Sensitivity analysis for missing dichotomous outcome data in multi-visit randomized clinical trial with...

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Longitudinal Data Analysis

Background:

• Clinical trials frequently encounter missing data, particularly in multi-visit studies.
• Dichotomous endpoints, common in trials, can be affected by missing observations.
• Missing data may be 'missing not at random' (MNAR), biasing results if not handled.

Purpose of the Study:

• To present a closed-form method for sensitivity analysis in randomized clinical trials.
• To address challenges posed by missing not at random (MNAR) dichotomous data in multi-visit studies.
• To provide adjusted estimates and covariance matrices for robust treatment comparisons.

Main Methods:

• A closed-form method for sensitivity analysis is developed.
• Missing data counts are mathematically redistributed to favorable and unfavorable outcomes.
• Mantel-Haenszel adjustment and randomization-based adjustment are used for treatment comparisons.
• The methodology is extended to ordinal endpoints in an appendix.

Main Results:

• Provides adjusted proportion estimates for dichotomous endpoints.
• Offers closed-form covariance matrix estimates for sensitivity analysis.
• Enables robust treatment comparisons over time, accounting for potential informative missingness.
• Illustrated with a practical clinical trial example.

Conclusions:

• The proposed method offers a robust approach to sensitivity analysis for clinical trials with MNAR dichotomous data.
• This technique enhances the reliability of treatment effect estimates in the presence of missing observations.
• The methodology is applicable to multi-visit randomized trials and adaptable for ordinal endpoints.