PROJECT SUMMARY/ABSTRACT
Social determinants of health are known to have a significant impact on patient outcomes. However,
controversies exist on how to best capture this information in routine care. Most SDH information is captured in
the form of a survey or unstructured free text or narrative and not regularly captured or screened for variety of
factors (e.g., time constraints, clinician experience/comfort in asking, patient fears of sharing potentially
stigmatizing information. This is a particularly rich and robust source of information, especially when trying to
identify patients' goals of care, preferences, or behavioral/social challenges that may exist. In this proposal, we
use natural language processing and generative AI models to capture SDH information from patients. We will
then process this information into discrete data elements that can then be passed into the EHR and acted upon
by clinical decision support system. We will pilot this intervention in a large academic medical center that
provides care to an at-risk patient population and community.
