A Bayesian multivariate mixture model for skewed longitudinal data with intermittent missing observations: An...

Area of Science:

• Developmental Pediatrics
• Biostatistics
• Child Health Research

Background:

• Infant motor development tracking is crucial for identifying risks of adverse outcomes.
• Modeling infant motor development presents statistical challenges including data skewness, missingness, and longitudinal correlations.

Purpose of the Study:

• To develop a flexible Bayesian mixture model for analyzing infant motor development data.
• To accommodate skewed, intermittently missing, and correlated longitudinal data in infant growth studies.

Main Methods:

• Utilized a Bayesian mixture model with matrix skew-normal distributions for developmental trajectories.
• Employed a Pólya-Gamma data-augmentation scheme for cluster-membership probability modeling.
• Imputed missing data using conditional multivariate skew-normal distributions with Gibbs sampling for Bayesian inference.

Main Results:

• The proposed model demonstrated improved inferences compared to methods ignoring skewness or using conventional imputation.
• Identified two distinct infant motor development clusters in the Nurture study cohort.
• Revealed detrimental effects of food insecurity on infant motor development.

Conclusions:

• The developed Bayesian model effectively analyzes complex infant motor development data.
• Findings highlight the importance of considering data skewness and missingness in developmental studies.
• The study identified specific developmental trajectories and the negative impact of food insecurity, informing targeted interventions.