Parametric latent class joint model for a longitudinal biomarker and recurrent events

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Analyzing longitudinal biomarkers alongside recurrent events presents statistical challenges.
• Existing models may not fully capture the complex interplay between biomarker trajectories and event occurrences.
• Population heterogeneity can influence both biomarker levels and event rates.

Purpose of the Study:

• To propose a flexible joint statistical model for longitudinal biomarkers and recurrent event data.
• To incorporate the effects of covariates, accumulated events, and post-event interventions.
• To address underlying population heterogeneity using a latent class structure.

Main Methods:

• Developed a joint modeling framework linking biomarker and recurrent event processes.
• Employed a latent class structure to account for unobserved population heterogeneity.
• Utilized the Expectation-Maximization (EM) algorithm for parameter estimation and penalized likelihood for determining the number of latent classes.

Main Results:

• The proposed joint model effectively accommodates covariates influencing both biomarker and event processes.
• The latent class structure successfully captured associations between the biomarker and recurrent events, reflecting population heterogeneity.
• Model performance was validated through simulation studies and demonstrated on a real-world epileptic seizure dataset.

Conclusions:

• The developed joint model provides a robust approach for analyzing complex longitudinal biomarker and recurrent event data.
• The latent class approach effectively handles population heterogeneity, offering deeper insights into disease dynamics.
• This methodology is valuable for understanding disease progression and the impact of interventions in clinical studies, such as epilepsy research.