Clinical Report: Triage Model Utilizing Machine Learning for Older Adults with Traumatic Brain Injury in Chinese Emergency Settings
Overview
This study presents an XGBoost-based triage model designed to optimize ICU allocation for elderly patients with traumatic brain injury (TBI). The model demonstrated high predictive accuracy, achieving an AUC of 0.93, and highlights the importance of integrating multidimensional data for effective triage in emergency settings.
Background
Elderly patients with traumatic brain injury present unique challenges in emergency departments due to their complex physiology and comorbidities. Traditional triage methods often fail to accurately identify high-risk cases, leading to potential over- or under-treatment. As the incidence of TBI in older adults rises, developing reliable triage models is crucial for optimizing resource allocation and improving patient outcomes.
Data Highlights
Metric
Value
AUC
0.93
Recall
0.9288
Precision
0.9274
F1 Score
0.9249
MCC
0.7241
Key Findings
The XGBoost model achieved an AUC of 0.93 for predicting ICU triage disposition.
Key predictors included symptoms, CT hematoma density, and contusion severity.
The model outperformed traditional triage methods, indicating its potential utility in clinical settings.
Decision curve analysis suggested a higher theoretical net benefit compared to standard triage strategies.
Further validation is necessary to confirm the model's effectiveness in reducing unnecessary ICU admissions.
Clinical Implications
The XGBoost triage model offers a promising tool for emergency physicians to enhance decision-making regarding ICU admissions for elderly TBI patients. Its ability to integrate complex data may lead to more accurate risk stratification and better resource allocation in emergency departments.
Conclusion
This study underscores the potential of machine learning models in improving triage processes for elderly TBI patients. However, prospective validation across multiple centers is essential to establish clinical utility and patient-centered outcomes.
