388MO Uncovering hidden social determinants of health in urological cancer patient notes using natural language processing

Individuals with multiple adverse social determinants of health (SDOH) face up to 2.75X higher mortality risk. SDOH features are often buried in unstructured notes, beyond the reach of traditional data extraction tools, and are time consuming for manual review, resulting in them often being missed. We evaluated how accurately an AI model could extract SDOH from urologic cancer notes to uncover the data that clinicians and health systems need to provide informed, patient centered, equitable care.