Geospatial Analysis Using Social Determinants of Health, Clinical Data, and Spatial Regression Methods

This chapter describes how to conduct geospatial analyses using social determinants of health (SDOH) data and clinical data derived from electronic health records (EHRs). We cover how to conduct neighborhood-level analyses where each census tract is the unit of analysis. These analyses combine information from the American Community Survey (ACS), Census, and open access datasets to explore a wide range of neighborhood-level factors that could be contributing to neighborhood-level differences in disease prevalence, incidence, or some other outcome of interest observed at the neighborhood level using clinical data from EHRs. We cover how to make interactive maps that enable users to click on census tracts to explore more details on those at-risk census tracts. We also quantify the spatial clustering seen visually in maps using a variety of statistical tests. We start with Moran’s I-test, which provides a high-level understanding of if spatial clustering is being observed for a covariate of interest in a particular geographic area. We then work with a few different spatial regression methods, including conditional auto regressive (CAR), simultaneous auto regressive (SAR), and spatial simultaneous autoregressive lag and spatial Durbin (mixed) models. We build on knowledge learned in other chapters and detail how to perform a neighborhood-wide association study (N-WAS) similar to a phenotype-wide association study (PheWAS). In an N-WAS, we assess each neighborhood-level covariate’s spatial association with an outcome of interest at the univariate level. We also cover how to construct a mixed model that accounts for individual, and neighborhood-level affects in the same model along with the computational complexity issues that this type of model brings. Finally, we cover how to involve the affected communities in the analysis by presenting preliminary, but robust, analyses to them for their insights, guidance, and feedback. After reading this chapter, you should be able to confidently answer the following questions: