Automatic Region-wise Spatially Varying Coefficient Regression Model: an Application to National Cardiovascular Disease Mortality and Air Pollution Association Study

TL;DR

This paper introduces a novel statistical framework that automatically detects and estimates region-specific associations between air pollution and cardiovascular mortality across the U.S., enhancing understanding of spatial health impacts.

Contribution

The study develops an automatic, region-wise spatially varying coefficient model combining spectral graph techniques and regression for analyzing areal health data.

Findings

Identified regions with distinct PM2.5 and mortality associations.

Validated the model's performance through simulation studies.

Provided insights for targeted environmental health policies.

Abstract

Motivated by analyzing a national data base of annual air pollution and cardiovascular disease mortality rate for 3100 counties in the U.S. (areal data), we develop a novel statistical framework to automatically detect spatially varying region-wise associations between air pollution exposures and health outcomes. The automatic region-wise spatially varying coefficient model consists three parts: we first compute the similarity matrix between the exposure-health outcome associations of all spatial units, then segment the whole map into a set of disjoint regions based on the adjacency matrix with constraints that all spatial units within a region are contiguous and have similar association, and lastly estimate the region specific associations between exposure and health outcome. We implement the framework by using regression and spectral graph techniques. We develop goodness of fit criteria for model assessment and model selection. The simulation study confirms the satisfactory performance of our model. We further employ our method to investigate the association between airborne particulate matter smaller than 2.5 microns (PM 2.5) and cardiovascular disease mortality. The results successfully identify regions with distinct associations between the mortality rate and PM 2.5 that may provide insightful guidance for environmental health research.