Lipocalin-2 is a Sensitive and Specific Marker of Bacterial Infection in Children

Jethro Herberg; Honglei Huang; Marie L. Thezenas; Victoria Janes; Michael Carter; Stuart Gormley; Melisa S. Hamilton; Benedikt Kessler; Michael Levin; Climent Casals-Pascual

Results. Patient groups. Table 1 shows the demographic features of the patient groups. Ethnicity and gender did not differ significantly. Children in the DV group were younger than those in PB and C groups, as expected for viral respiratory infection. The proportion of children with severe illness was higher in the DB than DV group. LCN2 and MMP-8 are associated with SBI. Concentrations of plasma LCN2 and MMP-8 were lowest in controls, and increased stepwise with likelihood of SBI (Table 1) . CRP, LCN2 and MMP-8 discriminated patients in the DB and DV or C groups ( Figure 2a ). We derived ROC curves for CRP, LCN2, MMP-8 for all 191 patients, and for neutrophil proportion and white cell counts in 175 patients with available values, based on the comparison of the DB group with the other febrile patients (combined PB, U and DV groups) (Figure 2b ). LCN2 used as a continuous variable predicted confirmed SBI with an AUROC of 0.88 (95% CI: 0.82-0.94), matching the AUROC for CRP (0.89 [95% CI: 0.84-0.94]). MMP-8, neutrophil proportion and white cell count predicted confirmed SBI less well than LCN2 and CRP (AUROC 0.80, 0.71 and 0.69 respectively) ( Figure 2b ). LCN2 performance is not confounded by creatinine concentration. As LCN2 levels increase with renal damage(7), we determined whether renal impairment influenced the diagnostic performance of LCN2. Creatinine levels contemporaneous with research bloods were available for 162 children (median and IQR: 39 and 34-50 µmol/L), and did not significantly differ between clinical groups (Kruskal-Wallis, P=0.36). The adjusted logistic regression analysis indicated that prediction of SBI by LCN2 was not affected by creatinine concentration (supplementary Table 1 ). Combination of LCN2 and CRP to predict SBI. CRP was used as a benchmark to evaluate the performance of LCN2 to predict confirmed SBI in 191 patients with available CRP and LCN2 data. The AUROC of LCN2 and CRP were strongly associated with confirmed SBI. Both biomarkers were highly correlated (r = 0.40, P<0.001). However, when both proteins were included into a logistic regression model, the combination was significantly superior to either marker alone (AU-ROC 0.92 [95%CI: 0.88-0.96]) (Figure 3 and Supplementary Table 1 ). LCN2 and CRP in uncertain cohorts. Without a sensitive gold-standard test for detection of SBI, the majority of SBI is 'missed', therefore we correlated LCN2 and CRP biomarker concentrations with clinical suspicion of SBI. There was a stepwise increase in biomarker concentration with increasing suspicion of SBI: from C, DV, U, PB, to B (Table 1, Figure 4 ). We derived cut-off values for LCN2 and CRP based on the concentration that showed the highest sensitivity and specificity to discriminate confirmed bacterial infection (DB) from other febrile patients (Supplementary Table 1 ). The sensitivity and specificity of LCN2 to predict DB were 83.3% and 82.5%. CRP had sensitivity and specificity of 85.7% and 83.8%. When the threshold value for LCN2 was applied to the other groups, there was a stepwise increase in prediction of SBI: 2 of 40 (5%) (C), 6 of 42 (14%) (DV), 9 of 33 (27%) (U), 15 of 34 (44%) (PB). The sensitivity and specificity of both markers combined (equal or greater than their 7th decile) was 90.4% and 86.5%. Children admitted with acute infection, sick enough to warrant blood tests, were recruited after presentation and before diagnostic studies were completed, at St Mary's Hospital, London and University Ho