Construction And Validation Of A Predictive Model For Nosocomial Infection After Neurosurgery

Objectives:To understand the current situation of nosocomial infection in patients after neurosurgery and analyze the risk factors of nosocomial infection.On this basis,the nomograph and decision tree risk prediction model of nosocomial infection in neurosurgery patients were constructed.And the two models are compared.Further deepen the understanding of researchers and medical staff on the prediction model of nosocomial infection,guide neurosurgery to further improve the content of nosocomial infection management,and timely find high-risk patients with nosocomial infection.Methods:The patients who received surgical treatment in neurosurgery of a hospital in Jinan from January 2020 to January 2021 were collected retrospectively.With the help of nosocomial infection surveillance system and hospital information system,eligible patients are screened according to inclusion and exclusion criteria.The self-designed "targeted surveillance questionnaire for nosocomial infection in patients after neurosurgery" was used to collect relevant research data.Subjects were randomly divided into two groups(prediction model building group and prediction model verification group)by 7:3 grouping method.The variables with statistical difference(P<0.05)in univariate analysis in the model establishment group were included in multivariate regression analysis,and the results were displayed by nomogram.The internal and external validation of the constructed prediction model is carried out respectively.AUC,Hosmer lemeshow test and calibration curve evaluate the discrimination and calibration of the model.DCA evaluates the clinical effects of the model.At the same time,C5.0 algorithm was used to construct the risk prediction model of decision tree for the above variables,and external verification was carried out.AUC and error classification table were used to evaluate the prediction effect of the model.Finally,according to AUC,the prediction effects of the two prediction models were analyzed,and compared to select the optimal model for clinical use.Results:1.A total of 1508 neurosurgical patients were included in this study.The average age of the patients was 51.21±12.697 years.Among them,there were 654 male patients and 854 female patients.After randomization,there were 1058 patients in the modeling group,with an average age of 51.28±12.610 years.There were 450 patients in the validation group,with an average age of 51.03±12.912 years.2.In this study,a total of 231 patients had postoperative nosocomial infection.The postoperative nosocomial infection rate was 15.32%,and the case rate of nosocomial infection was 16.45%.The postoperative nosocomial infection rate of the modeling group was 13.71%,and the case rate of nosocomial infection was 15.86%.The postoperative nosocomial infection rate in the validation group was 19.11%,and the nosocomial infection rate was 17.77%.3.The most common type of nosocomial infection after neurosurgery in this study is surgical site infection,with a constituent ratio of 78.92%.The other types of nosocomial infection are lower respiratory tract infection(constituent ratio is 12.80%),urinary tract infection(constituent ratio is 3.31%)and blood flow infection(constituent ratio is 0.41%).4.In this study,39 strains of pathogens were cultured,including 25 Gram-positive bacteria,13 Gram-negative bacteria and 1 fungus.The constituent ratios were 64.10%,33.33%and 2.56%respectively.5.Nomogram prediction model showed that craniocerebral tumor,craniotomy,operation time(≥4 hours),preoperative body temperature(℃),mechanical ventilation,indwelling catheter time(≥4 days),preventive antibiotics and hormones were independently related to nosocomial infection after neurosurgery.The internal and external validation of the model shows good prediction effect.The evaluation results of AUC,Hosmer lemeshow test and calibration curve show that the discrimination and calibration degree of the model are good.DCA showed that the model had good clinical benefits.6.In the decision tree model,a total of 9 variables enter each node of the model,namely endotracheal intubation,preventive antibiotics,operation time(h),indwelling urinary catheter time(day),brain tumor,preoperative body temperature(℃),and preoperative leukocyte count(×109/L),preoperative hospital stay(day),intraoperative blood transfusion.A total of 15 decision rules are generated.The external verification results show that the decision tree model is stable,and AUC and error classification table suggest that the model has good prediction effect.7.The AUC of nomogram model is greater than that of decision tree model,and there is a statistical difference.Conclusions:Postoperative nosocomial infection is common in neurosurgery.Hormone,prophylactic antibiotics,operation time(≥ 4 hours),preoperative body temperature(℃),craniocerebral tumor and indwelling catheter time(≥4 days)are independently related to postoperative nosocomial infection,which is an important entry point for clinical formulation of relevant preventive measures.Nomogram and decision tree model predict the infection of patients from different angles,and both show good prediction efficiency.The comprehensive use of the two tools can improve the prediction accuracy.