A Machine Learning Trauma Triage Model for Critical Care Transport

Area of Science:

• Prehospital emergency medicine
• Machine learning applications in healthcare
• Trauma patient triage

Background:

• Prehospital triage is critical for emergency care, but limited resources hinder accurate patient classification.
• Developing effective triage tools is essential for optimizing the use of limited resources in austere prehospital environments.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for prehospital triage of trauma patients.
• The model aims to predict the need for lifesaving interventions (LSIs) using continuous physiological waveform signals and vital sign patterns.

Main Methods:

• Retrospective cohort study of critically ill trauma patients transported by air ambulance (Jan 2018-Nov 2021).
• Analysis of physiological waveform signals and vital sign patterns within the first 15 minutes of care.
• Ensemble ML approach used to predict LSI occurrence based on preceding physiological features.

Main Results:

• The ML model demonstrated good performance in predicting overall LSI (AUC 0.810).
• Key metrics included high specificity (0.960) and negative predictive value (0.953).
• Model performance was robust across various LSI subcategories and patient subgroups.

Conclusions:

• An ML-based triage model accurately predicts lifesaving intervention needs in prehospital trauma patients.
• This AI-driven approach can streamline and enhance prehospital triage, improving patient outcomes.
• The findings support the deployment of ML modeling in field settings for critical care transport.