A Novel Linear B-spline Mixed Model for Repeated Measures (LB-MMRM) for Alzheimer's Disease Clinical Study Data

Mixed Models for Repeated Measures (MMRM) are widely used in neurological clinical trials, including Alzheimers disease studies, due to their robust statistical properties and regulatory acceptance. However, traditional MMRM treats time as a categorical variable, limiting its ability to incorporate unscheduled visits, harmonize trials with different visit schedules, or handle densely collected data from digital health technologies. To address these limitations, we propose a Linear B-Spline-based MMRM (LB-MMRM) model that uses time as a continuous variable while preserving compatibility with conventional MMRM when only scheduled visits are analyzed. LB-MMRM accommodates unscheduled visits by allocating observations to adjacent scheduled visits through spline-based weighting, thereby increasing effective sample size and statistical power. Simulation studies demonstrate that LB-MMRM maintains type I error control and improves power compared to traditional MMRM, particularly in scenarios involving irregular visit patterns or combined trial datasets. This approach offers a flexible and interpretable framework for modern clinical trials, supporting decentralized designs and integrated analyses without compromising regulatory standards.