Evaluation of Digital Therapeutics: Tutorial on Adaptive Design Methods (Preprint)

Le résumé fourni par la source

UNSTRUCTURED Software-based digital therapeutics (DTx) are increasingly integrated into routine health care. To enable early access for patients to innovative interventions, there are options for early market access/provisional reimbursement of DTx within regulatory frameworks, such as the German Digital Health Applications (DiGA) fast-track pathway, implemented by the Federal Institute for Drugs and Medical Devices (BfArM). They highlight both the opportunities and challenges of generating robust evidence under conditions of early market access (provisional registration for reimbursement). In particular, the widespread separation of pilot studies (for provisional registration) and confirmatory studies (for permanent registration) in such frameworks can lead to duplicate data collection, inefficient resource use, and practical difficulties in meeting evidence requirements within constrained timelines. This tutorial provides an accessible introduction to adaptive clinical trial designs and demonstrates their application to the evaluation of DTx in registration studies such as the German fast-track pathway. The focus is on the conditional-error-function approach. We explain how this approach enables integrating the pilot and confirmatory stages into a single, coherent trial design while maintaining strict control of type I error despite possible design modifications at interim analyses. This includes sample size recalculations. When deciding to conduct a single two-stage adaptive study, the pilot-stage data can be used for provisional registration, and the data from both stages (pilot and confirmatory) can be used for permanent registration. In addition, the tutorial addresses common complexities in DTx evaluation, including multiple relevant endpoints and heterogeneity across patient subgroups. We introduce methods for handling multiplicity, such as Bonferroni adjustments and the closed testing principle. We illustrate how these principles can be incorporated into adaptive designs to support flexible endpoint or subgroup selection after some data have been collected, without compromising statistical validity. Practical examples, inspired by real-world registration studies in the DiGA context, demonstrate how adaptive designs can improve efficiency by reducing overall sample size requirements, shortening study timelines, and enabling a more informative use of pilot-stage data. Beyond the German setting, we discuss the broader applicability of adaptive designs to international DTx evaluation frameworks. Many of these frameworks face similar challenges in balancing timely access to innovation with the need for robust confirmatory evidence. By translating established statistical methodologies into a form accessible to digital health researchers, developers, and regulators, this tutorial aims to support the adoption of more efficient and flexible evaluation strategies. In conclusion, adaptive trial designs represent a powerful, statistically valid, and practical approach for improving the evaluation of DTx.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.