Leveraging Healthcare Analytics to Uncover Social Determinants of Health Disparities in Congestive Heart Failure

TL;DR

This study uses healthcare data to show how social factors like age, financial insecurity, and depression contribute to heart failure disparities.

Contribution

The paper introduces a machine learning approach to identify social determinants of health patterns in heart failure patients using EHR data.

Findings

Age was the strongest predictor of CHF, with odds increasing nearly tenfold for those aged ≥75 years.

Random Forest achieved moderate performance (AUC = 0.67) with age, transportation barriers, and physical activity as top predictors.

Clustering revealed subgroups with co-occurring social needs like depression and financial strain.

Abstract

Background

Congestive heart failure (CHF) is shaped by both clinical risk factors and social determinants of health (SDOH), which influence disease progression, medication adherence, and care access. As healthcare systems incorporate SDOH screening into routine electronic health record (EHR) workflows, understanding the relationship between social needs and CHF outcomes is critical for risk stratification and population management.

Methods

A retrospective analysis was conducted using 30,534 inpatient EHR records, including 7,618 patients with CHF and 22,916 without CHF. Associations between CHF status and SDOH variables were examined using chi-square tests. Multivariable logistic regression evaluated adjusted predictors of CHF, and a Random Forest (RF) classifier assessed predictive importance of SDOH and demographic features. The Synthetic Minority Oversampling Technique (SMOTE) was applied to address class imbalance. Hierarchical clustering was performed to identify patterns of co-occurring social needs.

Results

Patients with CHF were significantly older and more frequently male and Black compared with patients without CHF (p < 0.001). Age was the strongest predictor of CHF, with adjusted odds increasing nearly tenfold among individuals aged ≥75 years compared with younger adults. Depression registry entry and financial insecurity increased odds, whereas being married or living with a partner was protective. All SDOH were significantly associated with CHF except tobacco use, although effect sizes were small. Random Forest achieved moderate performance (area under the receiver operating characteristic (ROC) curve (AUC) = 0.67, balanced accuracy = 62%), with age, transportation barriers, and physical activity ranking as top predictors. Clustering revealed distinct subgroups, including patients with co-occurring depression, alcohol use, stress, and another characterized by financial strain, food insecurity, and transportation limitations.

Conclusion

Demographic and social conditions contribute meaningfully to CHF burden, with age, socioeconomic stressors, and psychosocial vulnerability emerging as key factors. Despite small individual effect sizes, collective SDOH patterns demonstrated clinical relevance and predictive utility. Integrating SDOH variables into analytic models, supported by machine learning and oversampling techniques such as SMOTE, provides a scalable approach to improve CHF risk stratification and guide targeted intervention strategies.

Figures