Objective To determine whether the input of time from fever onset will change the accuracy of C-reactive protein (CRP) in diagnosing bacterial infections in febrile children.

Study design We performed a prospective observational study on febrile children presenting to the emergency department. The diagnostic performance of CRP at different time points from fever onset was compared using a receiver operating characteristic (ROC) curve.

Results Among 373 patients included, 103 (28%) had bacterial infection. The optimal cut-off for CRP suggesting bacterial infection changed with time from fever onset: 6 mg/dL for >12–24 h of fever; 10.7 and 12.6 mg/dL at >24–48 and >48 h of fever, respectively. The input of time from fever onset improved the area under the ROC curve from 0.83 (95% CI 0.78 to 0.88) for CRP overall to 0.87 (95% CI 0.77 to 0.96) and 0.90 (95% CI 0.84 to 0.97) at >24–48 and >48 h of fever, respectively. Duration of fever mostly affected the ability of CRP to correctly rule out bacterial infections. CRP level of 2 mg/dL obtained at ≤24 h of fever corresponds with a post-test probability for bacterial infection of 10%, whereas the same value obtained >24 h of fever reduces the risk to 2%.

Conclusions Clinicians should apply different CRP cut-off values depending on whether they are trying to rule in or rule out bacterial infection, but also depending on fever duration at the time of CRP testing.