Study objective: We estimated the accuracy of the total peripheral WBC count as a screen for bacteremia in febrile young infants.
Methods: We evaluated, retrospectively, the performance characteristics of linear and nonlinear (U-shaped) logistic models for predicting bacteremia that are based on the total peripheral WBC count. Research subjects were consecutive 0- to 89-day-old infants who had a temperature in triage of greater than or equal to 38 degrees C (> or =100.4 degrees F) and were evaluated for infection at a pediatric emergency department (1993 to 1999). Infants with leukemia were excluded. Areas under the receiver operator characteristic curves (AUC), as well as sensitivity, specificity, interval likelihood ratios, and the corresponding odds of bacteremia predicted at various thresholds of the test, were calculated.
Results: The rate of bacteremia was 1% (38/3,810). The U-shaped model was more accurate (AUC 0.69 versus 0.56); however, no threshold of the total peripheral WBC count had both good sensitivity and specificity. Sensitivity and specificity values were 79% and 5%, respectively, at a peripheral WBC count cutoff of 5,000 cells/mm(3), and 45% and 78%, respectively, at a cutoff of 15,000 cells/mm(3). The odds of bacteremia were not decreased substantially at any cutoff and were increased only modestly at values outside published norms of the test.
Conclusion: The total peripheral WBC count is an inaccurate screen for bacteremia in febrile young infants; thus, decisions to obtain blood cultures should not rely on this test.