Carbapenem-resistant gram-negative bacterial infection in intensive care unit patients: Antibiotic resistance analysis and predictive model development

Abstract

In this study, we analyzed the antibiotic resistance of carbapenem-resistant gram-negative bacteria (CR-GNB) in intensive care unit (ICU) patients and developed a predictive model. We retrospectively collected the data of patients with GNB infection admitted to the ICU of the First Affiliated Hospital of Fujian Medical University, who were then divided into a CR and a carbapenem-susceptible (CS) group for CR-GNB infection analysis. Patients admitted between December 1, 2017, and July 31, 2019, were assigned to the experimental cohort (n = 205), and their data were subjected to multivariate logistic regression analysis to identify independent risk factors for constructing the nomogram-based predictive model. Patients admitted between August 1, 2019, and September 1, 2020, were assigned to the validation cohort for validating the predictive model (n = 104). The Hosmer-Lemeshow test and receiver operating characteristic (ROC) curve analysis were used to validate the model's performance. Overall, 309 patients with GNB infection were recruited. Of them, 97 and 212 were infected with CS-GNB and CR-GNB, respectively. Carbapenem-resistant Klebsiella pneumoniae (CRKP), carbapenem-resistant Acinetobacter baumannii (CRAB) and carbapenem-resistant Pseudomonas aeruginosa (CRPA) were the most prevalent CR-GNB. The multivariate logistic regression analysis results of the experimental cohort revealed that a history of combination antibiotic treatments (OR: 3.197, 95% CI: 1.561-6.549), hospital-acquired infection (OR: 3.563, 95% CI: 1.062-11.959) and mechanical ventilation ≥ 7 days (OR: 5.096, 95% CI: 1.865-13.923) were independent risk factors for CR-GNB infection, which were then used for nomogram construction. The model demonstrated a good fit of observed data (p = 0.999), with an area under the ROC curve (AUC) of 0.753 (95% CI: 0.685-0.820) and 0.718 (95% CI: 0.619-0.816) for the experimental and validation cohort, respectively. The decision curve analysis results suggested that the model has a high practical value for clinical practice. The Hosmer-Lemeshow test indicated a good fit of the model in the validation cohort (p-value, 0.278). Overall, our proposed predictive model exhibited a good predictive value in identifying patients at high risk of developing CR-GNB infection in the ICU and could be used to guide preventive and treatment measures.

Keywords: area under the receiver operating characteristic curve; carbapenem-resistant gram-negative bacteria; logistic regression; predictive model; risk factor.

Figure 1

Nomogram for estimating the probability of CR-GNB infection in ICU patients. The scores of various independent variables were as follows: combination antibiotic treatment: 7.1 points, hospital-acquired infection: 7.8 points, and duration of mechanical ventilation ≥7 days: 10.0 points. The scores of the independent variables were summed to obtain the overall score, which can be used to predict the probability of CR-GNB infection in ICU patients. For instance, the predicted probability of CR-GNB infection occurrence in ICU patients was >80% when the total score was >16 points.

Figure 2

Hosmer−Lemeshow test provided a χ² value of 0.07 (p = 0.999) for the predictive model in the experimental cohort, suggesting that the model presented a good fit for the observed data.

Figure 3

Receiver operating characteristic (ROC) curve of the predictive model in the experimental cohort. The predictive model had a good discriminative power with an area under the ROC curve of 0.753 (95% confidence interval: 0.685–0.820).

Figure 4

Decision curve analysis (DCA) curves of the predictive model in the experimental cohort. The DCA curve showed that the maximum clinical benefit could be obtained with the model when the predicted probability was >0.7.

Figure 5

Hosmer−Lemeshow plot of the predictive model for the validation cohort. Assessment of the goodness of fit using the Hosmer−Lemeshow test showed an χ² value of 5.09 and a p-value of 0.278, suggesting that the model presented a good fit in the validation cohort.

Figure 6

Receiver operating characteristic (ROC) curve of the predictive model for the validation cohort. The area under the ROC curve was 0.718 (95% confidence interval: 0.619–0.816), indicating a good prediction performance of the model in the validation cohort.

Figure 7

Decision curve analysis (DCA) curves of the predictive model for the validation cohort. The DCA curve showed that the model provided a high practical value in clinical practice when the predicted probability was > 0.75 in the validation cohort.