Cohort characteristics and incidence of AKI
Over the study period, there were 494 admissions to the PICU (Fig. 1a), from which a total of 49 patients were recruited into the study. The most common cause for study ineligibility was the absence of a urinary catheter (219/445, 49.2 %) or central venous or peripheral arterial line (168/445, 37.8 %); the presence of a urinary catheter and central venous or peripheral arterial line were necessary for sample collection. This inclusion requirement led to recruitment of a cohort of particularly unwell PICU patients. Summary characteristics are presented in Table 1. Of the 49 patients, 26 (53 %) were female, and 25 (58.1 %) were White British; the median age of the entire patient cohort was 3 (range 0.04–15) years. The most common reason for admission was pneumonia (12/49, 24.5 %) and sepsis (12/49, 24.5 %). Ventilation was required by 43/49 (88 %) patients. The mortality rate was 5/49 (10.2 %). Of the 49 patients, four (8.2 %) were admitted electively, none of whom experienced an AKI episode, and 24 (49.0 %) experienced an AKI episode during PICU admission. Twelve patients reached pRIFLEmax R, nine patients reached pRIFLEmax I and three patients reached pRIFLEmax F (Fig. 1a).
Baseline creatinine clearance
In previous AKI studies in children, a presumed baseline eCCl (calculated using the Schwartz formula) of 120 mL/min/1.73 m2 has been used in the absence of pre-admission SCr readings [12, 19]. In our study, we examined whether an eCCl of 120 mL/min/1.73 m2 was representative of baseline renal function in our patient cohort. We obtained pre-admission and pre-discharge SCr values for study subjects to calculate eCCls. Pre-admission SCr values were recorded if available in the 3-month period prior to hospital admission. Pre-discharge SCr values were defined as the final SCr recording prior to hospital discharge, although this may have been affected by muscle loss during the PICU admission. We excluded patients who died prior to PICU discharge. Only 4/49 (8.2 %) of our patient cohort had pre-admission SCr values, and Fig. 2 shows that the majority of pre-admission (75.0 %) and pre-discharge (68.2 %) eCCl values were above the presumed baseline value of 120 mL/min/1.73 m2. These data suggest that the widely used presumed baseline eCCl value of 120 mL/min/1.73 m2 may underestimate baseline renal function in our patient cohort.
Biomarkers responsive to changes in eCCl
Biomarker profile plots were created for each patient to aid visualization of data, and an example plot from a patient with AKI (pRIFLEmax I) is shown in Fig. 3. This profile demonstrates an increase in Cys-C coincident with falling eCCl and peaks in pNGAL, uNGAL and KIM-1 prior to the decrease in eCCl. A representative set of biomarker plots from a patient not experiencing AKI is provided in Electronic Supplementary Material Fig. 1.
To determine whether eCCl values from all patients correlated with coincident biomarker values, we calculated a Spearman non-parametric correlation coefficient (r) for each biomarker. A significant correlation was observed for Cys-C (r = −0.77, p < 0.0001), a weak correlation was observed for pNGAL (r = −0.14, p = 0.043) and uNGAL (r = −0.13, p = 0.045), and there was no significant correlation with coincident eCCl and the urinary biomarker KIM-1 (r = −0.076, p = 0.26).
Biomarkers elevated in AKI
We investigated whether biomarker levels changed according to AKI severity. For the three groups we considered (no AKI, pRIFLE R, pRIFLE I/F), we found that pNGAL (p = 0.027) and uNGAL (p = 0.0079) levels were significantly higher in periods of pRIFLE I/F (Fig. 4) and that Cys-C levels were significantly higher in periods of both pRIFLE R and pRIFLE I/F (p < 0.0001). This result suggests that the biomarker levels of Cys-C, pNGAL and uNGAL may aid the diagnosis of AKI.
Biomarker sensitivity and specificity
To identify which of the biomarkers provided the highest sensitivity and specificity, we performed an ROC analysis using both ‘pRIFLE R or worse’, and ‘pRIFLE I or worse’ to define AKI. All biomarkers displayed superior predictive power for ‘pRIFLE I or worse’ compared to ‘pRIFLE R or worse’. The best performing biomarkers were Cys-C and pNGAL (Fig. 5). We also considered the variation of Cys-C with age. It has been reported that infants under the age of 18 months have a higher mean Cys-C value than older children . To determine if this pattern held true in our patient cohort, we compared median Cys-C values of patients not experiencing AKI under 18 months old to those older than 18 months and found no significant difference (≤18 months 0.75 mg/L, ≥18 months 0.78 mg/L; p = 0.66). We additionally examined whether a correlation between fluid status and biomarker levels existed. No significant correlation was observed for Cys-C, pNGAL or uNGAL, although a weak correlation was observed for KIM-1 (ESM Fig. 2).
The utility of pNGAL in AKI associated with sepsis
It has previously been reported that Cys-C and uNGAL levels are not altered by sepsis, whereas pNGAL levels rise in sepsis and cannot reliably discriminate AKI from no AKI in the septic state . As 24/49 (49.0 %) of our patients were admitted to the PICU with either sepsis or pneumonia, we examined whether pNGAL was a valid AKI biomarker in sepsis in our patient cohort. For patients admitted to PICU with either sepsis or pneumonia, we compared biomarker levels during periods of no AKI to periods of AKI using an AKI definition of pRIFLE R or worse (Fig. 6). Whereas Cys-C retained its ability to discriminate no AKI from AKI (p < 0.0001), there was no significant difference in pNGAL levels between no AKI and AKI periods in patients admitted with sepsis or pneumonia (p =0.97). We also performed an ROC analysis using median biomarker values for patients admitted to the PICU with either sepsis or pneumonia, using pRIFLE R as a cut-off between no AKI and AKI. Again, Cys-C retained its utility as an AKI biomarker in this patient group (AUC 0.86, p = 0.002), while pNGAL performed poorly (AUC 0.54, p = 0.71). Thus, our data support the previous data indicating that pNGAL is not a reliable biomarker for AKI associated with sepsis.
Identification of biomarker cut-off levels
By examining biomarker levels in periods of no AKI and AKI (Fig. 4) and the ROC analysis (Fig. 5), we concluded that Cys-C and pNGAL were the best performing biomarkers to diagnose AKI in our study group. We proceeded to identify biomarker cut-off values, including Cys-C and pNGAL for patients without sepsis or pneumonia, and Cys-C alone for patients with sepsis or pneumonia. Proposed biomarker cut-off levels were set according to the sensitivities and specificities provided by the ROC analyses. From these analyses we identified a Cys-C value of >0.91 mg/L (75 % sensitivity, 82 % specificity) and a pNGAL value of >258 ng/mL (88 % sensitivity, 62 % specificity) as cut-off levels for AKI.
A biomarker panel improves specificity for AKI
To test the hypothesis that biomarker specificity may be improved by introducing a requirement for abnormal Cys-C and pNGAL values to be recorded at the same time, patients without sepsis or pneumonia were split into three groups (Fig. 1b). pNGAL was not considered here as our results had already shown that it is not a reliable AKI marker in patients with sepsis or pneumonia. Group 1 contained patients not experiencing an AKI episode during PICU admission, Group 2 contained patients experiencing AKI on PICU admission (to test the hypothesis in patients with established AKI) and Group 3 consisted of patients with no AKI evident on PICU admission but who subsequently experienced an AKI episode during PICU admission (to test the hypothesis in patients prior to AKI onset). For this analysis, patients with pRIFLE R or worse were considered to have AKI. We found that the combination of Cys-C and pNGAL improved the specificity of the test [92.9 %, compared with the use of either Cys-C or pNGAL alone (85.7 % and 57.1 % respectively)] (Table 2). For patients in Group 2 (established AKI), this improved specificity came without a loss of sensitivity (compared to either test in isolation). However, contemporaneous combination of the two biomarkers in Group 3 (prior to AKI) reduced sensitivity to only 20 %. One possible explanation is that Cys-C and pNGAL peak at different times prior to an episode of AKI. We therefore proceeded to examine the temporal profiling of these biomarkers.
Temporal profiling of plasma biomarkers
We investigated the time course of Cys-C and pNGAL levels in the period prior to AKI onset for the four patients in Group 3 who were admitted to PICU without sepsis or pneumonia. Using the cut-off values of >0.91 mg/L for Cys-C and >258 ng/mL for pNGAL, an abnormal pNGAL value was recorded in three of these patients prior to AKI onset, while none of the patients had an abnormal Cys-C reading prior to AKI onset. Additionally, the highest recorded pNGAL level in a given patient’s biomarker profile occurred prior to AKI onset in all four patients, compared to two of the four patients for Cys-C. In this limited data set, pNGAL levels tended to peak earlier than Cys-C levels and were often declining by the time the Cys-C levels began to rise.
Temporal profiling of urinary biomarkers
The urinary biomarkers uNGAL and KIM-1 displayed a poor ability to diagnose AKI in the general patient group (Fig. 5). However, in an exploratory analysis, we examined the possibility that these biomarkers may have utility in predicting future AKI using data from all patients developing AKI after admission to PICU. We identified urinary biomarker values from six patients (pRIFLE R or worse) in the 24 h prior to an AKI episode and compared these to median values in the 24 h post-AKI. Five of these six patients had lower uNGAL levels in the 24 h after AKI diagnosis, while five displayed higher KIM-1 levels in the 24 h after AKI diagnosis. The possibility that uNGAL may peak before the diagnosis of AKI using pRIFLE criteria and thus act as an early marker of AKI could be explored in a future, appropriately powered study.
Patients with AKI and normal eCCl
Three patients in the study reached pRIFLE AKI criteria on the basis of low urine output alone (i.e. had normal eCCl). Biomarker profile plots are provided in ESM Figs. 3–5. No pNGAL plot is provided in ESM Fig. 5 since this patient was admitted with sepsis. The remaining two patients both had pNGAL readings which exceeded our proposed pNGAL cut-off value of 258 ng/mL prior to AKI diagnosis. All three patients recorded their highest uNGAL level prior to AKI diagnosis. Again, this result suggests that pNGAL and uNGAL may be earlier and more sensitive markers of AKI than SCr.