Analysis of crossover designs for longitudinal binary data with ignorable and nonignorable dropout

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Clinical Trial Methodology

Background:

• Longitudinal binary data in crossover designs frequently exhibit missingness due to ignorable and nonignorable dropout.
• Accurate analysis is crucial for reliable treatment effect estimation in such scenarios.

Purpose of the Study:

• To evaluate conditional and marginal models for longitudinal binary crossover data with missingness.
• To establish the relationship between conditional and marginal parameters for comparing treatment mean effects.
• To investigate model performance under various missing data mechanisms and distributions.

Main Methods:

• Extensive simulation studies were conducted under complete and missing data scenarios.
• Selection models were employed to handle nonignorable dropout.
• Generalized estimating equations (GEE) and generalized linear mixed-effects models (GLMM) with pseudo-likelihood estimation were utilized.
• A controlled multiple imputation method was proposed for sensitivity analysis.

Main Results:

• The study established the relationship between conditional and marginal model parameters.
• Simulation results demonstrated the performance of different models under various missing data conditions.
• GEE and GLMM with pseudo-likelihood provided valid and robust inference.
• The controlled multiple imputation method served as a valuable tool for assessing missing data assumption sensitivity.

Conclusions:

• Appropriate statistical models and methods are essential for valid inference in longitudinal binary crossover studies with missing data.
• Generalized estimating equations and generalized linear mixed-effects models are recommended for robust analysis.
• Sensitivity analysis using methods like controlled multiple imputation is crucial for evaluating the impact of missing data assumptions.