Analysis of crossover designs with nonignorable dropout

Area of Science:

• Biostatistics
• Clinical Trials
• Longitudinal Data Analysis

Background:

• Crossover designs are efficient but susceptible to dropout.
• Nonignorable dropout can bias treatment effect estimates.
• Handling missing data in longitudinal studies is crucial.

Purpose of the Study:

• To develop statistical methods for analyzing crossover designs with nonignorable dropout.
• To compare treatment mean effects in the presence of missing data.
• To assess the robustness of proposed methods.

Main Methods:

• Jointly modeling longitudinal outcomes and discrete time to dropout.
• Utilizing shared-parameter and mixed-effects selection models.
• Adapting linear-mixed effects and discrete-time hazards models.
• Employing maximum likelihood estimation and controlled multiple imputation.

Main Results:

• Proposed methods provide valid estimation of treatment effects under nonignorable dropout.
• Simulation studies demonstrate robustness across various missing data mechanisms.
• Successful application to continuous and binary outcome examples.

Conclusions:

• The developed statistical approaches effectively address nonignorable dropout in crossover trials.
• These methods enhance the reliability of treatment effect comparisons.
• Sensitivity analyses confirm the robustness of the findings.