Nonparametric Bayesian functional two-part random effects model for longitudinal semicontinuous data analysis.

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Longitudinal semicontinuous data, common in health and social sciences, present analysis challenges.
• Existing two-part random effects models (TPREM) have limitations in handling functional covariates and subgroup structures.
• Standard TPREM often assumes normality of random effects, which may not hold in practice.

Purpose of the Study:

• To propose a novel nonparametric Bayesian functional two-part random effects model (TPREM).
• To assess associations between longitudinal semicontinuous outcomes and various covariates, including functional ones.
• To relax the normality assumption for random effects, enabling subgroup identification.

Main Methods:

• Developed a nonparametric Bayesian functional TPREM.
• Utilized a Dirichlet process mixture of normals for random effects to capture heterogeneity.
• Applied the model to social insurance expenditure data and simulation studies.

Main Results:

• The proposed model effectively handles longitudinal semicontinuous data with functional covariates.
• The Dirichlet process mixture of normals successfully identified underlying subgroup structures.
• Demonstrated the model's utility and robustness through real-world data and simulations.

Conclusions:

• The nonparametric Bayesian functional TPREM offers a flexible and powerful approach for analyzing complex longitudinal data.
• This methodology advances the statistical toolkit for researchers in biomedical, epidemiological, and social sciences.
• The model provides a robust framework for uncovering hidden patterns and associations in semicontinuous longitudinal data.