Mortality associated with late-onset pneumonia in the intensive care unit:: results of a multi-center cohort study

Objective: To evaluate the attributable mortality associated with late-onset nosocomial pneumonia (LOP) while taking into account the severity at admission, the evolution of the patients during the first 4 days after admission to the ICU and the appropriateness of initial empiric antibiotic treatment. Design: Multicenter cohort study with prospective standardization of diagnostic interventions when nosocomial pneumonia develops. Setting: Medical and surgical ICUs of four university-affiliated teaching hospitals. Patients: Seven hundred sixty-four consecutive patients requiring ICU hospitalization for at least 4 days. Main outcome measures: The clinical and biological data as well as the therapeutic data and the outcome were prospectively recorded from the day of admission to ICU discharge. Simplified Acute Physiologic Score (SAPS II) and Logistic Organ Dysfunction (LOD) score were collected and computed within the first 4 calendar days of ICU admission. Variables associated with the outcome were selected using a stepwise Cox model. The time to acquisition of the first LOP was then introduced in the final model as a time-dependent covariate, The analysis was stratified by ICU center. Finally, as initial antibiotic therapy could have an impact on the increased risk of death induced by LOP, the Cox model was applied again introducing LOP immediately adequately treated and LOP not immediately adequately treated as two different time-dependent covariates. Results: Late-onset pneumonia developed in 89 patients (12%). A McCabe score of more than 1, SAPS II score and increases in SAPS between days 1 and 2, days 2 and 3, and days 3 and 4 were significantly associated with an increased risk of death. When the time to acquisition of the first episode of LOP was introduced into the Cox model, the LOP occurrence was associated with increased mortality, even adjusted over the selected prognostic parameters and after stratification by center (hazard ratio (HR)=1.53, 95% CI 1.02-2.3, P=0.04). When LOP immediately adequately treated and LOP not immediately adequately treated were separately introduced into the Cox model, inappropriately treated LOP remained significantly associated with an increased risk of mortality (HR=1.69, 95% CI 1.08-2.65, p=0.022), whereas appropriately treated LOP did not (HR=1.44, 95% CI 0.75-2.76, p=0.27). Conclusion: These data suggest that, in addition to severity scores, the underlying medical conditions and the evolution of severity within the first 4 days in ICU, late-onset pneumonia independently contribute to ICU patient mortality when empirical antibiotic treatment is not immediately appropriate.