Objective: To develop and validate a model for the prediction of bacteremia in hospitalized patients, and to identify subgroups of patients with a very low likelihood of bacteremia in whom a positive blood culture has a low positive predictive value.
Design: Prospective cohort study with clinical data on 1516 episodes collected from a random sample of all patients who had blood cultures done at one institution.
Setting: Urban, tertiary care hospital.
Patients: Derivation set: 1007 blood culture episodes sampled from all blood cultures done on patients at Brigham and Women's Hospital between October 1988 and February 1989. Validation set: 509 episodes, May 1989 to June 1989. The unit of evaluation was the episode, defined as a 48-hour period beginning after a blood culture was drawn.
Measurements and main results: True- and false-positive rates of blood cultures in the derivation set as assessed by independent reviewers were 7% (74 of 1007) and 8% (81 of 1007), respectively. Independent multivariate predictors of true bacteremia were temperature of 38.3 degrees C or higher, presence of a rapidly (less than 1 month) or ultimately (less than 5 years) fatal disease; shaking chills; intravenous drug abuse; acute abdomen on examination; and major comorbidity. In the low-risk group, defined by absence of these predictors, the misclassification rate of the model in the derivation set was 1% (4 of 303), and a positive blood culture had a positive predictive value of only 14% for true bacteremia. The model also identified a high-risk subset in which 16% (41 of 264) of episodes represented true bacteremia. The model was prospectively validated in 509 additional episodes, and the misclassification rate in the low-risk group was 2% (3 of 155).
Interventions: None.
Conclusion: These findings provide a means of stratifying hospitalized patients according to their risk for bacteremia. If prospectively validated in other settings, this model may be helpful when deciding whether or not to do blood cultures or start antibiotic therapy and, when evaluating a positive blood culture, to determine whether or not it is a true-positive.