Bayesian nonparametric latent class model for longitudinal data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Population Heterogeneity

Background:

• Latent class models identify unobserved population heterogeneity in longitudinal studies.
• Traditional models face challenges in determining the optimal number of latent classes.
• Predictors can influence class allocation probabilities in existing models.

Purpose of the Study:

• To propose a Bayesian nonparametric latent class model for longitudinal data.
• To allow the number of latent classes to be inferred directly from the data.
• To characterize latent classes of estradiol trajectories during menopausal transition.

Main Methods:

• Utilizes an infinite mixture model with predictor-dependent class allocation.
• Each individual's longitudinal trajectory is modeled using class-specific linear mixed effects models.
• Employs Markov chain Monte Carlo methods for parameter estimation.

Main Results:

• The proposed model successfully infers the number of latent classes from data.
• Demonstrates effective characterization of estradiol trajectories in a menopausal transition cohort.
• Validates the model's performance through simulation and real-world data analysis.

Conclusions:

• The Bayesian nonparametric latent class model offers a flexible approach for longitudinal data analysis.
• It overcomes limitations of traditional methods by data-inferring the number of classes.
• Provides robust characterization of population heterogeneity in health trajectories, exemplified by menopausal estradiol patterns.