A Prehospital Triage System to Detect Traumatic Intracranial Hemorrhage Using Machine Learning Algorithms

Area of Science:

• Emergency Medicine
• Neurosurgery
• Artificial Intelligence in Healthcare

Background:

• Effective prehospital triage for head trauma is crucial for patient prognosis.
• Current methods lack severity stratification tools for ambulance crews.
• Optimal medical institution selection is vital for improving outcomes.

Purpose of the Study:

• To develop and evaluate a machine learning-based prehospital triage system for head trauma severity.
• To stratify patients using machine learning techniques for improved prehospital decision-making.
• To assess the predictive accuracy of machine learning models in identifying traumatic intracranial hemorrhage.

Main Methods:

• Retrospective cohort study of 2123 head trauma patients.
• Development of machine learning models, including extreme gradient boosting (XGBoost), to predict traumatic intracranial hemorrhage.
• Evaluation of model performance using ROC-AUC, PR-AUC, sensitivity, and specificity.

Main Results:

• XGBoost achieved the highest predictive accuracy with ROC-AUC of 0.78 and PR-AUC of 0.46 in cross-validation.
• In the testing set, the XGBoost model demonstrated an ROC-AUC of 0.80, sensitivity of 74.0%, and specificity of 74.9%.
• The machine learning model showed comparable performance to the National Institute for Health and Care Excellence (NICE) guidelines.

Conclusions:

• Machine learning models can accurately predict traumatic intracranial hemorrhage using prehospital data.
• The developed model offers a valuable tool for prehospital triage and hospital selection in head trauma cases.
• Further validation in prospective, multicenter studies is recommended to confirm findings.