Integrating a risk prediction score in a clinical decision support to identify patients with health-related social needs in the emergency department

Abstract

Objectives: To improve the identification of patients with health-related social needs (HRSNs) in the emergency department (ED), we developed and integrated a risk prediction score into an existing Fast Healthcare Interoperability Resources (FHIR)-based clinical decision support (CDS). Materials and methods: We conducted 2 phases of individual semi-structured qualitative interviews with ED clinicians to identify HRSN risk score design preferences for CDS integration. Following this, we used patient HRSN screening survey, health information exchange (HIE), and clinical data to run logistic regressions, developing an HRSN risk score aligned with ED clinician preferences. Results: Emergency department clinicians preferred HRSN risk scores displayed via visual cues like color-coding with different ranges (low, medium, and high) with higher model sensitivity to avoid missing patients with HRSNs. The overall performance of the risk prediction model was modest. Risk scores for food insecurity, transportation barriers, and financial strain were more sensitive, aligning with users' preference for inclusivity and accurately identifying patients likely to screen positive for these HRSNs. Discussion: The design and risk score model choices, such as visual displays with additional data, higher sensitivity thresholds, and use of different thresholds for fairness, may support effective CDS use by ED clinicians. Conclusion: Using HIE data and an external CDS is a feasible route for including patient HRSNs information in the ED. We relied on clinician preferences for incorporation into the existing CDS and were attentive to performance fairness. While the predictive performance of our risk score is modest, providing risk scores in this manner may potentially improve the identification of patients' HRSNs in the ED.