implemented, adoptr allows the user to define his or her own performance measures in an easy way and to apply them in order to obtain an optimal design for each specific use case. Furthermore, arbitrary prior distributions on the effect size such as classical point, truncated normal or uniform priors can be incorporated.
In our talk we present the architecture and the usage of adoptr and illustrate its application by a variety of clinical trial examples. It is shown how optimal designs can be computed in a user-friendly way and how the set of scores which are pre-implemented in adoptr can easily be extended.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Jennison C, Turnbull B. Adaptive sample size modification in clinical trials: start small then ask for more? Statistics in Medicine. 2015;34(29):3793-3810.
- 2.
- Kunzmann K, Kieser M. Optimal adaptive single-arm phase II trials under quantified uncertainty. Journal of Biopharmaceutical Statistics. 2019 Apr 25:1-15.
