Objective: Surveillance of nosocomial bloodstream infection (BSI) is recommended, but time-consuming. We explored strategies for automated surveillance.
Methods: Cohort study. We prospectively processed microbiological and administrative patient data with computerized algorithms to identify contaminated blood cultures, community-acquired BSI, and hospital-acquired BSI and used algorithms to classify the latter on the basis of whether it was a catheter-associated infection. We compared the automatic classification with an assessment (71% prospective) of clinical data.
Setting: An 850-bed university hospital.
Participants: All adult patients admitted to general surgery, internal medicine, a medical intensive care unit, or a surgical intensive care unit over 3 years.
Results: The results of the automated surveillance were 95% concordant with those of classical surveillance based on the assessment of clinical data in distinguishing contamination, community-acquired BSI, and hospital-acquired BSI in a random sample of 100 cases of bacteremia. The two methods were 74% concordant in classifying 351 consecutive episodes of nosocomial BSI with respect to whether the BSI was catheter-associated. Prolonged episodes of BSI, mostly fungemia, that were counted multiple times and incorrect classification of BSI clinically imputable to catheter infection accounted for 81% of the misclassifications in automated surveillance. By counting episodes of fungemia only once per hospital stay and by considering all cases of coagulase-negative staphylococcal BSI to be catheter-related, we improved concordance with clinical assessment to 82%. With these adjustments, automated surveillance for detection of catheter-related BSI had a sensitivity of 78% and a specificity of 93%; for detection of other types of nosocomial BSI, the sensitivity was 98% and the specificity was 69%.
Conclusion: Automated strategies are convenient alternatives to manual surveillance of nosocomial BSI.