Biased from the Beginning: Disparities in Mental Health Language in Clinical Notes, an Original Video Vignette Survey Experiment

Fatima Fairfax, Duke University

Thursday, March 26, 2026 11:00 AM to 12:00 PM EDT
Johnson Center, Meeting Room F

From wearable biometrics to AI-driven decision support, technology is transforming clinical practice, enhancing precision, and redefining the delivery of care. While these innovations present a wealth of opportunities to improve healthcare in the U.S., greater attention is needed to evaluate the impact of these new technologies on healthcare quality, healthcare outcomes, and disparities in healthcare delivery. In this talk, I focus on how providers can infuse bias into new healthcare technologies and processes through their clinical records. Specifically, I examine if and how the content and quality of mental health information appear in clinical notes, and if that differs by race-gender presentation of the patient. Unstructured clinical notes are the key input into advanced healthcare technologies and algorithms and thus deserve more critical attention. Using an originally designed and administered video vignette survey experiment, we leverage the experimental design to isolate race-gender presentation as the key driver of difference in the clinical note. We find that clinical notes do vary by race and gender when holding presenting symptoms and social context constant. These findings highlight the importance of identifying and interrupting bias at multiple levels along the healthcare delivery system to ensure that new technologies can improve healthcare outcomes for all. 

Hosted by Department of Sociology and Anthropology.

Sponsored by Department of Sociology and Anthropology, College of Humanities and Social Sciences, Next System Studies.