A Bayesian Local Gaussian-Copula Allocation Model for Bivariate Discrete-Time First-Event Data With Time-Dependent...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Bivariate first-event outcomes are common in longitudinal biomedical research.
• Analyzing related events requires specialized statistical methods.

Purpose of the Study:

• To propose a Bayesian local Gaussian-copula allocation model for bivariate discrete-time first-event data.
• To accommodate baseline and time-dependent covariates in the analysis.
• To provide a flexible framework for modeling dependence structures between events.

Main Methods:

• Developed a Bayesian local Gaussian-copula allocation model.
• Utilized discrete-time decrement models for marginal survival estimation.
• Employed Gaussian copula to allocate probability across four terminal outcomes within intervals.
• Considered various specifications for the dependence parameter (constant, interval-specific, hierarchical).
• Summarized covariate effects using Kullback-Leibler projection.

Main Results:

• The proposed Gaussian-copula models demonstrated competitive performance against odds-ratio benchmarks for covariate effect estimation.
• The hierarchical dependence specification offered a regularized approach compared to unstructured interval-specific dependence.
• The model was successfully illustrated using data from the National Health and Aging Trends Study.

Conclusions:

• The Bayesian Gaussian-copula model is a viable and flexible tool for analyzing bivariate first-event data in longitudinal studies.
• The hierarchical model provides a robust alternative for dependence modeling.
• The method yields comparable conclusions to traditional approaches while offering greater flexibility.