A BAYESIAN GROWTH MIXTURE MODEL FOR COMPLEX SURVEY DATA: CLUSTERING POSTDISASTER PTSD TRAJECTORIES

Area of Science:

• Statistics
• Biostatistics
• Survey Methodology

Background:

• Growth mixture models (GMMs) are underutilized for complex survey data.
• Existing pseudo-likelihood methods may lead to efficiency loss due to survey weighting.
• Complex sample designs (stratification, clustering) require specialized analytical approaches.

Purpose of the Study:

• To propose a Bayesian GMM that incorporates complex sample design features.
• To improve estimation bias and efficiency for survey data analysis.
• To analyze longitudinal post-traumatic stress disorder (PTSD) trajectories in a hurricane-affected population.

Main Methods:

• Developed a Bayesian GMM incorporating sample design features as covariates or variance components.
• Implemented an efficient Gibbs sampler with closed-form conditional distributions.
• Applied the model to data from the Galveston Bay Recovery Study (GBRS) with a stratified multi-stage cluster design.

Main Results:

• Identified four clinically meaningful PTSD trajectory subgroups among residents of southeastern Texas post-Hurricane Ike.
• Characterized risk factors associated with different PTSD trajectory subgroup memberships.
• Demonstrated potential for reduced bias and increased efficiency compared to pseudo-likelihood methods.

Conclusions:

• The proposed Bayesian GMM provides a robust framework for analyzing complex survey data.
• The method offers advantages in bias reduction and efficiency, particularly when design features are informative.
• An accompanying R package, Bsvygmm, is available for practical implementation.