Identification and validation of an explainable prediction model for early neurological deterioration in moderate traumatic brain injury:a multicenter retrospective exploratory study
YANG Xiaoliang
ZHANG Rongjun
WANG Xiaofeng
LONG Qianfa
CAO Weidong
ZHAO Ming
ZHANG Honbin
ZHANG Min
QIAO Yu
YU Pengbo
WANG Weixi
XU Fan
WANG Yan
GUI Yinan
Abstract：Objective To integrate clinical and imaging features of patients with moderate traumatic brain injury and develop a prediction model for early neurological deterioration,so as to enhance risk assessment accuracy and provide a basis for early intervention.Methods A multicenter retrospective exploratory study was employed on patients with moderate traumatic brain injury admitted to the 987th Hospital of the Joint Logistics Support Force,Baoji Hospital of Traditional Chinese Medicine,and the Affiliated Xi'an Central Hospital of Xi'an Jiaotong University between January 2019 and July 2025.Nine machine learning methods-Logistic Regression,Decision Tree,Random Forest,Support Vector Machine,Neural Network,Extreme Gradient Boosting(XGBoost),Naive Bayes,Gradient Boosting Machine(GBM),and KNearest Neighbors(KNN)-were used to construct prediction models.Indicators such as the neutrophil-to-lymphocyte ratio,triglyceride-glucose index,venous-to-fingertip blood glucose ratio,and systemic inflammatory response index were calculated.Model performance was evaluated using the area under the curve(AUC),sensitivity,specificity,F1score,Youden index,decision curve analysis,and calibration curves.The Shapley Additive Explanations(SHAP)method was applied to interpret feature importance.Results A total of 160 patients were included,with 115 in the training set(60 in the deterioration group,55 in the non-deterioration group)and 45 in the external validation set(10 in the deterioration group,35 in the non-deterioration group).Among the nine machine learning models,Random Forest performed the best,achieving a training set AUC of 0.819 and an external validation AUC of 0.731.SHAP analysis further revealed the contribution of each feature to the prediction outcome.Conclusions This study presents a preliminary prediction model for early neurological deterioration in patients with moderate traumatic brain injury.The model shows effective predictive capability and provides an important reference for future large-scale clinical research and application.
Keywords：moderate traumatic brain injuryneurological deteriorationmachine learningprediction modelSHAP
Publication Date：2026-03-28
Online Publishing Date：2026-03-26(First online date of this platform, not the publication date of the document)
Pages：13( 22-34 )