Responding to Health Equity Crises in Underserved Populations

Abstract

Health equity crises—surges in maternal mortality, infectious disease outbreaks, chronic disease decompensation, and behavioral health emergencies that disproportionately affect low-income, rural, tribal, and minority communities—continue to outpace the capacity of conventional public health resource allocation systems. Allocation decisions remain largely reactive, anchored to annual needs assessments and static social vulnerability thresholds that cannot track the rapid, compounding shifts in community-level risk that characterise crisis conditions. This paper presents EquiNet, a spatiotemporal graph-AI framework for predictive risk stratification and equity-weighted resource allocation in underserved populations. The framework integrates a Spatiotemporal Graph Attention Network (ST-GAT) for community-level crisis risk prediction with an equity-weighted Isolation Forest ensemble for detecting emerging resource-access gaps, coordinated through a reinforcement-learning-based allocation optimiser operating over a continuously updated community–resource graph. Evaluated against 36 months of simulated public health surveillance data spanning 6,400 census tracts and 18.7 million residents across urban, rural, and tribal regions, EquiNet achieves a crisis-risk prediction AUC of 0.954, an F1-score of 0.895, and reduces mean time-to-intervention (MTTi) from 97 days under manual needs assessment to 3.2 days. Equity-gap indices across six historically underserved population subgroups improve by an average of 41.6%, and automated coverage of resource-allocation indicators across six crisis domains reaches 89.4% of the 187 indicators mapped to Healthy People 2030, the CDC/ATSDR Social Vulnerability Index, and HRSA Uniform Data System reporting requirements. These results indicate that graph-AI–native, equity-weighted allocation systems can convert fragmented public health surveillance data into proactive crisis response at a pace consistent with the speed at which health disparities compound.