Objective Clinical trials for Alzheimer's disease feature long follow-up periods and high dropout rates, with missing outcome data being commonplace, which can impair the accuracy of treatment effect estimation. This study aimed to compare the applicability and performance of several commonly used and regulatory-recommended missing data handling strategies, including the mixed-effects model for repeated measures (MMRM), standard multiple imputation (MI), reference-based imputation (RBI), and δ-adjusted multiple imputation, in Alzheimer’s disease clinical trials. Methods The data were derived from a multicenter, randomized, double-blind, parallel-group, placebo-controlled clinical trial for Alzheimer’s disease. The endpoint was the Alzheimer's disease assessment scale-cognitive (ADAS-Cog) score, and the change from baseline in ADAS-Cog score at Week 26 was the primary outcome, and the difference between the treatment and placebo groups was estimated. The primary analysis used MMRM under the missing at random (MAR) assumption. Sensitivity analyses were performed using standard MI, reference-based imputation (J2R, CR, CIR), and δ-adjusted multiple imputation. Effect estimates, standard errors, confidence intervals, and P-values were compared across methods. Results Treatment effect estimates were consistent in direction across all methods. Compared with MMRM and MI under the MAR assumption, RBI yielded more conservative estimates under the missing not at random (MNAR) assumption. Under conservative δ settings the conclusions remained robust (all P-values <0.001), indicating that the findings were stable against deviations from MAR to MNAR. Conclusion In this clinical trial dataset, treatment effect inferences show good consistency and robustness across multiple missing data handling methods and missing data mechanisms. The analytical workflow proposed in this paper can serve as a reference for missing data handling and sensitivity analysis in clinical trials for neurodegenerative diseases.
