VALIDATION OF A BACTEREMIA PREDICTION MODEL

OBJECTIVE: To validate a previously published model for predicting bac teremia in hospitalized patients. DESIGN: Application of a published b acteremia prediction model to a prospective validation cohort of patie nts and comparison of its predictability to that found in the derivati on cohort. SETTING: Urban, university-affiliated, 550-bed public hospi tal. PATIENTS: The validation cohort consisted of 342 patients with 55 9 blood culture episodes between October 14, 1992, and December 5, 199 2, Each blood culture episode was scored based on the presence or abse nce of seven predictors of bacteremia and the findings compared with p ublished results (derivation cohort).INTERVENTIONS: None. RESULTS: App lication of the bacteremia prediction model to the validation cohort i dentified episodes with a low risk (3%) and a high risk (17%) for true bacteremia, similar to the findings in the derivation cohort (1% and 16%, respectively). Comparison of the predictions of the model in the two cohorts by receiver operator characteristic curve analysis reveale d that the overall predictability of the model in the validation cohor t was not as good as in the derivation cohort. CONCLUSIONS: Although t he bacteremia prediction model did not perform as well overall in the validation cohort, the model still was able to clearly de fine two ext reme groups: those with a low risk and those with a high risk for true bacteremia. This predictive capability may aid physicians in prescrib ing empiric antimicrobial therapy and also may be useful to hospital e pidemiologists in assessing quality of care (Infect Control Hosp Epide miol 1995;16:203-209).