Researchers Fatimah D. Albalawi, Reem Alotaibi, S. Alshammari and colleagues evaluated GPT-5.2 for emergency triage in simulated Hajj scenarios, reporting a 79.7% relative reduction in under-triage across Emergency Severity Index (ESI) levels. The study, published in Diagnostics in 2026, used 1,000 balanced and stratified synthetic clinical dialogues to assess whether narrative information could improve triage decisions under controlled conditions.

In high-acuity cases, adding clinical dialogue was associated with a 91.9% relative reduction in under-triage, while GPT-5.2 achieved 94.5% recall for high-acuity cases. The researchers said the dialogues helped identify critical “red-flag” symptoms that structured data may not capture. ESI is a standardized triage protocol, but its use depends on human judgment.

The study frames the results as an evaluation of a conceptual clinical decision-support approach combining patient narratives with structured physiological indicators and electronic health records. The authors conclude that large language model systems may improve triage reliability in high-demand clinical environments, while describing the evaluation as a controlled decision-support setting.

Researchers Fatimah D. Albalawi, Reem Alotaibi, S. Alshammari and colleagues evaluated GPT-5.2 for emergency triage in simulated Hajj scenarios, reporting a 79.7% relative reduction in under-triage across Emergency Severity Index (ESI) levels. The study, published in Diagnostics in 2026, used 1,000 balanced and stratified synthetic clinical dialogues to assess whether narrative information could improve triage decisions under controlled conditions.

In high-acuity cases, adding clinical dialogue was associated with a 91.9% relative reduction in under-triage, while GPT-5.2 achieved 94.5% recall for high-acuity cases. The researchers said the dialogues helped identify critical “red-flag” symptoms that structured data may not capture. ESI is a standardized triage protocol, but its use depends on human judgment.

The study frames the results as an evaluation of a conceptual clinical decision-support approach combining patient narratives with structured physiological indicators and electronic health records. The authors conclude that large language model systems may improve triage reliability in high-demand clinical environments, while describing the evaluation as a controlled decision-support setting.