Sensitivity analyses in longitudinal clinical trials via distributional imputation.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Longitudinal Data Analysis

Background:

• Missing data is a common challenge in longitudinal clinical trials, impacting the reliability of study conclusions.
• Current methods often rely on the unverifiable 'missing at random' assumption, necessitating robust sensitivity analyses.
• Existing sensitivity analysis techniques, like multiple imputation, can be inefficient and yield suboptimal interval estimations.

Purpose of the Study:

• To propose a novel imputation method for sensitivity analysis in longitudinal clinical trials.
• To develop a statistically efficient and theoretically sound approach for handling missing data.
• To provide a consistent variance estimator for improved interval estimation in sensitivity analyses.

Main Methods:

• Introduced distributional imputation, which samples missing values from a target imputation model.
• Employed Monte Carlo integration principles to solve mean estimating equations for efficient estimation.
• Developed a weighted bootstrap method for consistent variance estimation, accounting for multiple sources of variability.

Main Results:

• The proposed distributional imputation method is fully efficient and possesses theoretical guarantees.
• The weighted bootstrap provides a consistent variance estimator, improving interval estimation accuracy.
• The framework demonstrated superior performance in simulation studies and a real-world antidepressant trial.

Conclusions:

• Distributional imputation offers a statistically superior alternative for sensitivity analyses in longitudinal clinical trials.
• This method enhances the robustness of study conclusions when faced with missing data and untestable assumptions.
• The proposed framework aligns with regulatory guidance (ICH E9(R1)) for handling missing data.