Development and validation of a predictive model for ventilator-associated pneumonia in patients with cardiogenic shock: based on the MIMIC-IV database

Mechanical ventilation is crucial for patients with cardiogenic shock (CS), while diagnosing ventilator-associated pneumonia (VAP) in CS patients is difficult. Therefore, there is an urgent need for an effective diagnostic model for VAP in CS. This study aims to develop an effective risk prediction model for VAP in CS patients based on clinical data. The study is a retrospective study conducted using the Medical Information Mart for Intensive Care IV (MIMIC-IV) dataset. Univariate and multivariate binary logistic regression analyses identified the variables for establishing a predictive model. Its clinical utility was assessed via decision curve analysis (DCA), as well as its discrimination and calibration through the concordance index (C-index) and calibration plots. Among the 807 CS patients admitted to the intensive care unit (ICU), 112 developed VAP. The results of this study suggest that the duration of invasive mechanical ventilation, the length of ICU stay (ICU LOS), concomitant hepatic insufficiency, and the presence of concomitant sepsis are independent risk factors associated with the development of VAP during hospitalization. The area under the curve for the model was 0.798. In addition, the clinical data of 90 CS patients in the South District of The Third Affiliated Hospital of Anhui Medical University were retrospectively analyzed for external verification, and the area under the external validation curve was 0.783. The Hosmer-Lemeshow P=0.47 indicates that the fit is acceptable, and the calibration curve proves that the predictive model has proper discrimination and good calibration. DCA revealed that the VAP prediction nomogram proved clinically valuable when interventions were considered at a VAP probability threshold ranging from 1% to 50%. We can apply the nomogram for predicting the development of VAP after admission to the ICU for patients with CS, utilizing readily accessible variables.