New prediction model for diagnosis of bacterial infection in febrile infants younger than 90 days

Abstract

Due to non-specific clinical presentation in febrile infants, extensive laboratory testing is often carried out to distinguish simple viral disease from serious bacterial infection (SBI). Objective of this study was to compare efficacy of different biomarkers in early diagnosis of SBI in infants <90 days old. Also, we developed prediction models with whom it will be possible to diagnose SBI with more accuracy than with any biomarkers independently. Febrile <90-day-old infants hospitalized in 2-year-period at Department of Pediatrics, University Hospital Centre Split with suspicion of having SBI were included in this study. Retrospective cohort analysis of data acquired from medical records was performed. Out of 181 enrolled patients, SBI was confirmed in 70. Most common diagnosis was urinary tract infection (68.6%), followed by pneumonia (12.9%), sepsis (11.4%), gastroenterocolitis (5.7%) and meningitis (1.4%). Male gender was shown to be a risk factor for SBI in this population (p=0.008). White blood cell count (WBC), absolute neutrophil count (ANC) and C-reactive protein (CRP) were confirmed as the independent predictors of SBI, with CRP as the best one. Two prediction models built by combining biomarkers and clinical variables were selected as optimal with sensitivities of 74.3% and 75.7%, and specificities of 88.3% and 86%. Evidently, CRP is a more superior biomarker in diagnostics of SBI comparing to WBC and ANC. Prediction models were shown to be better in predicting SBI than independent biomarkers. Although both showed high sensitivity and specificity, their true strength should be determined using validation cohort.