Joint Hidden Markov Model for Longitudinal and Time-to-Event Data with Latent Variables

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Simultaneous analysis of longitudinal and time-to-event data is crucial for understanding complex health trajectories.
• Latent variables play a significant role in mediating or confounding these relationships.
• Existing methods may not adequately capture the dynamic interplay between repeated measures and event occurrences.

Purpose of the Study:

• To develop a novel joint modeling framework for analyzing longitudinal and time-to-event data with latent variables.
• To investigate the influence of both observed and latent risk factors on time-to-event outcomes.
• To explore dynamic heterogeneity and risk factors within longitudinal processes.

Main Methods:

• A three-component model integrating a hidden Markov model (HMM) for longitudinal data, factor analysis for latent variables, and a proportional hazards model for time-to-event data.
• Inclusion of a shared random effect to account for correlation between longitudinal and time-to-event outcomes.
• Bayesian inference utilizing Markov chain Monte Carlo (MCMC) methods for statistical analysis.

Main Results:

• The proposed joint model effectively integrates longitudinal and time-to-event data streams.
• Simulation studies demonstrate the robustness and performance of the developed statistical approach.
• The model successfully identified significant risk factors influencing cognitive impairment and mortality in Chinese elders.

Conclusions:

• The developed joint modeling approach provides a powerful tool for analyzing complex health data with latent structures.
• This methodology offers enhanced insights into the dynamics of health outcomes and associated risk factors.
• The application highlights the model's utility in epidemiological research, particularly for aging populations.