Evaluating the Implementation of Social Determinants of Health Screening Across 18 Hospitals.
TL;DR
Despite a unified tool and training, screening rates varied widely across sites; clinical correlations validated the tool; the implementation process appears sustainable.
Abstract
Background
Social determinants of health (HRSN) significantly impact health outcomes, prompting mandates from the Centers for Medicare and Medicaid Services (CMS) for health system screening and reporting. Although health systems are rapidly implementing HRSN screening, the quality of the data collected has not been reported.
Objective
Evaluate aspects of data quality for inpatient HRSN screening tool implemented within an integrated health network of 18 hospitals, heterogeneity in screening performance, and explore clinical outcomes.
Design
Observational study using electronic health record (EHR) data from May to December 2024.
Participants
Adult patients admitted to 18 hospitals in the Atrium Health System (Southeastern United States). MAIN MEASURES We evaluated data quality of inpatient HRSN screening assessments using a data quality framework, including variation in screening performance by hospital site, assessed HRSN domain correlation, and finally explored the clinical outcomes of patients who screened positive for HRSN using descriptive statistics and mixed-effects logistic regression modeling. KEY
Results
HRSN assessment responses demonstrated conformance, completeness, and plausibility in the data quality framework. Of 197,305 eligible inpatient encounters, 172,519 (87.4%) had HRSN screens completed, with variation by market (92.1% vs. 72.4%). Among those screened, 12.7% (21,900) were positive for at least one HRSN, and this rate was similar across hospitals. There was moderate to strong HRSN intradomain correlation (e.g., food insecurity, Cramer V = 0.87) and weak correlations between domains (e.g., food and housing insecurity, Cramer V = 0.44). Patients screening positive were more likely to have Medicaid (23.1% vs. 13.7%), social vulnerability (median index 0.74 vs. 0.67), and increased 30-day readmission rates (14.3% vs. 11%; all p < 0.001).
Conclusions
Despite a unified tool and training, screening rates varied widely across sites; clinical correlations validated the tool. The implementation process appears sustainable. Qualitative work is needed to understand site-level differences in screening performance.