Integrating Multi-Source Social Determinants of Health (SDOH) Data into Predictive Analytics Frameworks to Mitigate Health Inequities and Reduce Long-term Uncompensated Care Costs in U.S. Urban Public Health Systems

Abstract

Background: U.S. urban public health systems face persistent health inequities and rising uncompensated care costs,

partly driven by unaddressed social determinants of health (SDOH). Objective: This study proposes and evaluates a

predictive analytics framework integrating multi-source SDOH data (housing, food security, transportation) to

identify high-risk patients, target interventions, and reduce long-term costs. Methods: A mixed-methods design was

employed, combining retrospective analysis of electronic health records (EHRs) from two large urban public health

systems (2018–2023) with semi-structured interviews of 45 healthcare administrators and data scientists. A machine

learning model (gradient boosting) was developed using EHR data and publicly available SDOH indices (e.g., Area

Deprivation Index). Findings: The integrated SDOH-EHR model improved high-risk patient identification by 34%

(AUC 0.89) compared to a clinical-only model (AUC 0.72). Predictive targeting reduced preventable emergency

department visits by 22% over 18 months and projected a 15–18% reduction in long-term uncompensated care costs.

Conclusion: Integrating multi-source SDOH data into predictive analytics frameworks can significantly enhance

health equity and financial sustainability in urban public health systems