Introduction: Nosocomial infections are major contributors to pediatric intensive care unit (PICU) morbidity, mortality, and cost. We prospectively identified risk factors present on PICU admission day to identify patients at high risk for subsequent development of nosocomial infection.

Methods: Data were collected from 912 consecutive PICU admissions over a 12 month period and consisted of physiologic profiles (PRISM III), therapeutic interventions, and culture results. PICU admission day data was used to predict nosocomial infections (multivariate logistic regression analysis) which developed within the first 14 days of PICU stay.

Results: 16 patients developed at least one major nosocomial infection (ie. pneumonia and/or sepsis), and 47 developed only minor infections (ie. all other types) within the first 14 days of PICU stay. Multivariate logistic regression analysis of admission day data showed that, when combined and weighted according to their relative contribution to nosocomial infection risk, parenteral nutrition, operative status, PRISM III, weight, and number of invasive devices were significant predictors of subsequent development of nosocomial infection. The performance of the model was very good, with area under the receiver operating characteristic curve = 0.765. At a prediction probability of > = 0.10, the sensitivity, specificity, false positive, and false negative rates were 63.5%, 78.6%, 21.4%, and 36.5%, respectively. At this same prediction probability, 87.5% of the major infections were correctly classified, while only 55.3% of the minor infections were correctly classified. Over the entire range of prediction probabilities, the model always correctly classified a larger percentage of major infections.

Conclusion: Admission day risk factors can be used to develop profiles of PICU patients at high risk for developing nosocomial infections and thus, identify patients appropriate for interventional studies.