Establishment and Validation of a Risk Prediction Model for Mortality in Patients with Acinetobacter baumannii Infection: A Retrospective Study

Abstract

Purpose

This study aims to establish a valuable risk prediction model for mortality in patients with Acinetobacter baumannii ( A. baumannii).

Patients and Methods

The 622 patients with A. baumannii infection from the First Affiliated Hospital of Anhui Medical University were enrolled as the study cohort. Univariate and multivariate logistic regression analysis was used to preliminarily screen the independent risk factors of death caused by A. baumannii infection, followed by LASSO regression analysis to determine the risk factors. According to the calculated regression coefficient, the Nomogram death prediction model is established. The area under the curve (AUC) and decision curve analysis (DCA) of the operating characteristic (ROC) curve of the subjects are used to evaluate the discrimination of the established prediction model. The calibration degree of the prediction model is represented by a calibration chart. A validation cohort that consisted of 477 patients admitted to the 901st Hospital was also included.

Results

Our results revealed that the source of infection, carbapenem-resistant A. baumannii, mechanical ventilation, serum albumin value, and Charlson comorbidity index were independent risk factors for death caused by A. baumannii infection. The AUC value of ROC curves of study cohort and validation cohort were 0.76 and 0.69, respectively. The probability range (30–80%) indicated a high net income of the modified model and strong capacity of discrimination. The calibration curve obtained by analysis swings up and down around the 45 diagonal line, which shows that the calibration degree of the prediction model is very high.

Conclusion

In this study, we have reconstructed a risk prediction model for mortality in patients with A. baumannii infections. This model provides useful information to predict the risk of death in patients with A. baumannii infection, but the specificity is not optimistic. If this prediction model is wanted to be applied to clinical practice, more analysis and research are necessary.