摘要
Abstract
Objective To build a prediction model for the risk of hyponatremia in patients with traumatic brain injury(TBI)by using five machine learning algorithms,and screen the optimal prediction model.Methods Based on a questionnaire on risk factors for hyponatremia,a retrospective analysis was conducted on the clinical data of 906 patients with TBI who were hospitalized in Baoding NO.1 Central Hospital from January 2022 to December 2024.They were randomly stratified into training set and test set in an 8:2 ratio.A total of 199 inpatients with TBI from April to July 2025 were collected as validation group for time period validation.In training set,predictor variables were screened through univariate analysis,LASSO regression and multivariate Logistic regression,and five prediction models,namely Logistic regression,decision tree,random forest,support vector machine(SVM)and extreme gradient boosting were constructed.The optimal model was selected by comparing the indicators such as area under the curve(AUC)and Fl value of test set and validation group.Results The incidence of hyponatremia in patients with TBI was 32.78％.Eight variables,including age,gender,Glasgow coma score(GCS),cerebral contusion and laceration,subarachnoid hemorrhage,frontal injury,cerebral edema,and eating patterns were its predictive variables.Internal tests and external verifications showed that the SVM model outperforms other models in terms of recall rate,F1 value and other indicators.SHAP analysis showed that the GCS score made the greatest predictive contribution to hyponatremia in patients with TBI.Conclusion The SVM model has the best predictive efficacy for hyponatremia in patients with TBI and can assist medical staff in identifying high-risk patients at an early stage and providing targeted interventions.关键词
机器学习/颅脑损伤/低钠血症/预测模型/护理Key words
Machine learning/Traumatic brain injury/Hyponatremia/Prediction model/Nursing分类
医药卫生
