The advanced machine learner XGBoost did not reduce prehospital trauma mistriage compared with logistic regression: a...

Area of Science:

• Emergency Medicine
• Data Science
• Trauma Surgery

Background:

• Accurate prehospital trauma triage is vital for patient care.
• Limited time and data in prehospital settings challenge triage model complexity.
• This study compares XGBoost and logistic regression for prehospital trauma mistriage.

Purpose of the Study:

• To assess if XGBoost reduces prehospital trauma mistriage compared to logistic regression.
• To evaluate the impact of training dataset size on model performance.

Main Methods:

• A simulation study using US National Trauma Data Bank (NTDB) and Swedish Trauma Registry (SweTrau) data.
• Predictors included systolic blood pressure, respiratory rate, Glasgow Coma Scale, and age.
• Outcome measured was the difference in undertriage and overtriage rates.

Main Results:

• XGBoost and logistic regression showed similar mistriage rates across datasets.
• XGBoost required larger training sets (up to 1000 events/parameter) for robust results in NTDB.
• Logistic regression achieved stable performance with smaller training sets (25 events/parameter) in NTDB.

Conclusions:

• XGBoost did not outperform logistic regression in reducing prehospital trauma mistriage.
• Logistic regression is recommended over XGBoost when predictors are few and categorical.
• Model performance is influenced by training dataset size and predictor characteristics.