UT Southwestern finds AI triage models moved some emergency patients who later got worse down the line after one stigmatizing detail was added
UT Southwestern Medical Center reported on October 20th, 2026 a study published in npj Digital Medicine. Its researchers paired emergency patients of similar starting severity, using records from Stanford Medicine and Beth Israel Deaconess Medical Center. Three open source large language models were asked which patient was more likely to deteriorate and should be seen first. Before any social or stigmatizing detail was added, all three picked the higher risk patient at rates well above chance. A line describing frequent emergency visits or past psychiatric emergencies most often moved a model away from the patient who later deteriorated. In some instances nearly 10% of patients were wrongly downgraded. The authors say bias varies by model and that each tool should be tested against real patient outcomes before use. No Texas hospital is reported using these models for triage.
How to take part
The work is a published study rather than a proceeding, so no comment window or hearing attaches to it. The medical center names the first and senior authors on its release, with a newsroom phone line.
Where
1 county. In Dallas-Fort Worth-Arlington.
Timeline
- filed
Date the medical center published its account of the study
- Today
How this decision moved
One dated line per check, oldest first. A line that says nothing changed means somebody looked and it had not.
- 2026-10-23
Admitted on the medical center's own account of the study. It names no Texas hospital using these models for triage, and the records came from hospitals in California and Massachusetts.
The evidence
Every fact above rests on one of these. The words are the source's own.
researchers at UT Southwestern Medical Center reported in a study published in npj Digital MedicineUT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
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.UT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
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.UT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
Before demographic, social, or potentially stigmatizing information was added, all three models identified higher-risk patients at rates well above chance.UT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
Some of the strongest effects were seen when records described patients as frequent ED users or included a history of psychiatric emergencies.UT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
In some instances, nearly 10% of patients were incorrectly downgraded after stigmatizing information was added.UT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
bias varies by model, so each AI tool should be tested against real patient outcomes before useUT Southwestern Medical Center, Artificial intelligence-enabled triage tools may reflect human biases Primary source, official · utsouthwestern.edu
Questions about this decision
Answered from the record itself. Every answer is assembled from stored fields, so an answer the record has no basis for is left out rather than guessed.
What is this decision?
UT Southwestern Medical Center reported on October 20th, 2026 a study published in npj Digital Medicine. Its researchers paired emergency patients of similar starting severity, using records from Stanford Medicine and Beth Israel Deaconess Medical Center. Three open source large language models were asked which patient was more likely to deteriorate and should be seen first. Before any social or stigmatizing detail was added, all three picked the higher risk patient at rates well above chance. A line describing frequent emergency visits or past psychiatric emergencies most often moved a model away from the patient who later deteriorated. In some instances nearly 10% of patients were wrongly downgraded. The authors say bias varies by model and that each tool should be tested against real patient outcomes before use. No Texas hospital is reported using these models for triage.
Who decides it?
University of Texas Southwestern Medical Center decides. The record names the deciding body for every entry it carries.
Can the public take part?
The work is a published study rather than a proceeding, so no comment window or hearing attaches to it. The medical center names the first and senior authors on its release, with a newsroom phone line. No dated public window is on the record. The deciding body is named and reachable.
Where in Texas does it apply?
It covers Dallas County in the Dallas-Fort Worth-Arlington, TX area.
Has it been decided?
It has been decided. The dates on the item page carry when.
What happens next?
No future date is on the record. The last dated step on it was filed on October 20th.
When did it start?
The earliest date on its record is October 20th, 2026.
What kind of decision is it?
It is filed under health and education.
What sources back it?
One source backs it. It is primary.
Is it on the ERCOT grid?
Yes. It sits inside the ERCOT interconnection.
When was it last checked?
Every fact on it was last verified against its source on October 23rd, 2026.
Cite this
Texas AI Docket, UT Southwestern finds AI triage models moved some emergency patients who later got worse down the line after one stigmatizing detail was added. Tracked since October 20th, 2026. Last verified October 23rd, 2026. https://texasaidocket.com/item/tx-2026-0228/. Reuse permitted under CC BY 4.0 with attribution. The same entry is in the docket JSON as item tx-2026-0228.
Beat
Filed under Health and education, with every other decision on that beat.