Methods
This prospective cohort study included 1509 episodes in which hospitalized patients had blood cultures performed in an urban tertiary-care hospital. The main outcome measures were gram-negative bacteremia and gram-negative sepsis.

Results
Of 1509 episodes, 115 (8%) represented bacteremia and 40 (3%) included gram-negative rods. Of these 40 patients, nine died in the hospital, including five patients who had gram-negative sepsis; all five had another rapidly fatal disease. Using criteria for treatment and exclusions from the HA-1A trial, three of the patients with gram-negative bacteremia would have been treated, while at least 52 patients without gram-negative bacteremia might have received HA-1A therapy (positive predictive value of criteria, 5.5%). Of the 1509 episodes, sepsis syndrome as defined by Bone was present in 34 (2.3%). While 32 of the 34 patients had suspected gram-negative bacteremia, only five had blood cultures positive for gram-negative bacteria.

Conclusions
In this population, current criteria for administration of monoclonal anti—endotoxin antibody therapy were not sensitive or specific for gram-negative bacteremia, and many patients with gram-negative sepsis were too ill from other conditions to benefit. Indiscriminate use of these therapies could thus be costly yet yield few benefits. To identify patients who should receive novel therapies, better risk-stratification methods and cost-effectiveness analysis are needed.

(Arch Intern Med. 1994;154:1241-1249)