Adaptive trial design and interim decision-making using incomplete longitudinal measurements: Methods and application to myasthenia gravis

Abstract

Sample size re-estimation designs using a promising zone framework are widely used adaptive trial methodologies that guide study continuation or modification during interim analyses. Conventional implementations often base interim calculations solely on participants with available primary endpoints, overlooking predictive information from baseline and earlier visits. This underutilization can lead to inefficient interim decision-making. In this work, we adapt semi-parametric efficient estimators that leverage baseline and intermediate data for use within a promising zone sample size re-estimation design. By incorporating information from participants who have not yet reached their primary endpoint, these estimators enable more precise interim estimators while maintaining strict Type I error control through the inverse normal combination function. Using data from the ADAPT study in generalized myasthenia gravis, we illustrate how these methods integrate into a promising zone sample size re-estimation framework. Simulations based on longitudinal profiles of anti-acetylcholine receptor antibody–seronegative participants demonstrate improved operating characteristics compared with the conventional approach, including increased overall power, especially for moderate effect sizes, without inflating the one-sided Type I error. Our findings highlight the practical benefit of applying existing semi-parametric estimators within promising zone sample size re-estimation designs, enabling more efficient and timely interim decision-making in settings with partially observed longitudinal data.