Model selection for semiparametric marginal mean regression accounting for within-cluster subsampling variability and...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal studies in elderly populations are crucial for understanding health outcomes like physical frailty.
• Semiparametric marginal mean regression is a common statistical approach for analyzing such data.
• Existing model selection criteria may not adequately handle informative cluster sizes, where the number of observations per subject relates to the outcome.

Purpose of the Study:

• To propose a novel model selection criterion, the Resampling Cluster Information Criterion (RCIC), for semiparametric marginal mean regression.
• To address the challenge of informative cluster sizes in longitudinal data analysis.
• To improve the accuracy of model selection in the context of elderly frailty studies.

Main Methods:

• Developed the Resampling Cluster Information Criterion (RCIC) based on within-cluster resampling principles.
• Ensured the RCIC implementation is computationally convenient, avoiding actual data resampling.
• Incorporated an additional component in RCIC to account for model variability across within-cluster subsampling.

Main Results:

• The RCIC method demonstrated remarkable improvements in selecting the correct model compared to existing methods.
• The criterion's effectiveness was validated regardless of whether the cluster size was informative or not.
• Application to a longitudinal frailty study identified key risk factors for physical frailty in the elderly.

Conclusions:

• The RCIC offers a computationally convenient and effective approach for model selection in semiparametric marginal mean regression with longitudinal data.
• The study identified female sex, advanced age, low income, low life satisfaction, and chronic health conditions as significant risk factors for physical frailty in the elderly.
• RCIC provides a valuable tool for analyzing complex longitudinal health data and identifying critical risk factors.