MMRM versus MI in dealing with missing data--a comparison based on 25 NDA data sets

Area of Science:

• Clinical Trials
• Biostatistics
• Drug Development

Background:

• Incomplete clinical trial data analysis is critical for drug development.
• Last-observation-carried-forward (LOCF) is a traditional but often suboptimal method.
• Multiple imputation (MI) and mixed-effects model repeated measures (MMRM) are advanced alternatives.

Purpose of the Study:

• To compare the performance of MI and MMRM in analyzing incomplete clinical trial data.
• To evaluate their robustness in controlling type I error rates and statistical power.
• To determine the optimal approach for hypothesis testing in drug efficacy studies.

Main Methods:

• Simulated incomplete data sets were analyzed.
• 25 New Drug Application (NDA) data sets from neuropsychiatric drug products were utilized.
• Comparative analysis focused on type I error rate and statistical power.

Main Results:

• The MMRM approach demonstrated superior performance compared to MI.
• MMRM effectively maintained the statistical properties of hypothesis testing.
• MI showed less favorable results in the analyzed datasets.

Conclusions:

• The MMRM approach is recommended over MI for analyzing ignorable missing data in clinical trials.
• MMRM offers better control over statistical properties for drug efficacy determination.
• This finding has significant implications for clinical trial data analysis and drug approval processes.