BACKGROUND
Traumatic brain injury (TBI) remains a critical public health challenge. Although studies have found several prognostic factors for TBI, a useful early predictive model for mortality has yet to be developed for TBI patients in the emergency room.
OBJECTIVE
The objective of this study was to use artificial intelligence (AI) and machine learning algorithms to develop predictive models for TBI patients in the emergency room triage. This could provide scientific data for healthcare providers which they could use as a reference when deciding which treatment to give and when informing and educating patient’s family members.
METHODS
From January 2010 to December 2019, this study retrospectively enrolled 18,249 TBI patients (9908 males and 8341 females; mean age: 57.85 ± 19.44 years) in the electronic medical records of three Chi-Mei Medical Centers, and investigated the 12 potentially predictive feature variables. Mortality during hospitalization was designated as the outcome variable. The correlation coefficient matrix was used to analyze the feature variables and mortality using Spearman rank order correlation methods. Further, the present study constructed six machine learning models including logistic regression (LR) random forest (RF), support vector machines (SVM), Light GBM, XGBoost and Multilayer Perceptron (MLP) to predict mortality risk. Next, following the model training and building, we conducted area under the receiver operating characteristic curve (AUC) for six models performance evaluation. Finally, we deployed and installed the model in the hospital information system for clinical practice in the triage setting.
RESULTS
The results showed that all six predictive models had high AUC from 0.851 to 0.925. Among these predictive models, LR-based model was the best model for mortality risk prediction with sensitivity of 0.812, specificity of 0.894, and accuracy of 0.89 for the 12 feature variables; thus, this was used to develop an application to assist in clinical decision making.
CONCLUSIONS
These results revealed that the LR model was the best model to predict the mortality risk in patients with TBI in the emergency room. Since the developed AI system can easily obtain the 12 feature variables during the initial triage, it can provide quick outcome prediction to clinicians to help them explain the patient’s condition to family members and to guide them in deciding further treatment.
Geographic Disparities in Mortality Risk Within a Racially Diverse Sample of U.S. Veterans with Traumatic Brain Injury
