Early mortality prediction after severe trauma using ensemble machine learning: a single-center retrospective study.

Area of Science:

• Emergency Medicine
• Data Science
• Critical Care

Background:

• Early identification of trauma patients at risk of in-hospital death is crucial for effective resuscitation and surgical planning.
• Machine learning offers potential for developing predictive models using routinely collected patient data.

Purpose of the Study:

• To develop and evaluate a machine learning framework for predicting in-hospital mortality in trauma patients.
• To assess the performance of various machine learning models and ensemble methods using early clinical data.

Main Methods:

• A retrospective study of 408 trauma patients using vital signs, laboratory, and blood-gas data within 30 minutes of emergency department arrival.
• Implementation and comparison of logistic regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, MLP, and ensemble (stacking, voting) models.
• Evaluation of model performance using Area Under the Receiver Operating Characteristic Curve (AUROC) and Area Under the Precision-Recall Curve (AUPRC).

Main Results:

• Ensemble models demonstrated high predictive performance, with the voting ensemble achieving an AUROC of 0.9506 and AUPRC of 0.8715.
• Key predictors identified by the stacking model included Injury Severity Score (ISS), base excess (BE), Glasgow Coma Scale (GCS), and pH.
• Individual models showed AUROC values ranging from 0.743 to 0.927.

Conclusions:

• Ensemble machine learning integrating early clinical data provides robust prediction of in-hospital mortality in severe trauma.
• The findings support the use of interpretable ensemble learning for early risk stratification in trauma patients.
• Injury Severity Score, Glasgow Coma Scale, and acid-base balance are significant contributors to mortality prediction.