Hospital-Acquired Pneumonia

Study objectives: To determine factors associated with antimicrobial-resistant hospital-acquired pneumonia (AR-HAP), to build an algorithm evaluating the risk for such a pneumonia, and to test this algorithm. Design: Combined observational and validation cohorts over two periods: January 1994 to December 1999, and January 2000 to March 2001. Setting: One ICU from a university-affiliated urban teaching hospital. Patients: One hundred twenty-four patients in the observational cohort and 26 patients in the validation cohort exhibiting bacteriologically documented hospital-acquired pneumonia (HAP). Interventions: Prospective data collection and multivariate analysis using the χ 2 automatic interaction and detection technique. Results: In the observational cohort, 39 antimicrobial-resistant bacteria were incriminated in 37 patients (30%). Multivariate analysis identified four independent variables allowing a binary stratification of risk. According to the presence or absence of prior antimicrobial treatment, neurologic disturbances on ICU admission, aspiration on ICU admission, and time elapsed between ICU admission and the onset of pneumonia, we were able to identify and separate patients at high, low, or even no risk for acquiring AR-HAP. In the validation cohort, nine antimicrobial-resistant bacteria were incriminated in nine patients (34.6%). In this cohort, the algorithm performed well allowing the identification of null risk categories: the absence of prior antimicrobial treatment, the presence of prior antimicrobial treatment with neurologic disturbances on ICU admission and an early-onset pneumonia, and the presence of prior antimicrobial treatment without neurologic disturbances but with aspiration on ICU admission were always associated with antimicrobial-susceptible HAP. Conclusion: We developed and tested a binary algorithm allowing the identification of patients at low risk for acquiring AR-HAP. An antibiotic strategy including an initial antimicrobial treatment guided by such an algorithm, followed, if possible, by a de-escalation when antimicrobial data are available, could increase the administration of adequate initial antimicrobial treatment and help prevent the emergence of antibiotic resistance in the ICU.