<img src="https://www.utsouthwestern.edu/newsroom/articles/year-2026/assets/nurse-patient-header.jpg” alt=”Mature nurse bends over patient as male doctor observes” loading=”lazy”>
Using patient records from two emergency departments to analyze how large language models prioritized patients, UT Southwestern researchers found the programs to be susceptible to the same clinical stigmas associated with human assessments. (Photo credit:
DALLAS – Oct. 20, 2026 – Artificial intelligence (AI)-driven tools used in emergency triage may be susceptible to the same clinical stigmas associated with human assessments, researchers at UT Southwestern Medical Center reported in a study published in npj Digital Medicine. Their findings suggest this could result in harmful biases and wrongful prioritization of patients.
“Wording matters. Because AI models process the words we use to describe patients, stigmatizing language can steer clinicians away from the patients who need care the most,” said first author Philip Jarrett, M.D., M.B.A., Assistant Professor of Emergency Medicine. “This study is part of a broader research program on how clinical information should be presented to AI so it can safely help prioritize emergency patients competing for limited resources.”
Philip Jarrett, M.D., M.B.A., is Assistant Professor of Emergency Medicine at UT Southwestern.
The team analyzed patient records from emergency departments (EDs) at two major academic institutions: Stanford Medicine in California and Beth Israel Deaconess Medical Center in Massachusetts. They paired patients with similar initial condition severity levels, then asked three open-source AI large language models (LLMs) to predict which patients were more likely to deteriorate and who should be prioritized for care. Before demographic, social, or potentially stigmatizing information was added, all three models identified higher-risk patients at rates well above chance.
Researchers then added different types of information to patient records – including mental health and substance use history, age, and patterns of ED use – and evaluated how the AI models’ decisions changed. For example, they compared neutral language and stigmatizing language: “established patient with multiple chronic conditions” vs. “17 visits in the past 12 months” and “opioid use disorder, on maintenance therapy” vs. “active IV drug use, heroin, last use today.”
The study found that details associated with clinical stigma often influenced how the AI systems prioritized patients. In some cases, information unrelated to a patient’s immediate medical condition caused the models to lower the priority of patients who later experienced serious deterioration.
Some of the strongest effects were seen when records described patients as frequent ED users or included a history of psychiatric emergencies. Those additions made the AI models more likely to deprioritize patients who were actually at higher risk of worsening illness, reflecting a similar pattern in human triage when patients labeled as “frequent flyers” receive lower priority. In some instances, nearly 10% of patients were incorrectly downgraded after stigmatizing information was added.
Andrew Jamieson, Ph.D., is Assistant Professor in the Lyda Hill Department of Bioinformatics at UT Southwestern.
“In this study, a single added detail sometimes shifted a model’s priority away from the patient who later deteriorated,” said the study’s senior author, Andrew Jamieson, Ph.D., Assistant Professor in the Lyda Hill Department of Bioinformatics.
Not every type of information affected the AI models in the same way. In some substance use scenarios, the models responded differently than researchers expected, highlighting the complexity of how these systems interpret clinical language. Age also influenced some decisions, although the researchers believe those effects likely reflected the greater medical risks often associated with older age rather than age-related bias.
“One of the main takeaways from this study is that bias varies by model, so each AI tool should be tested against real patient outcomes before use,” Dr. Jarrett said. The next step of this research is to develop models that predict clinical deterioration in the ED based on both structured data and clinical text, he added.
Overall, the findings suggest AI systems used to support emergency triage may be influenced not only by a patient’s medical condition, but also by how the patient’s history is described in clinical records. As hospitals increasingly explore the use of AI to help prioritize care, the authors conclude, the way patient information is documented could influence whether these systems make fair and accurate decisions.
“AI is expanding what we can do with complex information, and we’re still developing a scientific understanding of how these systems behave. A big part of that work is learning what shapes their answers, how those answers vary, and what that means in clinical practice,” said Dr. Jamieson, whose lab is focused on bridging the gap between research and patient care. “Clinical needs guide what we study, and what we learn helps improve the tools clinicians use to evaluate and solve real-world problems.”
Other researchers who contributed to this study are Peter Yun, M.D., Emmanuel Ohuabunwa, M.D., M.B.A., and Doreen Agboh, M.D., Assistant Professors of Emergency Medicine; and D. Mark Courtney, M.D., Professor and Executive Vice Chair of Academic Affairs in Emergency Medicine.
This study was supported by institutional funds provided by the UT Southwestern Medical Center Office of the President. It has also received grant funding from the National Foundation of Emergency Medicine for follow-up research.
About UT Southwestern Medical Center
UT Southwestern, one of the nation’s premier academic medical centers, integrates pioneering biomedical research with exceptional clinical care and education. The institution’s faculty members have received six Nobel Prizes and include 28 members of the National Academy of Sciences, 26 members of the National Academy of Medicine, and 14 Howard Hughes Medical Institute Investigators. The full-time faculty of nearly 3,400 is responsible for groundbreaking medical advances and is committed to translating science-driven research quickly to new clinical treatments. UT Southwestern physicians in more than 80 specialties care for more than 143,000 hospitalized patients, attend to more than 470,000 emergency room cases, and oversee nearly 5.3 million outpatient visits a year.
