A joint model for longitudinal and survival data based on an AR(1) latent process

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Nonignorable missing observations are a critical challenge in repeated measurement studies.
• Traditional joint modeling often assumes time-constant random effects, limiting applicability.
• Existing methods struggle with diverse data types and complex missing data patterns.

Purpose of the Study:

• To develop a flexible joint modeling framework for longitudinal and survival data.
• To incorporate time-varying subject-specific random effects using an autoregressive process.
• To accommodate various longitudinal response types (continuous, binary, count) and extended cases.

Main Methods:

• Introduced time-varying random effects following a first-order autoregressive (AR(1)) process.
• Employed a generalized linear model formulation for diverse longitudinal data.
• Utilized a hidden Markov model recursion for likelihood computation and quasi-Newton optimization for parameter estimation.

Main Results:

• The proposed joint model effectively handles nonignorable missing observations with time-varying random effects.
• The method demonstrated flexibility in accommodating continuous, binary, and count longitudinal data.
• Parameter estimation provided accurate standard errors via the information matrix.

Conclusions:

• The novel joint modeling approach offers a robust solution for analyzing longitudinal and survival data with complex missingness.
• The inclusion of time-varying random effects enhances the model's realism and applicability.
• Validated through simulation studies and real-world medical data analysis.