Association between Stroke Prevalence and PM2.5 : Classical and Spatial Regression Analyses Using Satellite Remote Sensing Imagery and 500 Cities Project Data

This chapter assesses the association between adult stroke mortality prevalence rate and long-term exposure to PM2.5 (ambient air pollution of particulate matter with an aerodynamic diameter of 2.5
m or less) adjusting for unhealthy lifestyles and other health conditions. Health data, based on 2017 or 2016 model-based small area estimates of chronic disease, was obtained from the “500 Cities Project” 2019 release. PM2.5 data for the year 2016 was acquired from the “The Global Annual PM2.5 Grids from MODIS, MISR and SeaWiFS Aerosol Optical Depth (AOD), V4.03 (1998-2019)” datasets. Average PM2.5 was calculated for each city using a GIS zonal statistics function. First, classic ordinary least (OLS) square regression modeling (univariate linear regression, and a multivariate linear regression model fitted using a generalized linear model via penalized maximum likelihood). However, Moran’s I test found significant positive spatial autocorrelation with stroke data, PM2.5, and the residues of the OLS multilinear regression model, which makes the OLS modeling unreliable. Therefore, two spatial regression models (spatial lag and spatial error) were further run to account for the spatial dependence. Both models have successfully explained away spatial dependency. The spatial error model has much better goodness-of-fit than the spatial lag model as indicated by the higher log-likelihood, lower Akaike info criterion (AIC), and lower Schwartz criterion. The spatial error model found that long-term exposure to ambient PM2.5 may increase the risk of stroke and that increasing physical activity, reducing smoking and body weight, enough sleep, and controlling diseases such as blood pressure, coronary heart disease, diabetes, and cholesterol, may mitigate the effect.