Analyzing spatial patterns of segregation, the built environment, and racialized health disparities

Abstract

This dissertation uses GIScience to examine the role of residential segregation and the built environment in shaping racialized health disparities. The first study uses a dimensionality reduction technique and spatial modeling approach to address the legacy of historic redlining in shaping the contemporary distribution of social and environmental determinants of health in Baltimore, MD, and Philadelphia, PA. By linking historic segregation mechanisms to present-day spatial inequities, this analysis provides evidence for the role of segregation as a fundamental cause of health disparities. The second study leverages electronic health records (EHRs) representing diabetes-related emergency department (ED) visits and an explainable machine learning approach to understand racialized disparities in acute diabetes outcomes and modifiable risk factors. Analysis in this study is stratified by patient race and by level of segregation, which prompts policy-relevant insights about the drivers of disparities in ED utilization. The third study is motivated by the recent growth of generative artificial intelligence, and harnesses advancements in pre-trained large language models (LLMs) to generate synthetic spatial data for use in downstream machine learning tasks as applied to health equity research. Overall, this dissertation advances methodological and substantive contributions at the intersection of GIScience, health geography, and social epidemiology. The work presented in the following chapters advances scholarly understanding of the etiology of racialized health disparities while introducing novel computational approaches to support geospatial health equity research.