Semi-parametric methods of handling missing data in mortal cohorts under non-ignorable missingness

Area of Science:

• Biostatistics
• Epidemiology
• Longitudinal Data Analysis

Background:

• Cohort studies often face challenges with missing data due to participant death or non-ignorable dropout.
• Accurate inference for the surviving cohort requires specialized statistical methods to address these missing data patterns.
• Existing methods may be sensitive to model assumptions, necessitating robust approaches.

Purpose of the Study:

• To develop and evaluate semi-parametric statistical methods for modeling repeated outcomes in cohort data with missingness due to death and dropout.
• To enable partly conditional inference for the cohort of individuals who are alive at any given time point.
• To provide robust statistical tools for analyzing longitudinal health data in aging populations.

Main Methods:

• Inverse Probability Weighting (IPW): Upweights observed subjects to represent those alive but unobserved.
• Outcome Regression: Imputes missing outcomes for living subjects using conditional mean imputation based on observed data.
• Augmented Inverse Probability (AIP): Combines IPW and outcome regression for double robustness against model misspecification.
• Methods are developed for both monotone and non-monotone missing data patterns.

Main Results:

• The proposed semi-parametric methods provide a framework for robust statistical inference in the presence of informative missing data.
• Application to the Health and Retirement Study cohort demonstrates the utility of these methods for analyzing elderly adult health.
• Sensitivity analyses confirm the robustness of the findings to potential violations of missing data assumptions.

Conclusions:

• The developed inverse probability weighted, outcome regression, and augmented inverse probability methods effectively address missing repeated outcomes in cohort studies.
• These methods facilitate reliable, partly conditional inference for the surviving cohort, crucial for understanding long-term health trends.
• The study provides valuable statistical tools for researchers analyzing longitudinal data in aging and health research.