SpGPCW

SpGPCW estimates spatially varying critical exposure windows by fitting a spatiotemporally structured Gaussian process within a hierarchical Bayesian logistic regression framework to relate time-varying environmental exposures (e.g., ozone, PM2.5) to pregnancy outcomes such as term low birth weight.

Key Features:

• Spatially Varying Gaussian Process Model: Employs a spatiotemporally structured Gaussian process within a hierarchical Bayesian logistic regression framework to estimate critical windows of susceptibility across geographic regions.
• Hierarchical Bayesian Framework: Uses Markov chain Monte Carlo (MCMC) sampling for parameter estimation and uncertainty quantification in complex data structures.
• Spatial Correlation and Temporal Smoothness: Models areal-level spatial correlation between lagged health effect parameters and enforces temporal smoothness across pregnancy periods.
• Application to Air Pollution Data: Applied to analyze associations between average weekly concentrations of ozone and PM2.5 during pregnancy and term low birth weight using North Carolina birth records.
• Bayesian Model Comparison: Implements Bayesian model comparison techniques to assess the presence and importance of spatial variability in critical window sets.

Scientific Applications:

• Critical Window Identification: Identifies pregnancy periods when exposure to pollutants like PM2.5 and ozone is most strongly associated with outcomes such as term low birth weight, accounting for spatial variability.
• Enhanced Inference and Model Comparison: Improves parameter inference by accounting for spatial variability and enables comparison of models with and without spatial structure.

Methodology:

Hierarchical Bayesian logistic regression with a spatiotemporally structured Gaussian process fitted via MCMC sampling to estimate lagged exposure–outcome associations while accounting for areal spatial correlation and temporal smoothness.