B1170
Title: Summary outcomes in longitudinal clinical trials: Robustness to MNAR and Modeling Assumptions
Authors: Matt Shotwell - Vanderbilt University Medical Center (United States) [presenting]
Abstract: Longitudinal designs are common in randomized clinical trials, but longitudinal assessments are often summarized prior to statistical analysis (e.g., symptom-free days). An alternative longitudinal analysis may be more efficient but also more sensitive to uncertainties regarding missing data and data-generating mechanisms. The efficiency and robustness of the two approaches in estimating a common estimand (e.g., difference in mean symptom-free days) are compared under missing-not-at-random (MNAR) and alternative correlation structures. The context and motivation are two large platform studies in hospitalized patients and outpatients with COVID-19: ACTIV-4 Host Tissue and ACTIV 6. These findings may be useful in selecting an analysis approach for studies that prioritize robustness to such uncertainties.