An Evaluation on the Potential of Large Language Models for Use in Trauma Triage

Large Language Models (LLMs) are becoming increasingly adopted in various industries worldwide. In particular, there is emerging research assessing the reliability of LLMs, such as ChatGPT, in performing triaging decisions in emergent settings. A unique aspect of emergency triaging is the process of trauma triaging. This process requires judicious consideration of mechanism of injury, severity of injury, patient stability, logistics of location and type of transport in order to ensure trauma patients have access to appropriate and timely trauma care. Current issues of overtriage and undertriage highlight the potential for the use of LLMs as a complementary tool to assist in more accurate triaging of the trauma patient. Despite this, there remains a gap in the literature surrounding the utility of LLMs in the trauma triaging process. This narrative review explores the current evidence for the potential for implementation of LLMs in trauma triaging. Overall, the literature highlights multifaceted applications of LLMs, especially in emergency trauma settings, albeit with clear limitations and ethical considerations, such as artificial hallucinations, biased outputs and data privacy issues. There remains room for more rigorous research into refining the consistency and capabilities of LLMs, ensuring their effective integration in real-world trauma triaging to improve patient outcomes and resource utilisation.