Enhancing Performance of the National Field Triage Guidelines Using Machine Learning: Development of a Prehospital...

Area of Science:

• Emergency Medicine
• Data Science in Healthcare
• Trauma Surgery

Background:

• Prehospital trauma triage is critical for directing severe trauma patients to appropriate care.
• Current national field triage guidelines by the American College of Surgeons demonstrate limited sensitivity in identifying severe injuries.
• There is a need for improved prehospital triage tools to enhance patient outcomes.

Purpose of the Study:

• To develop an advanced prehospital triage model for predicting severe trauma.
• To improve the accuracy and effectiveness of existing national field triage guidelines.
• To reduce undertriage rates in severe trauma patients.

Main Methods:

• A multisite prediction study utilizing data from the National Trauma Data Bank (2017-2019).
• Inclusion of patients aged 16+ transported by ambulance from the scene.
• Development of a model using extreme gradient boosting and Shapley additive explanation analysis, with data split into training, internal, and external validation sets.

Main Results:

• The model achieved a sensitivity of 0.799 for severe trauma prediction (Injured Severity Score ≥16) with an undertriage rate of 0.080.
• It demonstrated superior performance compared to the Glasgow Coma Score, Prehospital Index, revised trauma score, and national field triage guidelines RED criteria.
• Model performance was consistent across internal and external validation sets, indicating robustness.

Conclusions:

• The developed prehospital triage model shows significant promise for accurately predicting severe trauma.
• The model can achieve an undertriage rate below 10%, enhancing patient safety.
• Machine learning techniques effectively augment the performance of current prehospital field triage guidelines.