The diagnosis of serious bacterial infection (SBI) in young febrile children remains challenging. This prospective, multicentre, observational study aimed to identify new protein marker combinations that can differentiate a bacterial infection from a viral infection in 983 children, aged 7 days-36 months, presenting with a suspected SBI at three French paediatric emergency departments. The blood levels of seven protein markers (CRP, PCT, IL-6, NGAL, MxA, TRAIL, IP-10) were measured at enrolment. The patients received the standard of care, blinded to the biomarker results. An independent adjudication committee assigned a bacterial vs. viral infection diagnosis based on clinical data, blinded to the biomarker results. Computational modelling was applied to the blood levels of the biomarkers using independent training and validation cohorts. Model performances (area under the curve (AUC), positive and negative likelihood ratios (LR+ and LR-)) were calculated and compared to those of the routine biomarkers CRP and PCT. The targeted performance for added value over CRP or PCT was LR+ ≥ 5.67 and LR- ≤ 0.5. Out of 652 analysed patients, several marker combinations outperformed CRP and PCT, although none achieved the targeted performance criteria in the 7 days-36 months population. The models seemed to perform better in younger (7-91 day-old) patients, with the CRP/MxA/TRAIL combination performing best (AUC 0.895, LR+ 10.46, LR- 0.16). Although computational modelling using combinations of bacterial- and viral-induced host-protein markers is promising, further optimisation is necessary to improve SBI diagnosis in young febrile children.
Keywords: C-reactive protein (CRP); IFN-γ-induced protein 10 (IP-10); bacterial- and viral-induced host biomarkers; interleukin 6 (IL-6); machine learning; myxovirus resistance protein 1 (MxA); neutrophil gelatinase-associated lipocalin-2 (NGAL); paediatric serious bacterial infection; procalcitonin (PCT); tumor necrosis factor-related apoptosis-inducing ligand (TRAIL).