Using Artificial Intelligence to Detect Housing Status in Unstructured Electronic Health Record Data
Accurate data are essential for advancing medical care for people experiencing homelessness (PEH), yet current methods of identifying PEH in clinical datasets remains a challenge.
Standardized screening instruments often produce inaccurate results. Chart review is more accurate but not feasible at scale.
Large language models (LLMs) show promise for extracting social determinants of health from electronic health record (EHR) notes but evaluating housing identification performance has been difficult without an EHR–based reference standard.
This study found that, across 4,836 admissions, a large language model (LLM) identified more PEH than a non-standardized nursing intake form, offering a scalable solution for housing data extraction at minimal cost.
Authors: Abbott Gifford, BA; Shraddha Pandey, BA; Hannah Decker, MD; Margot Kushel, MD; Logan Pierce, MD;
Elizabeth Wick, MD.
Related Resources
