Development and validation of a nomogram model for predicting MDRO infections in elderly ICU patients with pulmonary infections

Abstract

Background: MDRO infections are increasingly problematic in ICUs, especially among elderly patients with lung infections, but knowledge about these infections in this group is limited. This study aimed to assess the status and risk factors of MDRO infections in elderly ICU patients and develop a risk prediction model to aid clinical decisions.

Methods: Using a retrospective cohort study, a total of 494 elderly patients with lung infections admitted to the ICU from January 2017 to December 2022 were selected, and the patients were divided into the MDRO group (259) and the non-MDRO group (235) based on whether or not the patients developed MDRO infections. Lasso and multifactorial logistic regression were applied to analyze the independent risk factors for multidrug-resistant bacterial infections in elderly patients with pulmonary infections, and to construct a nomogram model of the risk of MDRO infections. The differentiation, consistency and clinical benefit of the model were evaluated by receiver operating characteristic curve(ROC), calibration curves and decision curve analysis, respectively, and the stability of the model was verified by Bootstrap method.

Results: Duration of hospitalization before MDRO diagnosis, chronic obstructive pulmonary disease, personal history of cerebrovascular disease, tracheotomy and prior carbapenem exposure were found to be independent risk factors for multidrug-resistant bacterial infections in elderly patients with pulmonary infections in the intensive care unit (all p < 0.05). The nomogram model, constructed based on the results of logistic regression analysis, exhibited an area under the ROC curve of 0.748 with a 95% confidence interval of 0.705-0.790. The Hosmer-Lemeshow test indicated that the model predicted a good fit (p = 0.75), and the DCA curve suggested that the model had a good clinical utility.

Conclusion: Risk prediction model is effective in predicting the risk of MDRO infection in the ICU elderly pulmonary infection population and can be used to assess risk and inform preventive treatment and nursing interventions.

Keywords: Elderly people; Multidrug resistant bacteria; Risk factors.

Fig. 1

Study population flowchart

Fig. 2

LASSO regression with ten-fold cross-validation was used to select predictive variables. (a) The best lambda for LASSO regression was found by analyzing deviance using the minimum (left dotted line) and 1-SE criteria (right dotted line). (b) A plot of coefficient profiles versus the log of lambda values was created. Using the 1-SE criterion, five variables with non-zero coefficients were selected

Fig. 3

Nomogram for the prediction of multidrug-resistant bacterial infections in elderly patients with lung infections. COPD mean chronic obstructive pulmonary disease, MDRO multidrug resistant organisms, LOS mean length of hospital stay

Fig. 4

ROC curve for the predictive model and internal validation. (a) displays the model’s AUC, and (b) presents the internal validation AUC via bootstrap. The gray area indicates ROC curves from 1,000 bootstrap cross-validations

Fig. 5

Calibration plot for cohort nomogram prediction. The x-axis represents the nomogram’s predicted probability, while the y-axis represents its actual probability

Fig. 6

Decision curve analysis for the nomogram