Study objectives: To identify, in patients experiencing hospital-acquired pneumonia (HAP), prognostic factors present at disease onset and build an algorithm capable of stratifying mortality risk upon HAP onset.
Design: Observational cohort from January 1994 to December 2001.
Setting: One intensive care unit (ICU) from a university-affiliated, urban teaching hospital.
Patients: All consecutive patients exhibiting bacteriologically documented HAP either on ICU admission or during ICU stay.
Interventions: Data collection and multivariate analysis using Chi-Square Automatic Interaction and Detection technique.
Results: 168 patients were studied. The overall mean mortality rate was 49.4%. Upon onset of HAP, five independent variables allowed binary stratification of mortality risk. These consisted of underlying diseases (nonfatal versus ultimately and rapidly fatal diseases), Simplified Acute Physiology Score II (less than versus > or =37), platelet count (less than versus > or =150,000/mm3), chest x-ray involvement (1 versus >1 lobe), and PaO2/FiO2 (less than versus > or =167 mm Hg). A branching algorithm consisting of these five variables identified patients with HAP at both low (<35%) and high (>75%) risk of mortality.
Conclusion: Mortality in ICU patients with HAP may be predicted early, upon onset of HAP, by the cumulative use of prognostic factors in an algorithm.