基于机器学习的综合预测模型用于评估神经外科术后细菌性脑膜炎的风险OA

Objective To integrate clinical and cerebrospinal fluid(CSF)features using machine-learning algorithms to develop and validate a prediction model for early identification of high-risk patients with postoperative bacterial meningitis after neurosurgery.Methods A total of 196 patients with suspected postoperative bacterial meningitis after craniocerebral surgery admitted to the intensive care unit of Kunshan First People's Hospital Affiliated to Jiangsu University from December 2020 to December 2024 were enrolled retrospectively.Demographic characteristics,comorbidities,smoking and alcohol history,and CSF parameters were collected and preprocessed using blinded procedures,encoding,and standardization.Models were developed using Extreme Gradient Boosting(XGB),Random Forest(RF),Multilayer Perceptron(MLP),and Logistic Regression(LR).Patients were randomly split into a training set(n=118)and a testing set(n=78)at a 6∶4 ratio.SMOTE was applied to the training set to address class imbalance.Model performance was primarily assessed based on the area under curve(AUC)of receiver operating characteristic(ROC)and further evaluated using accuracy,F1-score,area under the precision-recall curve(AUPRC),and Brier score.Results Significant differences between the bacterial meningitis and non-bacterial meningitis groups were observed in comorbidities and lifestyle factors(hypertension,diabetes,smoking,and alcohol consumption),as well as CSF indices(lactate,fibronectin,white blood cells,protein,and glucose).Multivariable analysis identified eight independent predictors,including hypertension,smoking,alcohol consumption,CSF lactate,CSF fibronectin,CSF white blood cells,CSF protein,and CSF glucose.Among the four models,XGB achieved the best performance,with an AUC of 0.932 in the training set and an AUC of 0.968 in the testing set;the testing-set accuracy was 0.906,F1-score 0.916,AUPRC 0.989,and Brier score 0.07.Conclusions Comorbidities and unhealthy lifestyle habits were closely associated with postoperative bacterial meningitis.The XGB model incorporating the above clinical and CSF features demonstrated good early discriminatory performance and may support postoperative infection risk stratification and clinical decision-making.