Automating Injury Severity Score Calculation Using Large Language Models: A Feasibility Study With Large Language...

Area of Science:

• Medical informatics
• Trauma surgery
• Artificial intelligence in healthcare

Background:

• The Injury Severity Score (ISS) is vital for trauma assessment but manual calculation is error-prone and time-consuming.
• Current methods rely on manual scoring by registrars, leading to potential inaccuracies and delays.

Purpose of the Study:

• To evaluate the feasibility of using a large language model (LLM) for automated ISS calculation.
• To assess the accuracy and reliability of LLM-assisted trauma scoring compared to manual methods.

Main Methods:

• A retrospective study at a level I trauma center using 2022 patient data.
• LLM trained with structured prompts on trauma scoring principles.
• Validation using 100 cases, comparing LLM-generated ISS with registrar-calculated ISS via Pearson correlation, ICC, and Bland-Altman analysis.

Main Results:

• High agreement between LLM ISS and registrar ISS (ICC=0.981).
• LLM demonstrated high accuracy (0.91) with minimal mean bias (-0.03) in Bland-Altman analysis.
• Consistent performance across different ISS ranges.

Conclusions:

• LLM-generated ISS is a reliable and accurate automated method for trauma scoring.
• Potential to streamline clinical workflows and reduce human error in trauma assessment.
• Future research should focus on real-time integration and application to other scoring systems.