Common risk difference and corresponding confidence interval adjusted for multiple factors - comparison of seven...

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Modeling

Background:

• The U.S. Food and Drug Administration (FDA) highlights the need for covariate adjustment in clinical trial endpoint analysis.
• Traditional methods like Mantel-Haenszel (MH) weighting for dichotomous endpoints may not be optimally efficient.
• Existing Wald-type confidence intervals, including those from logistic regression, often exhibit undercoverage.

Purpose of the Study:

• To evaluate and propose improved methods for covariate adjustment in clinical trial analysis.
• To address the undercoverage issue prevalent in Wald-type confidence intervals.
• To compare the performance of various statistical methods through simulations.

Main Methods:

• Investigated Mantel-Haenszel (MH) weighting and minimum risk weighting strategies.
• Examined score-based methods for confidence intervals on differences in proportions.
• Developed and applied cell-wise adjustments and bias-reduction techniques for logistic regression models.
• Conducted extensive simulations to compare seven different statistical methods.

Main Results:

• MH weighting, while common, may not always provide optimal efficiency.
• Score-based methods demonstrated superiority over Wald-type methods in certain contexts.
• Proposed cell-wise adjustments and bias-reduction techniques significantly improved the coverage of Wald-type confidence intervals in logistic regression.

Conclusions:

• Cell-wise adjustments and bias-reduction techniques offer superior performance for logistic regression models.
• The proposed adjustments enhance the reliability and accuracy of confidence intervals in clinical trial analysis.
• These findings provide valuable insights for statistical analysis in clinical research, aligning with FDA recommendations.