Tubular injury and cell-cycle arrest biomarkers to predict acute kidney injury in noncritically ill children receiving aminoglycosides

Abstract

Aim:

NGAL, IL-18, KIM-1 as well as urinary TIMP2 and IGFBP7 and their mathematical product (TIMP2*IGFBP7) were evaluated for detecting pediatric aminoglycoside acute kidney injury (AG-AKI).

Methods:

In a prospective study, noncritically ill children received aminoglycosides (AG) ≥3 days. The area under the curve (AUC) for biomarkers to detect AKI was calculated by a) days before AKI onset; b) treatment days.

Results:

There were 113 AG episodes (68% febrile neutropenia). The AKI group had a higher proportion with febrile neutropenia. The AKI group had significantly lower NGAL 3 days before AKI, as patients with febrile neutropenia had a lower NGAL during AG treatment (p < 0.05). NGAL, IL-18 and TIMP2*IGFBP7 had AUC ≥0.73 at 3, 2 and 2 days before AKI onset.

Conclusion:

NGAL, IL-18 and TIMP2*IGFBP7 were modest early biomarkers of AG-AKI. Febrile neutropenia was associated with lower NGAL.

Nephrotoxicity is one of the most prevalent causes of acute kidney injury (AKI) in hospitalized patients [1]. Many adult studies have shown that AKI is associated with increased risk for chronic kidney disease, hypertension and mortality [2]. Recent data suggest that children treated with nephrotoxic medications during hospitalization who develop AKI, are at an increased risk for signs of chronic kidney disease at 6 months after discharge, compared with children who did not develop nephrotoxic AKI [3]. One of the most common and well-described nephrotoxic medications used in children are aminoglycosides (AGs) [4]. Although AG's are highly effective for treating Gram-negative bacterial infections, they are known to selectively damage cochlear hair cells and proximal tubular epithelial cells. Approximately 20–33% of noncritically ill, hospitalized children are treated with AG developing AKI [4].

AKI is currently diagnosed by a rise in serum creatinine (SCr), or a decrease in urine output [5]. However, SCr is a suboptimal biomarker for AKI. SCr only rises after substantial kidney reserve is lost and its concentrations are affected by factors independent of kidney function, including muscle mass and age [6]. Cessation of nephrotoxic medication in the early phases of AKI may increase the likelihood of reversibility and recovery from the injury, and reduce morbidity associated with AKI [7]. Thus, there is a need for earlier diagnosis of AG-induced AKI (AG-AKI) to enable more timely intervention to mitigate AKI severity and progression.

NGAL and KIM-1 have been widely studied for early AKI diagnosis in patients admitted to the intensive care unit, undergoing cardiac surgery, or in patients treated with AG for cystic fibrosis respiratory exacerbations [8–12]. They have been shown to be indicative of direct intrinsic tubule injury in animal studies of ischemic and nephrotoxic AKI [13,14]. While NGAL mRNA is primarily induced in the distal and collecting nephron during AKI, KIM-1 mRNA is upregulated in proximal tubular cells, where AG exerts significant toxicity [15]. IL-18 is also a proximal tubular damage marker and has been studied for early AKI diagnosis in several critically-ill hospitalized populations [16,17]. Given these biomarkers' association with kidney injury location and previous studies showing them as early diagnostic tests for AKI in other cohorts, they may be promising for AG-AKI diagnostic tests.

More recently, studied AKI diagnostic biomarkers are two markers of the cell-cycle arrest termed urinary TIMP2 and IGFBP7, and their mathematical product (i.e., TIMP2*IGFBP7) [18]. These biomarkers have shown promise for early AKI diagnosis in the context of septic shock, ICU, emergency room and cardiac surgery patients [19–21]. TIMP2*IGFBP7 has also recently been approved in the USA as a clinical test for AKI diagnosis in critically ill adults [22]. However, there is limited research of this biomarker in pediatric settings, and none in the context of AG-AKI.

The main objectives of this study were to determine excretion of biomarkers of direct tubular injury and cell-cycle arrest during AG treatment in noncritically children, and to determine the extent to which these biomarkers detect AKI and severe AKI earlier than SCr rise.

Methods

Design, setting & patient recruitment

This was a prospective cohort study conducted at the Montreal Children's Hospital (Montreal, Quebec, Canada) and Cincinnati Children's Hospital Medical Center (OH, USA) from 2008 to 2012. Patients in nonintensive care unit wards who were prescribed AG were identified every 1–3 days by the pharmacy department. Children aged 3 months–18 years old who expected to receive AG for ≥3 days were eligible for inclusion. Children admitted for kidney-related diagnoses (e.g., urinary tract infection), with known pre-existing kidney conditions (genito-urinary infection, chronic kidney disease, chronic kidney diagnoses, dialysis, transplant, tubulopathy), or prior treatment with AG were excluded. Patients were approached for the study if they were within 48 h of the first AG dose. Children ultimately treated with AG for less than 3 days, or with <1 SCr measured per 5 AG treatment days were excluded from analyses post hoc. Patients with AKI present on the first day of AG treatment (i.e., SCr at start of AG ≥ 50% higher than baseline SCr [bSCr], determined after data collection) were excluded from the analysis aimed at predicting AKI (as it was already present at treatment start), but retained for describing biomarker excretion patterns. Parent or guardian informed consent and child assent (if ≥7 years old) were obtained prior to initiating study procedures. Institutional ethical board approval from the McGill University Health Centre Research Institute and Cincinnati Children's Hospital Medical Center were obtained prior to initiating study activities.

Study procedure

Specimens were collected daily from recruitment to 3 days after AG treatment, for a maximum of 14 days. Participants were asked to provide a daily urine specimen (up to 30 cc) by clean catch, foley, or cotton balls in subjects wearing diapers. An attempt to obtain an extra 1.0–1.3 cc of blood during routine blood work was made daily. Leftover serum from routine blood draws was also retrieved from the hospital laboratory on days of missed study blood collection. Blood and urine specimens were centrifuged (2000 rpm, 15 min), aliquoted and stored at 80°C with emergency back-up power and continuous electronic monitoring. All plasma and urine biomarkers were measured on the first freeze–thaw cycle.

Clinical data collection

Clinical variables collected from charts include: demographic information, medical/medication history, medication during study, diagnoses, AG treatment details and reason for AG treatment. Data entry occurred periodically into a standardized database and queries were sent to study coordinators to clarify inconsistencies. To ensure consistency and accuracy, bSCr, SCr, height and treatment days were re-reviewed prior to statistical analysis.

Analyte measurements

SCr was measured with an isotope dilution mass spectrometry traceable assay at the Montreal Children's Hospital central biochemistry laboratory. Urine NGAL and IL-18 were measured at the Cincinnati Children's Hospital Medical Center Biomarker Laboratory, Ohio [23,24] with commercial ELISA kits per manufacturer's instructions. The NGAL (ng/ml, Bioporto, Gentofte, Denmark) and IL18 (pg/ml, Medical & Biological Laboratories, Nagoya, Japan) kits had an average intra/inter coefficient of variations were between 2.5–6.3%, and 3.0–10%, respectively. Measurements were performed while blinded to clinical data.

In 2019, additional banked urine samples were measured for KIM-1, TIMP2 and IGFBP7. Due to previous analyses, sample availability was limited. Urine samples from AG treatment day 2–4 were selected for measurement from all patients. In addition, for AKI patients, urine samples from 3 days before to the day of AKI onset were also selected. If less than two urine samples were available per patient using these identification methods, random urine samples (on AG treatment) were selected until all patients had at least two urine samples available to maximize the sample size. Urinary KIM-1, TIMP2 and IGFBP7 (pg/ml) were measured at the Cincinnati Children's Hospital Medical Center Biomarker Laboratory, Ohio, using ELISA kits (R&D Systems, Inc., MN, USA) per manufacturer's instructions. The inter and intra-assay CV's ranges were 2 and 7.8% for KIM-1, 5.3 and 8.6% for TIMP2, and 4.6 and 9.8% for IGFBP7 respectively. TIMP2 and IGFBP7 were expressed as ng/ml in the primary analyses. Laboratory investigators performing measurements were blinded to clinical outcome.

Outcome definitions

The primary outcome was occurrence of any stage AKI during AG treatment, defined based on the Kidney Disease: Improving Global Outcomes (KDIGO) SCr-based definition (≥50% SCr rise from baseline within 7 days or ≥0.3 mg/dl SCr rise within 48 h). A secondary outcome of the study was severe AKI (doubling of SCr from baseline or worse, need for dialysis, or an estimated glomerular filtration rate (eGFR) of <35 ml/min/1.73 m²) [5]. KDIGO urine output criteria was not used as noncritically ill patients are rarely catheterized, therefore, the accuracy of urine output data is limited. Moreover, AG-AKI is nonoliguric [25]. bSCr was defined as the lowest SCr value in the 3 previous months. In subjects with no available bSCr, bSCr was estimated by back-calculation using the Chronic Kidney Disease in Children (CKiD) bedside eGFR equation, assuming normal baseline glomerular filtration rate (GFR) (120 ml/min per 1.73 m2 in children over 2 years of age) [26,27]. For children aged under 2 years old, age-normative eGFR values were used instead [28]. When height (required for CKiD equation) and bSCr were unknown, the Hoste equation (height independent) was used to back-calculate bSCr (previously validated for use to estimate bSCr in AKI cohorts) [29]. For patients with recorded bSCr, eGFR was back-calculated with CKiD and Hoste equations, depending on availability of height. Both CKiD and Hoste equations have been validated in a Montreal population with wide range of GFR [30].

Statistical analyses

Variables were described as mean ± standard deviation (SD) (median), or proportion (n [%]) as appropriate based on variable distribution. Clinical variables were compared between AKI versus non-AKI patients using appropriate univariable analyses.

NGAL and IL-18 were examined when expressed as concentrations (i.e., per ml), and normalized to urine creatinine (uCr). KIM-1, TIMP2, IGFBP7 and TIMP2*IGFBP7 were only examined as concentrations to be consistent with prior research and due to a significant reduction in sample numbers when normalized to uCr.

Biomarkers were examined for early AKI diagnosis by evaluating AKI urine samples from 3 days before onset of AKI defined by SCr rise (Day -3 AKI) to the day of AKI onset (Day 0 AKI). For comparison to AKI urine biomarker concentrations, in non-AKI patients (who did not have AKI onset from which to anchor sample measurement), it was necessary to select appropriate urine samples. Since the average day of AKI onset was treatment day 3 and we were measuring AKI patient biomarkers from 3 days before AKI onset, we used biomarkers on AG treatment day 4 in non-AKI patients to compare with biomarker concentrations on the first day of AKI onset in AKI patients. Thus, in non-AKI patients, biomarker concentrations from AG treatment days 1–3 were used to compare with AKI patient biomarkers concentrations from Days -3 to -1 AKI in AKI patients. When a urine sample was not available from a specific AG treatment day in non-AKI patients, an alternative AG treatment day with urine specimens available was selected. For a more clinical approach, we evaluated biomarker concentrations from AG treatment days 2–7. Treatment day 1 was excluded due to low sample number. Biomarkers concentrations, stratified by day relative to AKI onset or treatment day, were plotted and compared between non-AKI and AKI groups (Mann–Whitney U test) and across days (Kruskal–Wallis test).

Biomarker diagnostic characteristics for predicting and discriminating for AKI and severe AKI, when drawn from 3 days before AKI to the day of AKI onset, and from AG treatment days 2, 3 and 4, were evaluated by calculating the area under the receiver operating curve (AUC) with associated 95% CI. For treatment day, biomarkers were only included from patients who had not yet developed AKI on the day of biomarker measurement, to avoid over-inflating biomarker AUC's (i.e., if a patient had AKI on AG treatment day 4, then AG treatment day 4 biomarker concentrations were not included in the analysis for AKI prediction).

Youden index (maximal combined sensitivity and specificity) was measured for an optimal threshold biomarker concentration to predict AKI. Additionally, combinations of biomarkers were examined for ability to predict AKI by calculating AUC from multiple logistic regression with different biomarker combinations.

Because febrile neutropenia is characterized by very low neutrophil counts and urinary NGAL may be partially neutrophil derived, we performed secondary analysis to compare NGAL (and other biomarkers) excretion in patients with versus without febrile neutropenia. All analyses were performed using STATA 12^®, College Station, TX, USA.

Results

Subject characteristics, AKI incidence & risk factors

Figure 1 shows the study flow. 113 treatment episodes were included, and 88 treatment episodes were studied to evaluate early AKI biomarker diagnosis. AKI occurred in 47/113 (41.6%) and severe AKI occurred in 27/113 (23.9%) of AG treatment episodes. 25 of 47 (53.2%) AKI patients had AKI on the first day of AG treatment.

Figure 1. . Study flow chart.

The flowchart displays total number recruited in the beginning to study, to final cohorts for analysis.

AG: Aminoglycoside; AKI: Acute kidney injury; SCr: Serum creatinine.

Table 1 shows that participants with AKI were more likely to have received concurrent, other nephrotoxic medication during AG treatment, to receive tobramycin and be under the hematology-oncology service. The AKI group were also more likely to have febrile neutropenia as the primary reason for AG treatment (83% of the AKI participants had febrile neutropenia). Higher baseline eGFR was associated with AKI (Median of AKI vs non-AKI patients: 287 vs120 ml/min/1.73 m², p < 0.001). Forty-four patients had baseline eGFR (>200 ml/min per 1.73 m²), all calculated from measured bSCr, as most (94%) of these eGFR calculations used the widely utilized CKiD equation (height dependent). The AKI group also had a significantly higher proportion of patients with concurrent nephrotoxic medications. Median (IQR) number of SCr values available per AG treatment day was 0.75 (0.2) for patients with AKI (approximately three SCr measures every 4 treatment days) and 0.67 (0.4) (approximately two SCr measures every 3 treatment days) for non-AKI patients.

Table 1. . Patient characteristics.

Parameters	All (n = 113)	No AKI (n = 66)	AKI (n = 47)
Continuous variables expressed as mean ± SD (median); categorical variables expressed as n (%)
Age (years)	7.89 ± 4.90 (8.44)	7.86 ± 5.34 (8.72)	7.93 ± 4.27 (8.44)
Hospital days before AG treatment	2.11 ± 0.59 (2)	2.11 ± 0.56 (2)	2.11 ± 0.63 (2)
Length of AG treatment (h)	182 ± 191 (145)	166 ± 98.9 (143)	203 ± 268 (151)
Length of AG treatment (days)	8.01 ± 7.52 (7)	7.55 ± 3.97 (7)	8.79 ± 10.7 (7)
Male gender	55 (48.7)	33 (50.0)	22 (46.8)
AG name†
– Gentamicin	18 (15.9)	15 (22.7)	(6.38)
– Tobramycin	93 (82.3)	50 (75.8)	43 (91.5)
– Amikacin	2 (1.77)	1 (1.52)	1 (2.13)
Reason for AG treatment‡
– Suspected or documented infection	29 (25.7)	21 (31.8)	8 (17.0)
– Cystic fibrosis	7 (6.19)	7 (10.6)	0 (0.00)
– Febrile neutropenia	77 (68.1)	38 (57.6)	39 (83.0)
Service‡
– General pediatric	15 (13.3)	14 (21.2)	1 (2.13)
– Surgery	9 (7.96)	7 (10.6)	2 (4.26)
– Hematology/oncology	89 (78.8)	45 (68.2)	44 (93.6)
Documented infection	54 (47.8)	32 (48.5)	22 (46.8)
Concurrent nephrotoxic medications†	80 (70.8)	41 (62.1)	39 (83.0)
Length of stay (days)	21.9 ± 36.4 (9)	22.7 ± 36.3 (10)	20.9 ± 36.8 (9)
eGFR‡	218 ± 153 (171)	140 ± 62.2 (120)	329 ± 173 (287)
Number of SCr per treatment day†	0.67 ± 0.22 (0.70)	0.67 ± 0.22 (0.67)	0.75 ± 0.20 (0.75)

^†p-value <0.05 for difference between AKI vs non-AKI groups.

^‡p-value <0.01 for difference between AKI vs non-AKI groups.

AG: Aminoglycoside; AKI: Acute kidney injury; eGFR: Estimated glomerular filtration rate; SCr: Serum creatinine.

Biomarker excretion by AKI status, relative to day of AKI onset

Figure 2 demonstrates biomarker concentration patterns, comparing AKI versus non-AKI patients, from 3 days before to the day of AKI onset. NGAL was lower in the AKI group at Day -3 (4.08 vs 57.1 ng/ml, p < 0.05), however, NGAL did not remain lower in the AKI group, and there was no difference in AKI versus non-AKI NGAL concentrations thereafter (Figure 2A). There was a nonsignificant rise in median NGAL concentrations after Day -3 in the AKI group but no appreciable change in the non-AKI group (Figure 2A). IL-18 was higher in the AKI group at Day -2 (105.5 versus 56.9 pg/ml, p < 0.01), after which concentrations dropped, as there was no appreciable change in IL-18 concentrations in the non-AKI group (Figure 2B). In AKI patients, KIM-1 concentration rose gradually (but not significant) from Day -3 to Day 0, and on Day 0, the AKI group had higher KIM-1 concentrations (533.3 versus 262.7 ng/ml, p < 0.01; Figure 2C). In general, there were nonsignificant higher concentrations of TIMP2, IGFBP7 and TIMP2*IGFBP7 in the AKI group from Day -3 to Day 0 (Figure 2D–F); however, at Day -2, TIMP2*IGFBP7 was higher in the AKI group (0.478 vs 0.366 (ng/ml)²/1000, p < 0.05; Figure 2F). Overall, when NGAL and IL-18 were expressed as normalized to uCr, biomarker excretion followed a similar pattern (Supplementary Figure 1).

Figure 2. . Biomarker excretion relative to the day of AKI onset (from day -3 to day 0 AKI).

Box plots are displayed with median (middle line), 25th and 75th percentiles (lower and upper borders of box respectively), stratified by non-AKI (white) and AKI (gray) status. NGAL (A), IL18 (B), KIM1 (C), TIMP2 (D), IGFBP7 (E) and TIMP2*IGFBP7 (F) are stratified by days before AKI onset (i.e., day -3 AKI: biomarker measured 3 days before AKI onset in AKI patients compared with a non-AKI biomarker measurement, as described in methods). Nonparametric tests used to compare between non-AKI and AKI groups (Mann–Whitney U), as well as across days (Kruskal–Wallis).

AKI: Acute kidney injury.

Biomarker excretion by AKI status, described by AG treatment day

Figure 3 shows biomarker concentrations from AG treatment days 2–7, in AKI and non-AKI groups. There was little evidence of AKI versus non-AKI differences in NGAL or IL-18 concentrations across AG treatment days (Figure 3A & B). KIM-1 was higher in the AKI group on treatment day 2 (884.9 vs 490.1 pg/ml, p < 0.05), treatment day 4 (411.9 vs 286.7 pg/ml, p < 0.05) and treatment day 5 (378 vs 117.3 pg/ml, p < 0.05). TIMP2 was generally higher in the AKI group across treatment days, particularly on treatment day 3 (2.17 vs 1.63 ng/ml, p < 0.05; Figure 2D). Patterns in AKI versus non-AKI biomarker concentration by treatment day for IGFBP7 and TIMP2*IGFBP7 were inconsistent (Figure 2E & F). Supplementary Figure 2A shows that NGAL/uCr was lower in the AKI group on treatment days 2 and 5 (p < 0.05) and remained generally lower across other treatment days in the AKI group. There was little AKI versus non-AKI difference in IL-18/uCr concentrations across treatment days (Supplementary Figure 2B).

Figure 3. . Biomarker excretion across aminoglycoside treatment days 2–7.

Box plots are displayed with median (middle line), 25th and 75th percentiles (lower and upper borders of box respectively), stratified by non-AKI (white) and AKI (gray) status. Biomarkers NGAL (A), IL18 (B), KIM1 (C), TIMP2 (D), IGFBP7 (E) and TIMP2*IGFBP7 (F) are plotted. Nonparametric tests used to compare between non-AKI and AKI groups (Mann–Whitney U), as well as across days (Kruskal–Wallis).

AKI: Acute kidney injury.

Biomarkers as early diagnostic tests

Table 2 shows the biomarker excretion values and diagnostic characteristics of biomarkers to predict and discriminate for AKI and severe AKI, from 3 days before AKI onset to the day of AKI onset. NGAL from Day -3 predicted AKI development with an AUC of 0.80 (0.60–1.00) (with lower NGAL values being predictive). At Day -2, IL-18 demonstrated an AUC of 0.80 (0.67–0.94) to predict AKI. KIM-1 had an AUC of 0.70 (0.49–0.91) to predict AKI on Day -2, and AUC of 0.68 (0.52–0.82) to discriminate for AKI on Day 0. At Day -2, TIMP2*IGFBP7 had an AUC of 0.73 (0.58–0.88) to predict AKI. Other AUC's are shown in Table 2. When the outcome was severe AKI (right side of Table 2), AUC's were generally stronger for predicting severe AKI, compared with predicting any AKI. Biomarkers expressed as normalized to uCr (NGAL and IL-18) showed similar findings (Supplementary Table 1).

Table 2. . Area under the receiver operating characteristics curve to predict acute kidney injury, relative to the day of acute kidney injury onset.

Day relative to AKI onset	AUC (95% CI) to predict any stage AKI	AUC (95% CI) to predict severe AKI
NGAL (ng/ml)
Day -3 AKI	0.80 (0.60–1.00)	0.84 (0.74–0.94)
Day -2 AKI	0.43 (0.18–0.68)	0.47 (0.20–0.74)
Day -1 AKI	0.54 (0.37–0.71)	0.51 (0.23–0.80)
Day 0 AKI	0.51 (0.34–0.68)	0.49 (0.27–0.70)
IL18 (pg/ml)
Day -3 AKI	0.61 (0.34–0.88)	0.55 (0.08–1.00)
Day -2 AKI	0.80 (0.67–0.94)	0.81 (0.63–0.99)
Day -1 AKI	0.54 (0.32–0.76)	0.44 (0.00–1.00)
Day 0 AKI	0.57 (0.40–0.73)	0.53 (0.23–0.82)
KIM1 (pg/ml)
Day -3 AKI	0.55 (0.33–0.78)	0.56 (0.34–0.79)
Day -2 AKI	0.70 (0.49–0.91)	0.66 (0.40–0.92)
Day -1 AKI	0.60 (0.39–0.81)	0.65 (0.27–1.00)
Day 0 AKI	0.68 (0.53–0.82)	0.62 (0.29–0.95)
TIMP2 (ng/ml)
Day -3 AKI	0.51 (0.16–0.86)	0.80 (0.41–1.00)
Day -2 AKI	0.68 (0.54–0.83)	0.72 (0.54–0.91)
Day -1 AKI	0.63 (0.47–0.80)	0.73 (0.45–1.00)
Day 0 AKI	0.53 (0.38–0.68)	0.53 (0.31–0.74)
IGFBP7 (ng/ml)
Day -3 AKI	0.63 (0.37–0.90)	0.76 (0.45–1.00)
Day -2 AKI	0.67 (0.49–0.86)	0.60 (0.33–0.87)
Day -1 AKI	0.43 (0.26–0.60)	0.55 (0.35–0.74)
Day 0 AKI	0.62 (0.47–0.78)	0.74 (0.54–0.94)
TIMP2*IGFBP7 (ng/ml)2/1000
Day -3 AKI	0.58 (0.31–0.86)	0.56 (0.28–0.85)
Day -2 AKI	0.73 (0.58–0.88)	0.70 (0.49–0.91)
Day -1 AKI	0.38 (0.21–0.55)	0.46 (0.23–0.69)
Day 0 AKI	0.39 (0.24–0.54)	0.73 (0.51–0.94)

Results stratified by days before AKI onset (i.e., Day -3 AKI: biomarker measured 3 days before AKI onset in AKI patients compared with a non-AKI biomarker measurement, as described in methods).

Bolded AUCs: 95% CI did not cross 0.5.

AKI: Acute kidney injury; AUC: Area under the curve.

When analyzing biomarkers by AG treatment day (Table 3), NGAL on treatment day 4 had an AUC of 0.69 (0.40–0.98) to predict AKI. IGFBP7 and TIMP2*IGFBP7 had AUC's of 0.69 (0.53–0.85) and 0.72 (0.57–0.87) to predict AKI, respectively, when measured on AG treatment day 2. Other biomarkers alone were not predictive of AKI and again, when the outcome was severe AKI, AUC's were generally much higher (right side of Table 3). Biomarkers expressed as normalized to uCr (NGAL and IL-18) showed similar findings (Supplementary Table 2). Biomarker threshold concentrations associated with the Youden index (maximizing sensitivity and specificity) on different study days are shown in Supplementary Table 3 & 4.

Table 3. . Area under the receiver operating characteristics curve to predict acute kidney injury, relative to aminoglycoside treatment day.

Treatment day	AUC (95% CI) to predict any stage AKI	AUC (95% CI) to predict severe AKI
NGAL (ng/ml)
2	0.43 (0.25–0.61)	0.40 (0.19–0.62)
3	0.39 (0.18–0.61)	0.47 (0.03–0.91)
4	0.69 (0.40–0.98)	0.87 (0.76–0.98)
IL18 (pg/ml)
2	0.63 (0.39–0.86)	0.59 (0.19–0.99)
3	0.33 (0.17–0.49)	0.81 (0.70–0.91)
4	0.52 (0.28–0.76)	0.52 (0.00–1.00)
KIM1 (pg/ml)
2	0.66 (0.45–0.86)	0.59 (0.23–0.94)
3	0.49 (0.23–0.76)	0.42 (0.00–0.92)
4	0.53 (0.34–0.70)	0.46 (0.21–0.71)
TIMP2 (ng/ml)
2	0.64 (0.48–0.80)	0.66 (0.45–0.87)
3	0.69 (0.49–0.88)	0.70 (0.19–1.00)
4	0.64 (0.44–0.83)	0.72 (0.40–1.00)
IGFBP7 (ng/ml)
2	0.69 (0.53–0.85)	0.71 (0.47–0.94)
3	0.46 (0.25–0.67)	0.48 (0.07–0.89)
4	0.60 (0.33–0.87)	0.71 (0.25–1.00)
TIMP2*IGFBP7 (ng/ml)2/1000
2	0.72 (0.57–0.87)	0.76 (0.55–0.96)
3	0.39 (0.17–0.62)	0.40 (0.00–0.90)
4	0.49 (0.27–0.72)	0.54 (0.18–0.90)

Bolded AUCs: 95% CI did not cross 0.5.

AKI: Acute kidney injury; AUC: Area under the curve.

Biomarker combinations

Table 4 shows biomarker combinations that resulted in AUC ≥0.75 for predicting AKI. Combining data from different biomarkers generally resulted in higher AUC's for predicting AKI. Combining NGAL and TIMP2*IGFBP7 resulted in AUC 0.82 to predict AKI when measured on Day -3. AUC's were much higher for predicting severe AKI (though sample size was limited). Multiple combinations of biomarkers demonstrated AUC > 0.90 to predict severe AKI, with the highest achieved by the combination of KIM-1 and TIMP2 (AUC: 0.95) measured on Day -3.

Table 4. . Biomarker combinations resulting in area under the receiver operating characteristic curve.

Day relative to AKI onset	AUC for any stage AKI (95% CI)	AUC for severe AKI (95% CI)
-3	NGAL + IL18 - 0.79 (0.62–0.95) NGAL + KIM1 - 0.80 (0.63–0.97) NGAL + TIMP2 - 0.82 (0.62–1.00) NGAL + IGFBP7 - 0.82 (0.61–1.00) NGAL + TIMP2*IGFBP7 - 0.82 (0.62–1.00)	NGAL + KIM1 - 0.85 (0.74–0.97) NGAL + IL18 - 0.80 (0.68–0.92) NGAL + TIMP2 - 0.94 (0.83–1.00) IL18 + TIMP2 - 0.85 (0.66–1.00) KIM1 + TIMP2 - 0.95 (0.83–1.00) NGAL + IGFBP7 - 0.85 (0.66–1.00) IL18 + IGFBP7 - 0.75 (0.41–1.00) KIM1 + IGFBP7 - 0.79 (0.49–1.00) NGAL + TIMP2*IGFBP7 - 0.80 (0.66–0.94)
-2	NGAL + IL18 - 0.81 (0.70–0.93) IL18 + KIM1 - 0.79 (0.62–0.97) IL18 + TIMP2 - 0.77 (0.63–0.91) IL18 + IGFBP7 - 0.80 (0.65–0.95) IL18 + TIMP2*IGFBP7 - 0.80 (0.65–0.94) KIM1 + TIMP2 - 0.76 (0.60–0.92)	IL18 + KIM1 - 0.82 (0.62–1.00) NGAL + IL18 - 0.83 (0.67–0.98) NGAL + TIMP2 - 0.75 (0.56–0.95) IL18 + TIMP2 - 0.82 (0.68–0.96) KIM1 + TIMP2 - 0.78 (0.58–0.98) IL18 + IGFBP7 - 0.80 (0.61–1.00) IL18 + TIMP2*IGFBP7 - 0.80 (0.60–1.00)
-1	–	IL18 + KIM1 - 0.77 (0.57–0.97) NGAL + TIMP2 - 0.78 (0.57–1.00) KIM1 + TIMP2 - 0.90 (0.80–0.99)
0	–	IL-18 + IGFBP7 - 0.75 (0.57–0.93) KIM1 + IGFBP7 - 0.75 (0.49–1.00) IL18 + TIMP2*IGFBP7 - 0.75 (0.59–0.91)
Treatment day	AUC for any stage AKI	AUC for severe AKI
3	KIM1 + TIMP2 - 0.75 (0.59–0.92)	–
4	NGAL + KIM1 - 0.75 (0.50–1.00) NGAL + IGFBP7 - 0.78 (0.49–1.00) NGAL + TIMP2*IGFBP7 - 0.77 (0.48–1.00) KIM1 + TIMP2 - 0.82 (0.70–0.93)	NGAL + IL18 - 0.87 (0.76–0.98) NGAL + KIM1 - 0.90 (0.80–1.00) NGAL + TIMP2 - 0.92 (0.84–1.00) NGAL + IGFBP7 - 0.91 (0.82–1.00) NGAL + TIMP2*IGFBP7 - 0.90 (0.80–1.00) IL18 + TIMP2 - 0.81 (0.70–0.93) KIM1 + TIMP2 - 0.84 (0.63–1.00)

≥0.75 to predict or discriminate for acute kidney injury (left) and for severe acute kidney injury (right), when measured relative to the day of acute kidney injury onset (top) and when measured relative to aminoglycoside treatment day (bottom). Results stratified by days relative to AKI onset (upper table) or aminoglycoside treatment day (lower table).

Bolded AUCs: 95% CI did not cross 0.5.

AKI: Acute kidney injury; AUC: Area under the curve.

NGAL & febrile neutropenia

Figure 4 shows that patients with febrile neutropenia had a lower NGAL concentration across AG treatment days 2–6 (p < 0.05; Figure 4A). Similar results were observed with NGAL/Cr (Supplementary Figure 3A). In general, other biomarker concentrations were very similar between patients with versus without febrile neutropenia (Figure 4B–D & F), as the exception was IGFBP7 on treatment days 2 and 3, which was lower in patients with febrile neutropenia (p < 0.05; Figure 4E).

Figure 4. . Biomarker excretion stratified by febrile neutropenia status, from aminoglycoside treatment days 2–7.

Box plots are displayed with median (middle line), 25th and 75th percentiles (lower and upper borders of box, respectively), stratified by febrile neutropenia (gray) and other reason for aminoglycoside treatment (white). Biomarkers NGAL (A), IL18 (B), KIM1 (C), TIMP2 (D), IGFBP7 (E) and TIMP2*IGFBP7 (F) are plotted. Nonparametric tests used to compare between groups (Mann–Whitney), as well as across days (Kruskal–Wallis).

AKI: Acute kidney injury.

Discussion

This is the first study of a cell-cycle arrest biomarker study in noncritically ill children. This is also the first study of renal injury biomarkers for predicting AKI performed in noncritically ill children receiving AG, not limited to only cystic fibrosis patients. We found that several AKI biomarkers are predictive of AG-AKI when measured at certain time points before AKI onset, that urine NGAL is lower in patients with AG-AKI and that when combining multiple biomarkers, prediction of AG-AKI is substantially higher. A mainstay of AKI management is early identification, in order to implement preventative measures in a timely fashion, to reduce morbidity [31]. As recommended by the Acute Dialysis Quality Initiative consensus, a goal of AKI research and treatment should include incorporating injury biomarkers to effectively diagnose AKI [32]. The costs of these biomarkers are currently relatively high. If AKI biomarkers can be maximally utilized (by applying them to the right patients, at the right time) and become integrated in routine clinical care laboratories, the cost of implementation will hopefully ultimately translate to reduced patient and healthcare system costs. This study attempted to address these goals in noncritically ill children treated with AG, who are known to be at high risk for AKI.

The incidence of AKI was 42% in 113 AG episodes, which is higher than previously reported rates in pediatric AG settings [1,27]. This may be due to usage of differing AKI definitions in other studies [33]. Also, our selection criterion to only include patients treated with AG for at least 3 days, likely led to a higher risk population. Half of the AKI patients had AKI on the first day of AG treatment. Comparatively, in a previous retrospective study on AG treatment in non-critically ill patients, up to 35% of patients had AKI on the first AG treatment day [1]. This suggests that children initiating treatment with AGs may already have AKI and are likely to be associated with their underlying illness or volume depletion. However, it remains unclear whether the high proportion of AKI on first day of AG treatment was ‘true AKI’ (with renal tubular injury) or simply a temporary decreased GFR from hypovolemia. Future studies could examine biomarkers to evaluate whether they can be used to distinguish ‘true AKI’ from temporary SCr rise, due to volume depletion. The AKI group also had a significantly higher proportion of participants with concurrent nephrotoxic medication treatment. This has been similarly seen in other cohorts (e.g., ICU patients), which highlights the consideration of other nephrotoxic medications as additional risk factors for AKI [34].

The goal of this study was to assess novel biomarkers of renal injury for predicting any stage AKI and severe AKI, prior to SCr rise. NGAL, IL18 and TIMP2*IGFBP7 demonstrated AG-AKI predictive behaviors only on specific days before AKI onset. KIM-1 was a later biomarker of AG-AKI. This latter finding has been similarly noted in pediatric cardiac surgery patients, where NGAL and IL-18 rose earlier, compared with KIM-1 and TIMP2*IGFBP7 [35]. NGAL and IL-18 were also predictive of severe AKI in early AG treatment. The higher AUC for severe AKI likely represents less misclassification of severe AKI (compared with any AKI). Combinations of biomarkers were found to better predict AKI than single biomarkers alone, particularly the combination of KIM-1 with TIMP2, 3 days before AKI. This is consistent with other studies in children after cardiac surgery, where combining biomarkers significantly increased the AUC's [35,36].

A surprising finding was the inverse association between NGAL and AKI. In other cohorts, NGAL has been described to be significantly higher in the AKI patients (neonates, pediatric cardiac surgery or in the pediatric intensive care unit) [11,35,37]. However, NGAL was found to be lower in AKI patients in our cohort. This led us to perform a secondary analysis to examine the relationship of NGAL with febrile neutropenia. Significantly lower NGAL concentrations were found in patients with febrile neutropenia, possibly since NGAL is secreted by human neutrophils. Since there was a higher proportion of patients with febrile neutropenia in the AKI group, this may explain our finding of lower NGAL in AKI patients. This study's small sample size limited the ability to investigate this, but future research should examine the use of NGAL in patients with and without febrile neutropenia. This demonstrates the importance of validating new biomarkers in different clinical and population characteristic contexts, prior to clinical application. Another potential explanation is that AG results primarily in proximal tubule injury, whereas NGAL mRNA and protein are induced exclusively in the distal and collecting nephron in animal and human models of AKI [38]. It is also likely that AG leads to an alternative injury mechanism compared with other causes of AKI, resulting in differing NGAL profiles. NGAL can be released as a compensatory mechanism to ameliorate toxicity by promoting cell survival and proliferation. If this mechanism was affected by AG, it may result in lower NGAL levels in patients with greater injury, leading to AKI. In mice studies, NGAL has been shown to inhibit apoptosis in tubular epithelial cells in endotoxemia and ischemia-induced AKI [39,40]. Our findings of significantly lower urine NGAL at Day -3 in subjects who subsequently developed AG-AKI may be reflective of the well-documented nephroprotective effects of NGAL [38].

Clearly established AKI-diagnostic thresholds for TIMP2*IGFBP7 have been published and exist as clinical tests for AKI in critically ill adults. A currently validated threshold for TIMP2*IGFBP7 is 0.3 (ng/ml)²/1000. In children undergoing cardiac surgery, a higher threshold value of 0.7 (ng/ml)²/1000 was found to be most diagnostic [41]. Our study, performed in noncritically ill children, demonstrated a much lower concentration threshold (∼0.053 (ng/ml)²/1000), demonstrating that optimal biomarker threshold concentrations may be specific to age groups, injury mechanisms and/or illness severity.

This study had limitations. KIM-1, TIMP2 and IGFBP7 were measured 7 years after sample collection in both AKI and non-AKI patients. One study demonstrated that over 18 months of storage, KIM-1 decreased by 1.7% [42]. Another study showed that urinary KIM-1 decreased less than 2% after 5 years of storage [43]. However, no data to our knowledge is available on TIMP2*IGFBP7 stability. The small study sample size limited our ability to perform subgroup analyses and led to wide 95% CI's for AUC estimation. However, since this was the first study performed in this patient population, our findings suggest that future larger studies should be performed to evaluate clinical utility of AKI biomarkers in the setting of AG-AKI. Baseline (pre-AG) biomarker measurements were lacking in a substantial proportion of participants. Although we attempted to collect these by using leftover urine specimens collected with routine care, this problem highlights the challenge of obtaining baseline biomarker measurements in studies involving noncritically ill children. Future studies may consider collecting urine specimens for biomarkers in all admitted patients, with retroactive consent to use those specimens in recruited patients, which has been shown to be feasible in a PICU population [37]. The lack of pre-AG treatment biomarker excretion data also resulted in the inability to investigate the extent to which pre-AG biomarker concentrations are lower in patients with neutropenia and whether biomarkers, independent of neutropenia, are associated with AKI development. Future studies should examine the potential interaction between biomarkers and febrile neutropenia with regard to AKI risk and diagnosis. Another limitation is that some participants were excluded because of limited SCr measures due to our study protocol to take blood only during routine blood work (to minimize venipunctures). This may have led to a somewhat biased study population, favouring more severely ill patients. However, this finding also highlights the relative infrequency of SCr monitoring in children receiving AG, as described in other studies [1]. Half of the AKI patients were excluded from the diagnostic characteristic analysis because of AKI onset on first AG treatment day. This early onset of AKI renders it difficult to collect urine samples before AKI to assess early diagnostic ability. The resulting smaller sample size may explain the lack of consistent biomarker elevation. Many participants had low bSCr and AKI participants had higher baseline eGFR. Since AKI ascertainment is based on percent SCr rise, lower bSCr will increase the likelihood of fulfilling AKI criteria or over-diagnose AKI. Low bSCr may be a marker of low muscle mass and has also been shown to be associated with increased risk of mortality, length of ICU stay in critically ill patients and may indicate chronic illness, therefore susceptibility to developing AKI [6,44]. Unfortunately, because of the lack of data collection on muscle mass, this speculation cannot be confirmed. Higher baseline eGFR in AKI groups has been consistently described in pediatric ICU and cardiac surgery cohorts [37,45], without a satisfactory explanation. This may be a marker of chronic illness as described above, but it may also be possible that AKI patients receive more intravenous hydration initially, thereby temporarily lowering bSCr values. However, insufficient fluid data was collected to investigate this. A high proportion of patients received tobramycin, had febrile neutropenia and were under the hematology/oncology service, each of which were associated with AKI development. Thus, generalizability of our findings to other noncritically ill populations should be interpreted with caution. Our findings suggest that AKI biomarkers may be highly influenced by comorbid conditions, such as febrile neutropenia, and timing of the biomarker application. This emphasizes the need for the biomarkers to be applied at their ‘ideal’ setting. Evaluating AKI biomarkers in a minimally heterogenous group of patients, as well as incorporating a clinical prediction score for AG-AKI may help improve our understanding of how to apply these AKI biomarkers.

Conclusion

In conclusion, NGAL, IL-18 and TIMP2*IGFBP7 were found to be modest early AG-AKI biomarkers, where a lower NGAL was associated with AG-AKI. The latter finding may be due to the association febrile neutropenia with lower NGAL concentrations. Combinations of biomarkers (KIM-1 and TIMP2) were found to substantially improve predictive performance in our study. However, these results will need to be confirmed in a validation study using a separate cohort. In order to make clinical recommendations and to select the best biomarker combinations for diagnosing AKI, validation studies in larger cohorts will be required.

Summary points.

• Almost half of patients treated with aminoglycoside (AG) developed AKI.

• Half of the AKI patients developed AKI on the first aminoglycoside treatment day.

• eGFR was significantly higher in AKI patients than non-AKI patients.

• NGAL was significantly lower in AKI patients prior to AKI onset, versus non-AKI patients, which may have resulted from the higher proportion of patients with febrile neutropenia in the AKI group.

• IL-18, KIM-1 and TIMP2*IGFBP7 were higher in AKI patients than non-AKI patients at 2 days before, the day of and 2 days before AKI onset, respectively.

• NGAL, IL-18, TIMP2*IGFBP7 may be modest early biomarkers of aminoglycoside-induced AKI, when measured at 3 days before, 2 days before and 2 days before AKI onset, respectively.

• Biomarkers were generally more predicting of severe AKI (KDIGO stage 2 AKI or worse) than any stage AKI.

• Combination of biomarkers were very predictive of severe AKI before AKI onset.