ABSTRACT
Objective@#To create a model to predict nosocomial infections in emergency intensive care units (EICU), so as to provide insights into early identification and interventions among patients with nosocomial infections. @*Methods@#All nosocomial infections were collected from patients hospitalized in the EICU of a large tertiary hospital from 2017 to 2020. The 2017-2019 data were selected as the training set to create a logistic regression model, and the fitting effectiveness of the predictive model was evaluated using Hosmer-Lemeshow test. The 2020 data were selected as the test set to evaluate the external validation of the predictive model. In addition, the value of the model for prediction of nosocomial infections was examined using the receiver operating characteristic (ROC) curve analysis. @*Results @#Totally 1 546 inpatients in EICU were enrolled, and the prevalence of nosocomial infections was 7.18%. Multivariable logistic regression analysis identified hospital stay duration of >7 days (OR=21.845, 95%CI: 7.901-60.398), use of ventilators (OR=3.405, 95%CI: 1.335-8.682), and surgery (OR=1.854, 95%CI: 1.121-3.064) as risk factors of nosocomial infections. The predictive model was p=ey/(1+ey), y=-6.105+(3.084×duration of hospital stay)+(1.225×use of ventilators)+(0.617×surgery). The area under ROC curve was 0.806 (95%CI: 0.774-0.838) for the training set and 0.723 (95%CI: 0.623-0.823) for the test set, and if the 0.065 cut-off of the predictive model created by the training set was included in the test set, the predictive value yield a 0.739 sensitivity and 0.642 specificity for prediction of nosocomial infections among patients hospitalized in EICU. @*Conclusion@#The created predictive model for nosocomial infections among patients hospitalized in EICU presents a high accuracy, which shows a satisfactory predictive value for high-risk nosocomial infections.