BACKGROUND: With enhanced digitization of large-scale health and socio-economic data, it has generated new opportunities and challenges in the processes and planning of smart cities in the cities and public health. It is increasingly becoming necessary to have combined analytical frameworks that are able to work out these data into actionable information in dealing with health disparities. The paper hypothesizes a geospatial analytical model, the Spatial Health Information Tool (SHIT), which can be used to study health inequity at the county level in the United States METHODS: The framework was constructed with the help of the CDC PLACES 2024 data that comprised 3,142 counties and 40 health measures. To determine the spatial autocorrelation and geographic hot spots of unfavorable health outcomes, spatial statistical tools, such as Global Moran I, Local Moran I and Getis-Ord statistics were used. DBSCAN was used to perform density-based clustering to identify spatial groups of counties that were similar in their risk profiles. Besides, a random forest regression was also used to assess predictive relations between identified risk factors and health outcomes, which allowed considering the relative significance of determinants that led to health disparities. In an attempt to further encode spatial relationships among neighboring regions, a Graph Neural Network (GNN) was used to encode spatial relationship across counties. Moreover, a composite Health Vulnerability Index was computed by summing up the choice of health indicators into a normalized score in order to determine counties with high cumulative health risk. To enhance the explanatory power of the analytical findings, the Explainable Artificial Intelligence (XAI) methods were included to determine the most significant determinants of health inequalities. This element enables the framework to connect the spatial health patterns with their socioeconomic or behavioral risk factors. The framework has been equipped with interactive visualization modules, such as choropleth maps, scatterplots, spatial queries, and three-dimensional prevalence surface maps as support of the exploratory spatial analysis. RESULTS: The presented framework was able to find statistically significant spatial clusters of health disparities at the county-level in the U.S and found specific geographic patterns of high disease burden and high prevalence of risk factors. The clustering based on density also demonstrated clusters of counties with comparable vulnerability patterns. The random forest analysis revealed that there are significant predictive relationships between socio-economic and behavioral risk factors and unfavourable health outcomes and some of the major determinants are consistently at the top of feature importance. The predictions of the Graph Neural Network also demonstrated patterns of spatially correlated patterns of health risk between neighboring counties, proving the existence of regional health disparities. Besides that, an index of Health Vulnerability helped identify the counties where several health risk factors are combined, which is a convenient measure to inform the priorities of specific public health interventions. Explainable AI analysis also demonstrated the most important determinants and linked with unfavorable health outcomes and offered a way to interpret the reasons behind the regional health disparities. The interactive modules allowed the stakeholders to visually examine these patterns and relationships in an easy and user-friendly way. CONCLUSIONS: The SHIT framework is a powerful decision-support system that can be used to investigate the issue of health inequities because it combines open health data, spatial statistics, and clustering methods with machine learning in a single and interactive system. The method produces policy-relevant details that might be used to inform specific interventions, equal distribution of resources and infrastructural planning. The present work illustrates the positive value of spatial big data analytics in enhancing smart city management plans and helping to advance the quality and health of the population.
Chauhan et al. (Mon,) studied this question.