Prediction model, risk factor score and ventilator-associated pneumonia: A two-stage case-control study

Abstract

Background: Ventilator-associated pneumonia (VAP) is one of the most important hospital acquired infections in patients requiring mechanical ventilation (MV) in the intensive care unit, but the effective and robust predictable tools for VAP prevention were relatively lacked. Methods: This study aimed to establish a weighted risk scoring system to examine VAP risk among a two-stage VAP case-control study, and to evaluate the diagnostic performance of risk factor score (RFS) for VAP. We constructed a prediction model by least absolute shrinkage and selection operator (LASSO), random forest (RF), and extreme gradient boosting (XGBoost) models in 363 patients and 363 controls, and weighted RFS was calculated based on significant predictors. Finally, the diagnostic performance of the RFS was testified and further validated in another 177 pairs of VAP case-control study. Results: LASSO, RF and XGBoost consistently revealed significant associations of length of stay before MV, MV time, surgery, tracheotomy, multiple drug resistant organism infection, C-reactive protein, PaO2, and APACHE II score with VAP. RFS was significantly linearly associated with VAP risk [odds ratio and 95 % confidence interval = 2.699 (2.347, 3.135)], and showed good discriminations for VAP both in discovery stage [area under the curve (AUC) = 0.857] and validation stage (AUC = 0.879). Conclusions: Results of this study revealed co-occurrence of multiple predictors for VAP risk. The risk factor scoring system proposed is a potentially useful predictive tool for clinical targets for VAP prevention.