Ventilator-associated pneumonia and attributable mortality: a systematic review of observational studies

Abstract number: O136

Melsen W.G., Rovers M.M., Bonten M.J.M.

Objective: To determine the attributable mortality of Ventilator-associated Pneumonia (VAP) based upon the results of observational studies.

Methods: A systematic review and meta-analysis was performed. The studies were identified by performing a systematic search strategy using PubMed, Web of Science and Embase through February 2007. Only studies reporting mortality rates of patients with and without VAP were included. The data were extracted using standardised forms and the quality of all studies was determined by a validated scoring system.

Results: 52 studies with a total of 17.347 patients met the inclusion criteria. Mortality rates of patients with VAP ranged from 14 to 78%. Pooling of all studies resulted in an I² statistic of 69%, indicating considerable heterogeneity and, therefore, precluding estimation of an overall effect. A significant association between VAP and mortality, in univariate analyses, was observed in 17 of 52 studies (33%). Yet, nine studies also performed multivariate analyses, and significance of independent association persisted only in four of these studies. The origin of heterogeneity could not be explained by differences in study design, study quality and diagnostic approach. However heterogeneity was limited for studies investigating only trauma patients (I²=1.3%) or acute respiratory distress syndrome (ARDS) patients (I²=0%). The estimated relative risk of mortality of patients with VAP in these patient groups were 1.09 (95% CI 0.87–1.37) among trauma patients and 0.86 (95% CI 0.72 to 1.04) among ARDS patients.

Conclusions: The widely held belief that VAP is associated with attributable mortality cannot be based on the available evidence from observational studies. In fact, for two specific patient groups (trauma and ARDS) there is evidence of absence of such an association. Only analyses including more patient specific data, thereby allowing to adjust for possible confounders, might confirm or reject the presumed association for other patient groups.