Risk factors for carbapenem-resistant Klebsiella pneumoniae infection in ICU-hospitalized patients and its predictive value of BP neural network
OBJECTIVE To investigate the risk factors for carbapenem-resistant Klebsiella pneumoniae infection in ICU-hospitalized patients and its predictive value of BP neural network and logistic regression models.METHODS Information on patients admitted to the ICU of a military tertiary hospital from Apr.1,2019 to Apr.1,2023 was collected,including demographic characteristics,history of underlying diseases,history of antimicrobial drug ap-plication,types of antimicrobial drug,cumulative days of antimicrobial drug use,history of invasive operation,and coexisted fungal infection.Among them,137 cases of CRKP nosocomial infection were defined as CRKP group,and 381 patients with Klebsiella pneumoniae(CSKP)infections who were sensitive to carbapenems anti-microbial drugs were selected according to the case-control study method in 1:3 ratio,and were defined as CSKP group.The CRKP infection prediction model was constructed using artificial neural network and logistic regression algorithm,and the accuracy of the data and prediction model was evaluated using the receiver operating character-istic(ROC)curve.RESULTS From April 2019 to April 2023,a total of 1 507 strains of Klebsiella pneumoniae were isolated from ICU,of which 338 were CRKP,with an overall isolation rate of 22.43％.The annual isolation rates were 26.83％,22.55％,19.30％,and 19.05％in order.The annual incidences of CRKP nosocomial infec-tion were 3.79‰,3.64‰,2.91‰,and 3.84‰ respectively.Logistic multivariate regression analysis showed that malignant tumor,surgery during hospitalization,indwelling urethral catheter,use of carbapenem antibiotics,use of glycyl tetracycline antibiotics,and ventilator use time were the risk factor for CRKP nosocomial infection(P＜0.05).The BP neural network model structure was {30-6-1},and its prediction accuracy was 80％.The area un-der the ROC curve of the logistic regression model was 0.781,and the area under the ROC curve of the BP neural network model was 0.786,with a model prediction accuracy of 80％.CONCLUSION The CRKP detection rate showed consistent performance with the overall trend of the national CRKP detection rate,and the accuracy of the two prediction models was better,with the prediction performance of the BP neural network model slightly better than that of the logistic regression model.
NETL
NSTL
万方数据
