To construct and validate an intelligent triage model for elderly TBI patients using machine learning techniques based on real-world emergency department data, addressing significant triage challenges.
Key Findings:
The XGBoost model achieved an AUC of 0.93, recall of 0.9288, precision of 0.9274, F1 score of 0.9249, and MCC of 0.7241, outperforming comparator models.
Key predictors included symptoms, CT hematoma density, and contusion severity.
Decision curve analysis indicated a higher theoretical net benefit compared to 'treat all' and 'treat none' strategies.
Interpretation:
The XGBoost model serves as an interpretable tool for predicting ICU triage disposition in elderly TBI patients, potentially aiding emergency physician decision-making and improving patient outcomes.
Limitations:
The study's endpoint was a single-center process-of-care surrogate, necessitating further prospective and multicenter validation against patient-centered outcomes.
Validation against outcomes such as mortality, neurological deterioration, and functional status is required.
Conclusion:
The model may support triage decisions but should not replace clinical judgment; validation against real-world outcomes is necessary to ensure its applicability.
