Assessing non-inferiority for binary matched-pairs data with missing values: a powerful and flexible GEE approach...

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Methodology

Background:

• Clinical studies frequently assess treatment non-inferiority using binary matched-pairs data.
• Traditional methods often assume complete data, limiting power when missingness occurs.
• Handling missing data is crucial for robust statistical power in these studies.

Purpose of the Study:

• To introduce a flexible generalized estimating equations (GEE) approach for non-inferiority tests with partially observed binary matched-pairs data.
• To evaluate the performance of the proposed GEE method against existing approaches using simulated data.
• To reassess sample size calculations for studies with binary matched-pairs data incorporating missing observations.

Main Methods:

• Development of a generalized estimating equations (GEE) approach to estimate confidence intervals for risk differences, accommodating partially observed pairs.
• Comparison of the GEE approach with methods for complete data and methods handling missing data via simulations.
• Application of the methods to recalculate sample sizes for binary matched-pairs observational studies.

Main Results:

• The GEE approach demonstrates comparable performance to complete-data methods in large sample sizes.
• In scenarios with missing data (MCAR/MAR), the GEE method yields higher statistical power and narrower interval widths.
• The GEE approach is non-inferior to multiple imputation and other specialized missing data methods, and leads to smaller sample size requirements.

Conclusions:

• The proposed GEE approach is a powerful and flexible alternative for non-inferiority testing with binary matched-pairs data, effectively handling missing observations.
• This method enhances analytical flexibility by enabling the inclusion of additional covariates.
• The GEE approach can be utilized even when initial sample size calculations were based on different statistical methods.