Comparison of population-averaged and subject-specific approaches for analyzing repeated binary outcomes

Area of Science:

• Epidemiology
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal studies generate binary outcomes, requiring specialized statistical modeling.
• Existing methods often fall into population-averaged or subject-specific categories.
• Clarity on the relationship between these methods and traditional analyses is lacking for epidemiologists.

Purpose of the Study:

• To compare different statistical models for longitudinal binary data.
• To clarify the relationship between population-averaged and subject-specific approaches.
• To examine the interpretation of covariates under various models.

Main Methods:

• Analysis of a binary outcome from the Midwestern Prevention Project, a longitudinal smoking prevention trial.
• Comparison of stratified analysis, standard logistic models, conditional logistic models, generalized estimating equations (GEE), and random-effects models.
• Evaluation of models using two and seven repeated measurements.

Main Results:

• The study provides a detailed comparison of model outputs and interpretations.
• Differences in estimating effects of time-varying and time-invariant covariates are highlighted.
• The practical implications for applying these models in epidemiologic research are discussed.

Conclusions:

• Understanding the nuances between different longitudinal data analysis models is crucial for accurate epidemiologic inference.
• The choice of model impacts the interpretation of covariate effects.
• This comparison aids researchers in selecting appropriate methods for longitudinal binary outcomes.