The diagnostic performance of serum C-reactive protein (CRP) in prediction of bacteremia among febrile patients visiting an emergency department (ED) was analyzed.

During randomly selected 96 days between August 2006 and July 2007, a prospective study of febrile adults visiting the ED of a medical center was conducted to analyze the clinical characters associated with bacteremia.

Of the total 454 febrile adults enrolled, their mean age was 54.1 years, and 232 (54.6%) were women. Major comorbidities included cardiovascular disease (137 patients, or 30.1%) and diabetes mellitus (105, or 23.1%). Seventy-four patients (16.2%) had true bloodstream infections with the predominance of monomicrobial gram-negative bacteremia in 49 patients (10.7%). Four risk factors, including low platelet count (<100 000/mm³; odds ratio [OR], 4.19; 95% confidence interval [CI], 1.85-9.47; P = .001), high blood urea nitrogen (>20 mg/dL; OR, 4.61; 95% CI, 2.56-8.31; P < .001), high fever (>39.0°C; OR, 3.67; 95% CI, 2.05-6.59; P < .001), and high Pittsburg bacteremia scores (≧4 points; OR, 2.95; 95% CI, 1.01-8.57; P = .04) were independently associated with bacteremic episodes. Of note, high CRP (>150 mg/dL; OR, 1.75; 95% CI, 0.73-3.99; P = .21) was not an independent risk factor. In further analysis, the difference of serum CRP levels between bacteremic and nonbacteremic adults was significant only when the period from fever onset to ED arrival was more than 12 hours.

The CRP level was not reliable to distinguish the bacteremia from nonbacteremic infection, whereas duration after fever onset was less than 12 hours. Clinicians must consider the history of fever onset to improve the accuracy of early prediction of serum CRP before the microbiological results of blood cultures is available.