Handling of missing data in long-term clinical trials: a case study

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Longitudinal Data Analysis

Background:

• Missing data is a pervasive challenge in clinical trials, often oversimplified by classical statistical methods.
• High discontinuation rates (nearly 50%) in long-term trials can compromise data integrity and analysis validity.

Purpose of the Study:

• To apply established missing data theory to analyze efficacy data from a long-term open-label trial.
• To assess the impact of various missing data handling techniques on treatment satisfaction estimates.
• To evaluate the robustness of original analyses through sensitivity analyses.

Main Methods:

• Re-examination of treatment satisfaction data using multiple imputation, selection models, and pattern-mixture models.
• Application of missing at random (MAR) and missing not at random (MNAR) assumptions for sensitivity analyses.
• Comparison of adjusted estimates with original paired t-test analysis that ignored missing data.

Main Results:

• Sensitivity analyses using MAR and MNAR techniques yielded adjusted effect size estimates for treatment satisfaction improvement over 12 months.
• Effect sizes remained statistically significant and clinically relevant across different missing data modeling approaches.
• The robustness of the original analysis was supported by consistent findings under various missing data assumptions.

Conclusions:

• Advanced missing data techniques (MAR/MNAR) are valuable for analyzing longitudinal data in trials with high dropout rates.
• Sensitivity analyses enhance the credibility of findings by exploring the impact of different missing data assumptions.
• The study demonstrates the utility of robust statistical methods for ensuring reliable efficacy conclusions in challenging clinical trial settings.