Study On Prediction Model Of Secondary Central Nervous System Infection After Craniotomy Based On Machine Learning

Objective:By analyzing the risk factors of central nervous system infection after neurosurgical craniotomy,an individual model for predicting the risk of CNSIs was established.Methods:A total of 1599 patients who underwent craniotomy in the Department of Neurosurgery of the affiliated Hospital of North Sichuan Medical College from January 2019 to December 2020 were selected as the study object.1599 patients with postoperative CNSIs were divided into the case group and the control group randomly.The influencing factors of secondary CNSIs after craniotomy were determined by literature review,and the independent risk factors for modeling were screened out by univariate analysis and multi-factor Logistic regression.The modeling variables were inputted into the model,and the prediction models of secondary CNSIs after craniotomy were established by using Logistic regression,naive Bayesian,random forest,LightGBM and Adaboost algorithms based on machine learning.The prediction performance of the five models was evaluated by accuracy,accuracy,recall rate,F1 value and the area under ROC curve,and the clinical application effect of the model was verified.Results:Univariate analysis showed gender,alcohol consumption,GCS score,emergency operation,number of operations after admission,operation time,intraoperative blood loss,intraoperative use of microscope,indwelling lumbar cistern drainage tube,indwelling time of lumbar cistern drainage tube,indwelling epidural drainage tube during operation,postoperative incision effusion,number of indwelling drainage tube during operation,indwelling ventricular drainage tube,postoperative cerebrospinal fluid leakage,endotracheal intubation,postoperative tracheotomy,postoperative albumin content,there was significant difference in staying in ICU(P<0.1).The results of multivariate Logistic regression analysis showed that gender,GCS score,operation time,indwelling lumbar cistern drainage tube,indwelling lumbar cistern drainage tube,indwelling epidural drainage tube,indwelling ventricular drainage tube,emergency operation and the number of operations after admission were independent risk factors for secondary CNS Is after neurosurgical craniocerebral surgery(P<0.05).The prediction model based on Adaboost algorithm has better prediction performance than the other four models.Under 50%discount cross-validation,the accuracy,precision,recall,Fl-score and area under the ROC curve are 0.80,0.69,0.85,0.76 and 0.897 respectively.The top three important variables of Adaboost model are operation time,indwelling time of lumbar cistern drainage tube and indwelling lumbar cistern drainage tube during operation.In addition,the Adaboost model with the best prediction performance was used for clinical verification,and the prediction results were compared with the discharge diagnosis of patients.The results showed that the prediction accuracy of Adaboost model for occurrence of CNSIs was 60%,for non-occurrence of CNSIs was 92%,and the overall prediction accuracy was 76%.Conclusion:1.Gender,GCS score,operation time,indwelling lumbar cistern drainage tube,indwelling lumbar cistern drainage tube,indwelling epidural drainage tube,indwelling ventricular drainage tube,emergency operation and the number of operations after admission were independent risk factors for secondary CNSIs after craniotomy.2.In this study,the Adaboost model based on machine learning has a good ability to predict secondary CNSIs in patients after craniotomy,can guide clinical individualized prevention and control of secondary CNSIs in patients after craniotomy,and has high clinical application value.