Confidence intervals for correlated proportion differences from paired data in a two-arm randomised clinical trial

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Medical Data Analysis

Background:

• Disease progression monitoring often relies on composite outcome measures derived from paired domains.
• Scores within paired domains are frequently correlated, necessitating specialized analytical approaches.
• Existing methods may not adequately account for this correlation, potentially biasing treatment effect estimations.

Purpose of the Study:

• To develop and evaluate statistical methods for analyzing treatment effects in paired domains while accounting for score correlation.
• To compare the performance of different confidence interval construction methods for analyzing responder proportions in randomized trials.
• To apply these novel methods to a real-world clinical trial setting for scleroderma.

Main Methods:

• Utilized profile likelihood and asymptotic score methods for correlated data analysis.
• Developed and compared three simple asymptotic methods for confidence interval construction.
• Evaluated estimator performance based on coverage probabilities, non-coverage rates, and confidence widths.
• Applied methods to a multi-center, two-arm randomized clinical trial.

Main Results:

• The study presents a comparative analysis of various statistical methods for handling correlated paired domain data.
• Performance metrics such as coverage probability and confidence width were systematically assessed for each method.
• The developed methods were successfully applied to analyze a clinical trial in scleroderma patients.

Conclusions:

• Accounting for correlation in paired domains is crucial for accurate treatment effect analysis in disease progression studies.
• The proposed statistical methods offer improved precision and reliability for analyzing clinical trial data.
• These advancements can enhance the interpretation of treatment efficacy in various medical conditions.