BACKGROUND: Nosocomial bloodstream infections occur at a rate of 1.3 to 14.5 per 1000 hospital admissions and are believed to lead directly to 62,500 deaths per year in the United States. Measures of the incidence and the proportion of all hospital deaths related to deaths from these infections provide estimates of their impact. The objectives of the study were to characterize the secular trends in nosocomial bloodstream infection at a single institution and to estimate the population-attributable risk for death among patients experiencing the infection. METHODS: A 12-year retrospective study using prospectively collected data from a hospital-wide surveillance system for nosocomial infections in a 900-bed tertiary care institution. All patients (N = 260,834) admitted to the institution between 1980 and 1992 were included in the study. Bloodstream infection rates were calculated for the 10 leading groups of pathogens, and trends were analyzed using simple linear regression. In-hospital mortality rates from patients who did or did not develop nosocomial blood stream infections were compared. RESULTS: Between 1980 and 1992, a total of 3077 patients developed 3464 episodes of nosocomial bloodstream infection. The crude infection rates increased linearly from 6.7 to 18.4 per 1000 discharges (0.83 to 1.72 episodes per 1000 patient-days) during the 12-year study period (r =.87). Increases in the infection rates were due to gram-positive cocci (r =.96) and yeasts (r =.95) and essentially explained by infections caused by coagulase-negative staphylococci, Staphylococcus aureus, enterococci, and Candida species, respectively. Although the crude mortality in patients with nosocomial bloodstream infections decreased from 51% in 1981 to 29% in 1992, the in-hospital population-attributable mortality among infected patients increased from 3.55 deaths per 1000 discharges in 1981 to 6.22 per 1000 discharges in 1992 (r =.67). The etiologic fraction or the proportion of deaths in patients with bloodstream infection to all deaths occurring in the hospital increased from 11.4% in 1981 to 20.4% in 1992 (r =.59). CONCLUSIONS: The incidence, the etiologic fraction, and the population-attributable risk for death among patients experiencing nosocomial bloodstream infections increased progressively during the last decade.