SC05: Revolutionizing Longitudinal Endpoint Analysis in Clinical Trials: Novel Non-Linear and Parametric Models in Frequentist and Bayesian Frameworks

Traditional analysis of longitudinal endpoints in clinical trials relies on mixed models for repeated measures (MMRM), with efficacy inference typically based on mean change from baseline at a pre-specified visit (e.g., the final study visit). This approach is well established, broadly accepted for regulatory submissions, and appropriate for many pivotal studies with pre-specified visit schedules and short-to-moderate study durations. At the same time, increasing data density, unscheduled assessments, and interest in leveraging valuable information from intermediate visits have motivated the use of modeling strategies that extend beyond the standard categorial-time MMRM framework. In this short course, we demonstrate the application of novel non-linear and parametric models that leverage data from multiple or all visits for efficacy inference while handling unscheduled assessments. Several models are specifically suited for analyzing high-density data from electronic apps and "Decision-Making in the Evolving Digital Era." Covered models include MMRM across all/multiple visits, proportional MMRM, cubic/B-spline mixed-effects models, proportional cubic/B-spline mixed-effects models, and hierarchical proportional/non-proportional cubic/B-spline mixed-effects models. Course objectives: (i) illustrate implementation in frequentist and Bayesian frameworks; (ii) highlight optimal covariate adjustment strategies in these models; (iii) introduce the "time savings in delaying disease progression" concept for interpreting treatment effects; (iv) reveal underlying connections among these models; and (v) provide regulatory perspectives on key topics. Participants will gain the knowledge and skills to independently apply these approaches post-course.