Procalcitonin to Detect Suspected Bacterial Infections in the PICU

Objectives: Nonspecific clinical symptoms frequently lead to suspicion of bacterial infection in critically ill children. Clinicians send bacterial cultures for suspected infection and begin an empiric course of antibiotics while microbiology results are pending. We investigated whether the biomarker procalcitonin could be useful to predict confirmed bacterial infection in critically ill children in the PICU, before culture results are available. Design: Prospective, blinded single-center study. Setting: Tertiary PICU and cardiothoracic ICU. Patients: There were one hundred forty-four patients with suspected bacterial infections that had bacterial cultures sent by clinicians. Interventions: Procalcitonin samples were obtained at three time intervals: as close to the time of the initial culture as possible (up to 12 hr after) and 24 and 72 hours after the initial culture. Patients were stratified into clinical outcome groups based on microbiology results and clinical symptoms using Centers for Disease Control and Prevention criteria. These assignments were blinded to procalcitonin levels. Primary outcome was the presence of culture-proven bacterial infection. Measurements and Main Results: There was a statistically significant difference in initial and subsequent median procalcitonin values between patients with confirmed bacterial infections and patients with low suspicion of bacterial infection (p < 0.02). However, there was extremely high variability in procalcitonin values among all groups. Procalcitonin had only a fair ability to predict bacterial infection, with area under the curve of receiver operating characteristic plots ranging between 0.63 and 0.71. When using serial procalcitonin values to predict bacterial infection, positive likelihood ratios were near 1 and negative likelihood ratios were between 0.3 and 0.4. Conclusions: Procalcitonin levels were higher in children with documented confirmed bacterial infection as compared with those with low suspicion of infection. However, neither single nor serial procalcitonin measurements were able to predict the presence or absence of confirmed bacterial infection with enough certainty to be clinically useful as to recommend initiating or withholding antibiotics.