Background.The objective of this study was to examine central venous catheter (CVC)-related nosocomial blood stream infection risks of umbilical venous, percutaneous and Broviac catheters, as well as variations in CVC use and CVC-related risk for nosocomial blood stream infection in the neonatal intensive care unit (NICU).Methods.A cohort study was performed based on 19 507 infants admitted to 17 NICUs in the Canadian Neonatal Network from January, 1996, through October, 1997. Information on these subjects was prospectively collected by trained abstractors. Incidence of infection was measured as infection episodes per 1000 patient days. The risk ratio (RR) of CVC use for nosocomial blood stream infection was calculated as the infection rate during catheter days divided by the infection rate during noncatheter days. Using a Poisson regression model we examined the adjusted RR of CVC use for nosocomial blood stream infection, controlling for patient characteristics and illness severity at admission. Interinstitutional variations in CVC-related infection risks were examined by stratified analyses.Results.CVC were used in 22.5% of patients. The incidence of nosocomial blood stream infection was 2.9 per 1000 noncatheter days, 7.2 per 1000 umbilical venous catheter days, 13.1 per 1000 percutaneous catheter days and 12.1 per 1000 Broviac catheter days. The RR for nosocomial blood stream infection, adjusted for differences in patient characteristics and admission illness severity, was 2.5 for umbilical venous catheter, 4.6 for percutaneous catheter and 4.3 for Broviac catheter (P< 0.05). There were significant (P< 0.05) risk-adjusted variations in CVC-related infection risks among NICUs.Conclusions.CVC use increased the risk of nosocomial blood stream infection. The risk of nosocomial blood stream infection in percutaneous and Broviac catheters was 70 to 80% higher than in umbilical venous catheters. There was significant variation in CVC-related infection risks among Canadian NICUs.