ABSTRACT:
Health outcomes in urban populations are strongly influenced by spatial pat- terns of underlying risk factors, making the understanding of spatial depen- dencies essential for accurate modeling. Spatial weight matrices quantify how health-related variables at one location relate to those at another. Traditional approaches often rely on Euclidean distance, assuming straight-line prox- imity between locations. However, in geographically disjoint urban areas, Euclidean distance can misrepresent true spatial relationships, potentially biasing health risk assessments. In this study, we introduce a connectiv- ity structure derived from the underlying road network to more realistically capture spatial relationships between households. We compare spatial de- pendency modeling using Euclidean distance versus road network distance across three geographically separated areas in South Africa: Melusi, At- teridgeville, and Hillbrow. Road network distances were calculated using shortest-path algorithms, providing a more accurate representation of how individuals and households are connected in urban settings. We applied Lo- cal Indicators for Categorical Data (LICD) to detect spatial clustering of key health-related categorical variables, including seropositivity, access to exclu- sive toilets, hygiene practices, and employment status. The results revealed distinct pockets of elevated health risks that were more accurately identified using network-based distances. Beyond mapping spatial dependencies, our analysis highlights demographic disparities across urban areas, and examines how factors such as population density influence health outcomes. By un- covering the relationships between urban infrastructure, population density, and epidemiological risk factors, this research offers actionable insights for public health planning and urban policy aimed at reducing health inequities and improving disease prevention strategies.