by Joseph M. Swanson, G. Christopher Wood, Lijing Xu, Lisa E. Tang, Bernd Meibohm, Ramin Homayouni, Martin A. Croce, Timothy C. Fabian
Ventilator-associated pneumonia (VAP) carries significant mortality and morbidity. Predicting which patients will become infected could lead to measures to reduce the incidence of VAP. Methodology/Principal Findings
The goal was to begin constructing a model for VAP prediction in critically-injured trauma patients, and to identify differentially expressed genes in patients who go on to develop VAP compared to similar patients who do not. Gene expression profiles of lipopolysaccharide stimulated blood cells in critically injured trauma patients that went on to develop ventilator-associated pneumonia (n?=?10) was compared to those that never developed the infection (n?=?10). Eight hundred and ten genes were differentially expressed between the two groups (ANOVA, P<0.05) and further analyzed by hierarchical clustering and principal component analysis. Functional analysis using Gene Ontology and KEGG classifications revealed enrichment in multiple categories including regulation of protein translation, regulation of protease activity, and response to bacterial infection. A logistic regression model was developed that accurately predicted critically-injured trauma patients that went on to develop VAP (VAP+) and those that did not (VAP-). Five genes (PIK3R3, ATP2A1, PI3, ADAM8, and HCN4) were common to all top 20 significant genes that were identified from all independent training sets in the cross validation. Hierarchical clustering using these five genes accurately categorized 95% of patients and PCA visualization demonstrated two discernable groups (VAP+ and VAP-). Conclusions/Significance
A logistic regression model using cross-validation accurately predicted patients that developed ventilator-associated pneumonia and should now be tested on a larger cohort of trauma patients.