Background: The detection of infectious bacteria in blood culture samples is important for diagnosis and treatment, but this requires 1-2 days at least, and is not adequate as a rapid test. Therefore, we have investigated the diagnostic ability and the optimal cutoff value of procalcitonin (PCT) and C-reactive protein (CRP) for predicting the bacteremias using receiver operating characteristic (ROC) curves and relative cumulative frequency distribution (RCD) curves.
Methods: A case-control study was performed in inpatients (852 subjects: 426 positive cultures and 426 negative cultures) from January 1 to December 31, 2014. We retrospectively investigated their blood culture and blood chemistry findings recorded in this period using electronic medical records.
Results: Area under the ROC curve of PCT and CRP were 0.79 and 0.66, respectively. The optimal cutoff values were 0.5 μg/L with a sensitivity of 70% and specificity of 70% for PCT and 50.0 mg/L with a sensitivity of 63% and specificity of 65% for CRP. When the optimal cutoff value was treated as a reference, the odds ratio (OR) was 71.11 and the hazard ratio (HR) was 6.27 for PCT >2.0 μg/L, and the risk of blood culture positivity was markedly elevated. PCT levels were significantly higher in the population with Gram-negative rod (GNR) infections than in the population with Gram-positive coccal (GPC) infections.
Conclusions: The elevation of CRP and PCT were significantly associated with bacteremias. PCT was superior to CRP as a diagnostic indicator for predicting bacteremias, for discriminating bacterial from nonbacterial infections, and for determining bacterial species.
Keywords: C-reactive protein; bacteremias; confounding factors; procalcitonin; receiver operating characteristics curve; relative cumulative frequency distribution curve.