Semiparametric regression models and sensitivity analysis of longitudinal data with nonrandom dropouts

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Nonrandom dropouts pose a significant challenge in longitudinal studies, potentially biasing results.
• Existing methods may not adequately address complex dropout mechanisms in outcome analysis.
• Accurate analysis of longitudinal data requires robust methods to account for missingness patterns.

Purpose of the Study:

• To propose a novel family of regression models for analyzing longitudinal outcomes with nonrandom dropouts.
• To develop a shared random effects model for a more interpretable dropout selection parameter.
• To conduct a sensitivity analysis to assess the robustness of inferences under different model assumptions.

Main Methods:

• Development of generalized linear models with random effects incorporating a dropout selection mechanism.
• Formulation of a shared random effects model for interpretable dropout parameter estimation.
• Implementation of semiparametric and parametric models within a sensitivity analysis framework, addressing identifiability concerns through parameter fixing and global sensitivity testing.

Main Results:

• The proposed semiparametric model demonstrated a substantial reduction in bias compared to the parametric model.
• Bias reduction was particularly notable in scenarios with high dropout rates or misspecified dropout models.
• The dropout selection parameter in the shared random effects model offered meaningful interpretation.

Conclusions:

• The proposed regression modeling framework effectively adjusts for nonrandom dropouts in longitudinal data.
• The semiparametric approach offers improved robustness against bias compared to parametric methods.
• The methodology provides a valuable tool for reliable analysis of longitudinal data with complex missingness.