Identifying predictors of nosocomial pneumonia in trauma patients admitted to a level-1 trauma center

Abstract

Background

Nosocomial pneumonia is common in trauma patients and associated with an adverse prognosis. We recently externally validated and recalibrated an existing formula to predict nosocomial pneumonia risk. Identifying more potential predictors could aid in a more accurate prediction of nosocomial pneumonia risk in level-1 trauma patients. This study aims to identify predictors of nosocomial pneumonia in level-1 trauma patients available in the emergency department or shortly after, and examine their added predictive value to an existing prediction model by Croce.

Methods

This retrospective cohort study included all consecutive trauma patients (≥ 16 years) admitted for > 24 h to our level-1 trauma center in 2017. Excluded were patients with active infection upon admission, transfer from another hospital, or in-hospital mortality < 48 h. Multiple imputations were used for missing values. Multivariable logistic regression analysis and Ridge penalization were performed to assess the association of predictors with nosocomial pneumonia and evaluate predictor stability. The predictive performance in addition to the existing prediction model was evaluated as well.

Results

The study included 809 patients [median age 51 (IQR 32–68) years, 66.9% male, median ISS 10 (5–17), median GCS score 15 (14–15)]. Pneumonia incidence was 10.6% (n = 86). Age (OR 1.03 per year), ISS (OR 1.10 per point), GCS score (OR 0.91 per point), pulmonary contusion (OR 2.77), male sex (OR 1.36), hypertension (OR 1.86), diabetes (OR 1.20), number of rib fractures (OR 1.05 per fractured rib), and thoracic spine fracture (OR 1.51) were found to be predictors of nosocomial pneumonia. All variables showed added predictive value in addition to the existing model.

Conclusion

Patient history, injury severity, thoracic trauma, and traumatic brain injury are essential components of nosocomial pneumonia prediction and add to the predictive value of an existing model. Our results further build a basis for more accurate prediction.

Level of evidence

Level III, prognostic/epidemiological.