AI-Driven Social Determinants of Health (SDoH) Risk Scoring: Integrating Geospatial and Clinical Data to Prevent Readmissions in Vulnerable Patient Populations

• Sunday Sunday

  Ladoke Akintola University of Technology

  Author

Hospital readmissions within 30 days of discharge represent a persistent challenge for healthcare systems, contributing to elevated costs, adverse patient outcomes, and systemic inefficiencies. While traditional readmission prediction models have predominantly relied on clinical variables derived from electronic health records, mounting evidence suggests that social determinants of health-the conditions in which patients live, work, and age-are equally consequential predictors of post-discharge outcomes. Despite this recognition, existing approaches suffer from three critical limitations: they employ narrow SDOH proxies rather than comprehensive multidimensional indicators, they fail to leverage geospatial patterns at meaningful granularity, and they lack validated frameworks that integrate clinical and community-level data in real-time. This study addresses these gaps by developing and validating a machine learning framework that integrates 752 geocoded SDOH indicators with 42 clinical variables extracted from electronic health records to predict 30-day all-cause readmissions across a cohort of 33,579 patients with heart failure. Using a hybrid modeling approach that systematically compared Logistic Regression, Random Forest, and XGBoost algorithms, the XGBoost model incorporating expanded SDOH predictors achieved superior performance (ROC-AUC = 0.671; 95% CI: 0.658-0.684), representing a 6.2% improvement over clinical-only models and a 3.9% improvement over models using traditional composite SDOH indices. Feature importance analysis revealed that environmental predictors-including housing cost burden and air quality-ranked among the top predictors alongside clinical biomarkers, underscoring the multidimensional nature of readmission risk. Furthermore, algorithmic fairness analysis demonstrated that expanded SDOH models improved predictive equity across racial groups, reducing the equalized odds ratio from 0.329 to 0.437 when compared to traditional indices. The findings contribute a replicable, interpretable framework for SDOH-informed risk stratification that enables proactive, targeted interventions for vulnerable patient populations. For healthcare administrators and policymakers, this research provides an evidence-based pathway for moving beyond clinical-only prediction models toward more equitable and accurate risk assessment that addresses the fundamental drivers of preventable readmissions.

• pdf

Copyright (c) 2026 Sunday Sunday (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.