Biomarker Predictors of Adverse Acute Kidney Injury Outcomes in Critically Ill Patients: The Dublin Acute Biomarker Group Evaluation Study

(1)

Patient-Oriented, Translational Research: Research Article

Am J Nephrol

Biomarker Predictors of Adverse Acute Kidney Injury Outcomes in Critically Ill Patients: The Dublin Acute Biomarker Group Evaluation Study

Blaithin A. McMahon^{a, e} Marie Galligan^b Lynn Redahan^b Terri Martin^b Edel Meaney^b Eoin J. Cotter^b Niamh Murphy^b Claire Hannon^b Peter Doran^b Brian Marsh^c Alistair Nichol^d Patrick T. Murray^b

aDepartment of Medicine, Medical University of South Carolina, Charleston, SC, USA; ^bSection of Medicine and Medical Specialties, School of Medicine, University College Dublin, Dublin, Ireland; ^cDepartment of Intensive Care Medicine, University College Dublin, Mater Misericordiae University Hospital, Dublin, Ireland; ^dDepartment of Intensive Care Medicine, University College Dublin, St. Vincent’s University Hospital, Dublin, Ireland; ^eJohns Hopkins School of Medicine, Baltimore, MD, USA

Received: February 19, 2019 Accepted: April 2, 2019 Published online: June 14, 2019

Nephrology

American Journal of

Prof. Patrick T. Murray

DOI: 10.1159/000500231

Keywords

Acute kidney injury · Biomarkers · Clinical model · Diagnosis · Outcome

Abstract

Background: The Dublin Acute Biomarker Group Evaluation (DAMAGE) Study is a prospective 2-center observational study investigating the utility of urinary biomarker combinations for the diagnostic and prognostic assessment of acute kidney injury (AKI) in a heterogeneous adult intensive care unit (ICU) population. The objective of this study is to evaluate whether serial urinary biomarker measurements, in combination with a simple clinical model, could improve biomarker performance in the diagnostic prediction of severe AKI and clinical outcomes such as death and need for renal replacement therapy (RRT). Methods: Urine was collected daily from patients admitted to the ICU, for a total of 7 post- admission days. Urine biomarker concentrations (neutrophil gelatinase-associated lipocalin [NGAL], α-glutathione S- transferase [GST], π-GST, kidney injury molecule-1 [KIM-1], liver-type fatty acid-binding protein [L-FABP], Cystatin C, creatinine, and albumin) were measured. Urine biomarkers

were combined with a clinical prediction of AKI model, to determine ability to predict AKI (any stage, within 2 days or 7 days of ICU admission), or a 30-day composite clinical outcome (RRT – or death). Results: A total of 257 (38%) patients developed AKI within 7 days of ICU admission. Of those who developed AKI, 106 (41%) patients met stage 3 AKI within 7 days of ICU admission and 208 patients of the entire study cohort (31%) met the composite clinical endpoint of in-hospital mortality or RRT within 30 days of ICU admission. The addition of urinary NGAL/albumin to the clinical model modestly improved the prediction of AKI, in particular severe stage 3 AKI (area under the curve [AUC] of 0.9 from 0.87, p = 0.369) and the prediction of 30-day RRT or death (AUC 0.83 from 0.79, p = 0.139). Conclusion: A clinical model incorporating severity of illness, patient demographics, and chronic illness with currently available clinical biomarkers of renal function was strongly predictive of development of AKI and associated clinical outcomes in a heterogeneous adult ICU population. The addition of urinary NGAL/albumin to this simple clinical model improved the prediction of severe AKI, need for RRT and death, but not at a statistically or clinically significant level, when compared to the clinical model

(2)

Introduction

Despite advances in critical care medicine, the mor- bidity and mortality rate in hospitalized patients with acute kidney injury (AKI) are still rising, in particular in those patients requiring renal replacement therapy (RRT) [1, 2]. The use of serum creatinine and urine output (functional markers) in the diagnostic criteria has limited the timely and accurate diagnosis of AKI [3, 4]. This has led to the discovery of numerous novel biomarkers, which have been assessed in various AKI models and in diverse patient populations. However, their integration into current clinical practice has been hampered by variable and often moderate diagnostic performance in heterogeneous patient populations [5–15]. Some biomarkers have failed to show a clear advantage beyond the conventional methods in clinical decision-making in patients with AKI [16].

We conducted a large, prospective, 2 center study to evaluate whether consecutive urinary biomarker measurements, and biomarker measurement in combination with standard clinical covariates, relevant to renal function and outcomes in critical illness, could improve biomarker performance in the diagnostic prediction of AKI and associated 30-day clinical outcomes in a heterogeneous adult intensive care unit (ICU) population.

Materials and Methods Aim, Design, and Study Setting

The Dublin Acute bioMArker Group Evaluation Study is a prospective observational study investigating the utility of urinary AKI biomarkers on admission to the ICUs of 2 academic medical centers in Dublin, Ireland. We prospectively enrolled 710 adults between September 2011 and December of 2014. Due to the incapacitating nature of critical illness, deferred consent to collect urinary samples was approved, assent was obtained from a legal authorized representative, and consent from the patient if they recovered prior to analysis. The study was approved by the research Ethics Committees of the Mater Misericordiae University Hospital and St. Vincent’s University Hospital, Dublin, Ireland.

Study Participants and Outcome Definitions

Urine was collected from patients admitted to the ICU for a total of 7 post-admission days. Inclusion criteria were males and females 18 years of age and older, admitted to the ICU (within the previous 24 h), and presence of an indwelling urinary catheter before enrollment. Exclusion criteria were patients on chronic RRT, patients admitted after recent (<3 months) kidney transplant, expected ICU discharge ≤24 h, patients with a recurring AKI during the same hospitalization episode in ICU, and participation in another conflicting clinical study at the time of this study. Urinary biomarkers were collected at the time of enrolment in the study, within 24 h of ICU admission. The primary outcome was develop-

ment of any stage of AKI during the first 2 or 7 days following ICU admission, defined and staged according to serum creatinine- based criteria per the Kidney Disease Improving Global Outcomes (KDIGO) criteria [17]. Hourly urine output data (but not weight- based urinary output) were available in this study; therefore, the KDIGO urinary output criteria were not used in the staging of AKI. First serum creatinine collected on ICU admission was used as the reference creatinine in order to define AKI cases that developed subsequently [18]. A set of clinical covariates relevant to renal function and to outcomes in critical illness (Acute Physiologic As- sessment and Chronic Health Evaluation [APACHE] II score, age, gender, creatinine, blood urea nitrogen [BUN], and hourly urine output) was predefined, for combination with urinary biomarker concentrations during analysis.

Urinary Biomarker Measurements

All urinary biomarkers were measured serially on ICU admission and daily for 7 days. Only biomarker measurements at the time of ICU admission were used for the prediction of AKI at 48 h or 7 days’

post ICU admission. Urinary Cystatin C assays were performed using a Roche Tina-Quant Cystatin C (Gen.2) assay (Roche GmBH, Mannheim, Germany), run on a Roche Cobas 701 instrument (assay range 0.40–6.80 mg/L, total coefficient of variation [total CV%] of

≤2.4%). Urinary neutrophil gelatinase-associated lipocalin (NGAL), microalbumin, and creatinine were performed on an Architect ci4100 analyser, using the Architect Urine NGAL assay (Abbott Di- agnostics, Wiesbaden, Germany; range 10.0–1,500 ng/mL [undiluted], limit of 6,000 ng/mL [1/4 dilution], total CV% ≤4.9%), the Ar- chitect/Aeroset Urine Microalbumin assay (Abbott Diagnostic, Wi- esbaden, Germany; range 5–500 μg/mL [undiluted], limit of 2,000 μg/mL [1/4 dilution], total CV% ≤3.79%) and the Architect Creati- nine (Enzymatic) assay (Abbott Diagnostic, Wiesbaden, Germany;

range 2.5–400 mg/dL, total CV% of ≤3.6%). Liver-type fatty acid- binding protein 1 (L-FABP1) and kidney injury molecule-1 (KIM-1) assays were performed using commercially available ELISA kits obtained from CMIC (CMIC Holdings Ltd., Tokyo, Japan) and R&D Systems (R&D Systems Limited, Minneapolis, MN, USA), respectively (L-FABP1 assay range 3–400 ng/mL intra/inter assay CV%

≤10/10%; KIM1 assay range 0.156–10.0 ng/mL, intra/inter CV% of

≤4.4/7.8%). π and α glutathione S-transferase (GST) assays were performed by Argutus Medical Ltd./EKF diagnostics Ltd. (Surrey, UK), using EKF π GST and α GST EIA (π GST assay range 3.13–100 µg/L, inter/intra assay CV% of ≤10/5.6%; α GST assay range 6.25–200 µg/L, intra/inter assay CV and of ≤5.4/8.5%). Sensitivity analysis was used to evaluate whether normalized biomarkers to urinary creatinine would impact on our conclusions [19].

Statistical Methods

Distributions of serum creatinine, urine output per hour, BUN, and all urine biomarker concentrations were right skewed, and hence were log-transformed for statistical analysis. Values below or above the limits of detection were imputed. Receiver operating char- acteristic (ROC) curve analysis was used to evaluate the diagnostic and prognostic ability of urine biomarkers for AKI and death or dialysis. ROC curves were constructed from fitted logistic regression models including predictors of the clinical covariates alone; then including as predictors clinical covariates combined with urine biomarkers. Models were fitted using repeated 10-fold cross validation (CV) with 301 repeats. Diagnostic performance is characterized using ROC metrics (area under the curve [AUC], sensitivity and spec-

(3)

ificity), summarized across the repeats of cross validation as median values with 95% CIs based on 2.5th and 97.5th percentiles. ROC curves were constructed for each repeat of CV and plotted the median line across repeats shown in bold. Each ROC curve was constructed by evaluating sensitivity and specificity over a range of probability cutoffs. DeLong’s test was used to compare the AUCs (taken as median across CV repeats) of models fitted with biomarkers and clinical covariates to models fitted with clinical covariates alone; a statistically significant result (p < 0.05) is taken as evidence of a significant difference in predictive ability of a model including biomarkers with clinical covariates and a model including clinical covariates alone. Statistical analysis was carried out in R version 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria). Logistic regression models were fitted using the caret package in R, while ROC analysis was carried out using the pROC package.

Results

Characteristics of the Study Cohort

Of 2,130 patients screened in 2 adult ICUs, 1,398 were deemed ineligible (predominantly due to anticipated death within 24 h, absence of urinary Foley catheter, or

expected discharge from ICU within 24 h) and 15 refused consent (Fig. 1). The remaining 717 were enrolled in this study, and 669 adults were evaluable, following exclusion of 48 subjects due to incomplete patient information. Of these 669 evaluable patients who were admitted to ICU, 257 patients (38%) developed AKI, defined by the KDIGO system. Given the limitation with using the first serum creatinine on ICU admission, we used an AKI recovery definition defined as a 0.33 decrease from baseline serum creatinine (reverse of KDIGO criteria) to capture those patients with AKI on presentation who recovered (Fig. 1;

online suppl. Table 2a, b; for all online suppl. material, see www.karger.com/doi/10.1159/000500231) [20, 21]. Serial urine biomarkers were analyzed daily for 7 days after enrolment and included NGAL, L-FABP, KIM-1, πGST, αGST, and cystatin C. Clinical covariates determined on admission to ICU included the following: APACHE II score, gender, sCr (µmol/L), hourly urine output (mL/h), and BUN. One hundred and twenty eight cases (50%) were classified as having reached KDIGO stage I within 7 days, 23 patients (9%) KDIGO stage II, and 106 partici-

ICU admission screened (n = 2,130)

Included cohort (n = 669/717)

No AKI (n = 412) AKI (n = 257) (38%)

AKI after 48 h but before 7 days (n = 42)

(16%) AKI during first

48 h (n = 215)

(84%) KDIGO stage:

Stage 1: 128 (50%) Stage 2: 23 (9%) Stage 3: 106 (41%) Clinical assessment on

admission to ICU:

• APACHE II • Urinary Output • SCr

• BUN • Gender

Biomarker measurements daily × 7 days:

• NGAL • Albumin • KIM-1 • L-FABP • Cystatin C • πGST • αGST

- Refused consent (n = 15)

- Did not meet enrollment criteria (n = 1,398)

Recovery AKI (n = 177)

434 (65%) Total AKI + recovery cases

Fig. 1. Study characteristics. ICU, intensive care unit; AKI, acute kidney injury; KDIGO, Kidney Disease Improving Global Out- comes; APACHE, Acute Physiologic Assessment and Chronic Health Evaluation; SCr, serum creatinine; BUN, blood urea nitro-

gen; KIM-1, kidney injury molecule-1; L-FABP, liver-type fatty acid-binding protein; GST, glutathione S-transferase; NGAL, neutrophil gelatinase-associated lipocalin.

Color version available online

(4)

pants (41%) developed severe AKI (KDIGO stage III) defined by either receipt of RRT, a 3-fold increase in creatinine, or creatinine rise to >4.0 mg/dL (>353.6 µmol/L).

Patient Characteristics on ICU Admission and Clinical Outcomes

The demographic and clinical characteristics and outcomes of 669 adult patients are reported in Table 1. Pa- tients who developed AKI had a higher BUN and serum creatinine and a lower estimated GFR and urinary output on admission to ICU (Table 1). No differences were not- ed between patients with and without AKI with respect to race and gender (Table 1). Patients who developed any stage of AKI were more likely to have a higher APACHE II score, have preexisting chronic kidney disease (exclud- ing chronic kidney disease [CKD] stage V), and have a

higher proportion of cases with hospital stays exceeding 30 days (Table 1). Eighty-eight participants (34%) died in the AKI cohort, compared to 64 (16%) in the non-AKI cohort. Ninety nine of 669 participants (15%) required RRT within 7 days of admission to ICU; 16% within 30 days. Totally, 56% of the AKI group developed the composite endpoint of RRT and/or death, compared to 16%

of non-AKI participants.

Diagnostic Testing of AKI by Urinary Biomarkers Diagnostic performance of the clinical model (APACHE II score, gender, sCr, hourly urine output [mL/min], and BUN) and each urine biomarker (NGAL, albumin, L- FABP, KIM-1, πGST, αGST, and cystatin C) measured on ICU admission were used for the detection of any stage of AKI within 48 h and 7 days of ICU admission as shown in

Table 1. Patient demographics, clinical characteristics, and outcomes

Risk factor AKI,

257 (38), n (%) Non-AKI,

412 (62), n (%) All,

669, n (%) p value Demographics

Age >75 years 60 (23.3) 76 (18.4) 136 (20.3) 0.13

Gender, female 90 (35) 147 (35.7) 237 (35.4) 0.86

Estimated GFR, mL/min, median (Q1–Q3)* 51 (28–79) 80 (57–97) 69 (45–93) <0.001

Serum creatinine, μmol/L (Q1–Q3)* 116 (83–185) 82 (66–107.3) 93 (71–128) <0.001

BUN, mmol/L, median (Q1–Q3)* 8 (6.0–14) 6 (5–9) 7 (5–10) <0.001

Urine output, mL/min, median (Q1–Q3)* 53 (29–94) 89 (58–127) 75 (48–118) <0.001

APACHE II, median (Q1–Q3)* 24 (18–30) 19 (14–26) 21 (15–27) <0.001

Comorbid illness

IDDM 16 (6.2) 18 (4.4) 34 (5.1) 0.31

NIDDM 33 (12.8) 51 (12.6) 84 (12.7) 0.92

Hypertension 97 (37.7) 147 (36.2) 244 (36.8) 0.69

Sepsis* 51 (20) 54 (13) 105 (16) <0.05

Immunocompromised* 16 (6.2) 12 (2.9) 28 (4.2) <0.05

CKD (not ESKD)* 32 (12.5) 23 (5.7) 55 (8.3) <0.05

Cirrhosis 16 (6.2) 17 (4.1) 33 (4.9) 0.22

Chronic heart failure 26 (10.1) 30 (7.3) 56 (8.4) 0.20

Malignancy 40 (15.6) 70 (17.2) 110 (16.6) 0.57

Nephrotoxin 31 (12.1) 50 (12.1) 81 (12.1) 0.98

Chronic obstructive airway disease 38 (14.8) 52 (12.8) 90 (13.6) 0.47

Peripheral vascular disease 16 (6.2) 14 (3.4) 30 (4.5) 0.09

Outcomes

Number of days in ICU, median (Q1–Q3)* 7 (3–15) 3 (2–7) 4 (2–9) <0.001

≥30 days in hospital* 85 (33) 100 (24) 185 (28) <0.05

Mortality* 88 (34) 64 (16) 152 (23) <0.001

Requirement for dialysis (7 days)* 99 (39) 0 (0) 99 (15) <0.001

Requirement for dialysis (30 days)* 104 (40) 4 (1) 108 (16) <0.001

Death and/or dialysis* 143 (56) 65 (16) 208 (31) <0.001

* p < 0.001–0.05.

AKI, acute kidney injury; GFR, glomerular filtration rate; BUN, blood urea nitrogen; APACHE, Acute Physiologic Assessment and Chronic Health Evaluation; CKD, chronic kidney disease; ICU, intensive care unit.

(5)

the appendix (online suppl. Table 1a, b). Based upon ROC analysis, no single biomarker was outstanding for diagnos- ing AKI. The AUC for the detection of any stage of AKI at 48 h for urine NGAL was 0.70, albumin was 0.68, L-FABP was 0.69, KIM-1 was 0.65, πGST was 0.59, αGST was 0.53, and Cystatin C was 0.63 at 48 h. For the detection of any stage of AKI during first 7 days of ICU admission, the AUC for urine NGAL was 0.68, L-FABP was 0.64, KIM-1 was 0.57, πGST was 0.52, αGST was 0.64, and Cystatin-C was 0.70. Sensitivity and specificity to detect AKI within 48 h and 7 days for each of the individual urine biomarkers are listed in the appendix (online suppl. Table 1).

Risk Prediction for Severe Stage III AKI

The diagnostic performance of all urinary biomarkers or combination urinary NGAL and urinary albumin (measured on ICU admission) together with clinical model were used for the detection of severe AKI (stage 3) within 7 days of ICU admission and are illustrated in Figure 2.

The AUC was modestly improved when urinary NGAL and albumin were added to the clinical model for the prediction of severe stage 3 AKI (AUC improvement from

0.87 to 0.9; p = 0.369; Fig. 2a). The AUC-ROC was also marginally higher for the combination of all urinary biomarkers with the clinical model to predict severe stage 3 AKI (AUC for all biomarkers with clinical model was 0.89, versus 0.85 for the clinical model alone; p = 0.4; Fig. 2b).

Risk Prediction at 48 h and 7 Days Post ICU Admission

The performance of all urinary biomarkers, or the combination of urinary NGAL and urinary albumin, together with the clinical model (assessed on admission to ICU) in predicting any stage of AKI at 48 h and 7 days’

post ICU admission is illustrated in Fig. 3. AUC-ROC for the diagnosis of AKI at 48 h and 7 days using the combination of urinary NGAL/albumin with clinical model were 0.78 and 0.76, respectively, versus clinical model alone with an AUC of 0.76 (p = 0.464) and 0.74 (p = 0.463), respectively (Fig. 3a, b). AUC-ROC for the diagnosis of AKI at 48 h and 7 days using a combination of all urinary biomarkers and clinical model were 0.80 and 0.78, respectively, versus clinical model alone with an AUC of 0.78 (p = 0.555) and 0.75 (p = 0.463), respectively (Fig. 3c, d).

Predicting stage III AKI, within 7 days of ICU admission

Model (n = 668) AUC

Clinical model only 0.87

Clinical model + uNGAL + uALB 0.9

DeLong's test D = 0.899, df = 1,309.3

p = 0.369

Model (n = 355) AUC

Clinical model + all urinary biomarkers 0.89

DeLong's test D = 0.841, df = 679.7

p = 0.4 uNGAL + uALB

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

Model

Clinical model + uNGAL + uALB Clinical model

Model

Clinical model + all urinary biomarkers Clinical model

All urinary biomarkers

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

a b

Fig. 2.a, ^b Predicting stage III AKI (within 7 days of ICU admission). AKI, acute kidney injury; ICU, intensive care unit; AUC, area under the curve; uNGAL, urinary neutrophil gelatinase-associated lipocalin; uALB, urinary albumin.

(6)

Addition to the clinical model of urinary NGAL/albumin combination did have an improvement in the prediction of AKI at 48 h and 7 days post ICU admission; however, the improvement in the AUC was not any more than for the complete panel of urinary biomarkers versus the clinical model alone.

Risk Prediction at 48 h and 7 Days Using AKI Recovery Criteria

While 38% of 669 subjects developed AKI by the standard definition by KDIGO, we identified an additional 177 cases of AKI recovery over the 7 days from ICU admission. The diagnosis of AKI using recovery diagnosis

Admission biomarkers predicting all stages of AKI, within 48 h of ICU admission

Admission biomarkers predicting all stages of AKI, within 7 days of ICU admission

Model (n = 668) AUC

DeLong's test D = 0.732, df = 1,331.8

p = 0.464

Model (n = 355) AUC

DeLong's test D = 0.59, df = 706.68

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

Model

a

Model

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

c

Model (n = 668) AUC

DeLong's test D = 0.735, df = 1,331

p = 0.463

Model (n = 355) AUC

DeLong's test D = 0.734, df = 706.05

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

Model

b

Model

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

d

Fig. 3.a, ^b Admission uNGAL and uALB predicting all stages of AKI with 48 h and 7 days of ICU admission, respectively. ^c, ^d Admission biomarkers predicting all stages of AKI with 48 h and 7 days of ICU

admission, respectively. AKI, acute kidney injury; ICU, intensive care unit; uNGAL, urinary neutrophil gelatinase-associated lipocalin; AUC, area under the curve; uALB, urinary albumin.

(7)

increased the number of cases from 215 (32%) to 310 (46%) within 48 h of ICU admission and from 257 (38%) to 339 (50%) within 7 days of ICU admission. The performance of all urinary biomarkers, or the combination of urinary NGAL and urinary albumin, together with the clinical model (assessed on admission to ICU) in predicting any stage of AKI including all cases of renal recovery at 48 h and 7 days’ post ICU admission is illustrated in online supplementary Table 2a and b, respectively.

Risk Prediction for in-Hospital Mortality or RRT at 30 Days

The AUC-ROCs for death or need for RRT at 30 days using combination of urinary NGAL/albumin and clinical model or all biomarkers and clinical covariates is illustrated in Figure 4a and b, respectively. At 30 days post ICU admission, the AUC for in-hospital mortality or initiation of RRT using a combination of urinary NGAL/

albumin with clinical covariates was 0.83, improved from clinical covariates alone with an AUC of 0.79 (p = 0.139), whereas the AUC for all urinary biomarkers with clinical model was lower at 0.77, but also marginally increased over the clinical model alone, AUC of 0.74 (p = 0.406).

Discussion

We demonstrate that creating a clinical model incorporating severity of illness, patient demographics and chronic illness with currently available clinical parameters of renal function (serum creatinine, hourly urine output, and BUN) was moderately predictive of development of all stages of AKI diagnosis (within 1 week) and associated clinical outcomes (RRT initiation and in-hospital mortality at 30 days). The combination use of urinary NGAL/

Albumin with the clinical model within 48 h and up to 7 days of ICU admission modestly enhanced model performance (Fig. 3a, b) and performed best for severe stages of AKI (stage III; Fig. 2a; AUC improvement from 0.87 to 0.90, p = 0.369). This signal was less robust for the diagnosis of all stages of AKI when compared to severe stage III AKI only. The value of combination NGAL and ALB was most useful for stage 3 AKI, which is not surprising, as more severe kidney damage is associated with lower likeli- hood of reversible AKI. The use of a 7 biomarker panel together with the clinical model also showed modest improved predictive value for all stages of AKI diagnosis (at 48 h and 7 days; Fig. 3c, d) and also performed best for

Biomarker prediction of death or RRT, 30 days

Model (n = 668) AUC

DeLong's test D = 1.48, df = 1,319.4

p = 0.139

Model (n = 355) AUC

DeLong's test D = 0.831, df = 703.95

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

Model

1 – Specificity 0

0.25 0.50 0.75 1.00

Sensitivity

0 0.25 0.50 0.75 1.00

a b

Fig. 4.a, ^b Biomarker prediction of death or RRT (30 days). RRT, renal replacement therapy; uNGAL, urinary neutrophil gelatinase-associated lipocalin; AUC, area under the curve; uALB, urinary albumin.

(8)

severe stage III AKI, however not at a statistically significant level (Fig. 2b; AUC improvement from 0.85 to 0.89, p = 0.4). Equally adding both biomarkers or 7 biomarker panel with the clinical model showed a modest improvement in the prediction of in-hospital mortality or RRT initiation at 30 days post ICU admission (AUC improvement from 0.79 to 0.83, p = 0.139 and 0.74 to 0.77, p = 0.406, respectively; Fig. 4a, b).

Interestingly, the use of AKI recovery diagnostic criteria did result in some minor improvement in biomarker performance after the addition of this alternative catego- rization of the outcome; however, the differences in AUCs between the clinical model and the biomarker model re- mained marginal. Thus, accounting for the addition of recovery cases in the AKI classification did not change the conclusions from this study (online suppl. Table 2a, b).

Despite the modest improved risk prediction for AKI with the use of 2 biomarker combinations or panel of 7 biomarkers with the clinical model (raising the AUC to 0.9; Fig. 2), it must be emphasized the clinical model alone performed better than any single urine AKI biomarker, or their combination (online suppl. Table 1). The lower individual performance by the panel of biomarkers in this study is consistent with prior multicenter reports and may be due to many factors [5, 9–15, 22–26]. The incon- sistent performance of biomarkers across studies in critically ill adult patients can be explained by several factors, such as indiscriminate use of various biomarkers without knowledge of patient clinical status [25, 27], heterogeneous patient population with unknown timing and eti- ology of AKI, lack of gold standard marker [28, 29], presence of comorbid illness, in particular, CKD [22, 30, 31], the presence of volume overload and muscle wasting in critically ill patients and finally lack of clinical context and clinical recommendation on biomarker use [29, 32]. The use of the clinical model for the future risk prediction of severe AKI, need of RRT and/or death noticeably increas- es the AUC without the use of any biomarkers despite the heterogeneous nature of our ICU population with multiple comorbid illnesses and presence of CKD. This is not surprising; the use of clinical models to enrich the study population into low- and high-risk categories is a well- proven approach [33]. We used creatinine, BUN, and urinary output as part of our clinical model for several rea- sons, first, to identify high-risk patients likely to require RRT based upon their degree of kidney dysfunction at time of diagnosis, and secondly, to evaluate the use of novel biomarkers of kidney damage in combination with the kidney function biomarkers that are currently used for decision-making in clinical practice.

The TRIBE study [15] used a clinical prediction model (defined as age [per year], gender, Caucasian race, non- elective surgery, CPB time ≥120 min, RACHS ≥3, estimated glomerular filtration rate percentile, and site) with postoperative levels of urine interleukin-18 and NGAL for prediction of AKI. Urine interleukin-18 and plasma NGAL improved the ROC area from 0.69 to 0.76 and 0.75, respectively. In the SAPPHIRE and TOPAZ studies by Gunnerson et al. [34], a multivariate clinical model was created to assist in the identification of AKI within 12 h of TIMP2 IGFBP7 assessment. With the addition of TIMP2. IGFBP7, the model’s ability to predict moderate- to-severe AKI increased AUC-ROC from 0.77 to 0.88 (p = 0.01). The ROC areas obtained in our current study are similar to those obtained in the surgical subanalysis of SAPPHIRE and TOPAZ studies for prediction of severe and clinically relevant AKI (Fig. 2). In another study by Siew et al. [11], the addition of a clinical risk model including APACHE II to urinary NGAL was limited in predicting AKI in critically ill patients.

This study has shown that biomarkers in combination with the clinical model could predict severe AKI, RRT, or death with high AUC (0.9) in a heterogeneous ICU population, and this level of biomarker performance is superior to level of renal risk published by many other AKI biomarker studies in the heterogeneous ICU population [12, 15, 22, 26]. However, biomarker performance is limited, since no one biomarker or biomarker combinations were superior to clinical model alone, emphasizing the point that more should be done to focus on risk stratification model building in AKI. Adding biomarkers to AKI clinical risk models (e.g., Renal Angina Index AKI risk model) has been shown to further enhance biomarker utility [35]. It is not surprising that a single biomarker or panel of biomarkers alone would achieve superior AKI diagnostic accuracy in the absence of clinical model as AKI is not a single disease entity, but a multifactorial clinical syndrome that can be the result of an array of renal insults. The future challenges of novel biomarkers in AKI research are unclear but may involve their use for the risk stratification of early AKI; to predict severe AKI and RRT;

and to guide patient triage, avoidance of nephrotoxins, and implementation of KDIGO/ESICM guidelines on the prevention or management of AKI [17]. There are several strengths to our observational study. We utilized a prospective cohort design and measured urinary biomarkers consecutively over 7 days from the time of ICU admission. We utilized a precise protocol to collect urinary specimens, followed by blinded measurements. We enrolled a large heterogeneous cohort of ICU subjects

(9)

with a high incidence of AKI (257 patients, 38%) with a high incidence of severe AKI (106 of the 669 patients developed stage 3 AKI during the 7-day study period) and a high number of outcomes such as death and dialysis. Pa- tients were enrolled from 2 large academic centers allow- ing for validation of the biomarkers in >1 clinical setting over a 3-year setting. Finally, all biomarker data were normalized to urinary creatinine, which did not change the conclusion of this paper (online suppl. Table 2c–f).

This study does have limitations. First, the definition of AKI used in this study is based on the serum creatinine criteria, which raises the challenge of using an imperfect variable to analyze the performance of novel biomarkers, but we also assessed patient-centered clinical outcomes such as death and RRT requirement. Second, our cohort had a predominance of Caucasian participants, and the effect of other race on these biomarkers could not be stud- ied effectively. Finally, we used NGAL in combination with urinary albumin as our biomarker combination of choice for testing in all subjects as both tests were available in Europe on clinical laboratory platforms at the time of study initiation. Several other biomarkers were tested in smaller subsets in exploratory analyses. Other AKI biomarkers (including Nephrocheck) were unavailable when this study was conducted and may or may not show superior AKI risk prediction compared to NGAL/albumin, but was not tested in this cohort.

Conclusion

Using a large, 2-center, heterogeneous critically ill cohort, we have demonstrated that on ICU admission, the use of a simple clinical model (incorporating severity of illness and standard clinical biomarkers of renal function) was moderately predictive of development of severe AKI (within 1 week) and associated clinical outcomes of RRT and death (within 30 days). The addition of AKI biomarkers to this simple clinical model modestly improved prediction of AKI and associated adverse clinical outcomes, but not at a statistically or clinically significant level, when compared to the clinical model alone.

Acknowledgments

The GST urine biomarker assays were donated by Argutus Di- agnostics/EKF Diagnostics, Ireland. Urine NGAL assays were donated by Abbott Diagnostics. We are indebted to our ICU staff nurses for their assistance and to our patients and their families for their participation.

Ethics Approval

The study was approved by the research Ethics Committees of the Mater Misericordiae University Hospital and St. Vincent’s University Hospital, Dublin, Ireland.

Disclosure Statement

P.T.M. is a scientific advisor to: FAST Biomedical; Sphingotec;

and AM-Pharma. All other authors declare that they have no com- peting interests.

Consent to Participate

Due to the incapacitating nature of critical illness among our participants deferred consent to collect urinary samples was approved, assent was later obtained from a legal authorized representative, and consent from the patient if they recovered prior to analysis.

Consent for Publication Not applicable.

Availability of Data and Material

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Funding Sources

The research reported in this article was supported by Health Research Board (Ireland) (HRA_POR 2011 128), and Health Re- search Board (Ireland) CRF/2007/1 (Dublin Centre for Clinical Research Consortium Network, Ireland). Professor Nichol is supported by a Health Research Board clinical trial network award (HRB CTN-2014-012).

Author Contributions

P.T.M.: designed the study and obtained funding for the study.

B.A.M., T.M., E.M., N.M., and C.H.: enrolled, consented, collected specimens, and patient information and maintained ethical stan- dards for this study. E.J.C.: preformed biomarker measurements.

M.G.: analyzed the results and provided critical input into statistical analysis. B.A.M.: drafted the paper. B.M. and A.N.: oversaw patient eligibility and enrollment in the ICUs. All authors made substantial contributions, interpreted the results, and read and approved the final manuscript.

(10)

References

1 Uchino S, Kellum JA, Bellomo R, Doig GS, Morimatsu H, Morgera S, et al.; Beginning and Ending Supportive Therapy for the Kidney (BEST Kidney) Investigators. Acute renal failure in critically ill patients: a multinational, multicenter study. JAMA. 2005 Aug;294(7):813–8.

2 Ricci Z, Cruz D, Ronco C. The RIFLE criteria and mortality in acute kidney injury: A sys- tematic review. Kidney Int. 2008 Mar;73(5):

538–46.

3 Waikar SS, Betensky RA, Bonventre JV. Cre- atinine as the gold standard for kidney injury biomarker studies? Nephrol Dial Transplant.

2009 Nov;24(11):3263–5.

4 Devarajan P, Murray P. Biomarkers in acute kidney injury: are we ready for prime time?

Nephron Clin Pract. 2014;127(1–4):176–9.

5 Koyner JL, Vaidya VS, Bennett MR, Ma Q, Worcester E, Akhter SA, et al. Urinary biomarkers in the clinical prognosis and early detection of acute kidney injury. Clin J Am Soc Nephrol. 2010 Dec;5(12):2154–65.

6 Wagener G, Jan M, Kim M, Mori K, Barasch JM, Sladen RN, et al. Association between in- creases in urinary neutrophil gelatinase-associated lipocalin and acute renal dysfunction after adult cardiac surgery. Anesthesiology.

2006 Sep;105(3):485–91.

7 Xin C, Yulong X, Yu C, Changchun C, Feng Z, Xinwei M. Urine neutrophil gelatinase-associated lipocalin and interleukin-18 predict acute kidney injury after cardiac surgery. Ren Fail. 2008;30(9):904–13.

8 Che M, Xie B, Xue S, Dai H, Qian J, Ni Z, et al. Clinical usefulness of novel biomarkers for the detection of acute kidney injury following elective cardiac surgery. Nephron Clin Pract.

2010;115(1):c66–72.

9 de Geus HR, Bakker J, Lesaffre EM, le Noble JL. Neutrophil gelatinase-associated lipocalin at ICU admission predicts for acute kidney injury in adult patients. Am J Respir Crit Care Med. 2011 Apr;183(7):907–14.

10 Makris K, Markou N, Evodia E, Dimopoulou E, Drakopoulos I, Ntetsika K, et al. Urinary neutrophil gelatinase-associated lipocalin (NGAL) as an early marker of acute kidney injury in critically ill multiple trauma patients.

Clin Chem Lab Med. 2009;47(1):79–82.

11 Siew ED, Ware LB, Gebretsadik T, Shintani A, Moons KG, Wickersham N, et al. Urine neutrophil gelatinase-associated lipocalin moderately predicts acute kidney injury in critically ill adults. J Am Soc Nephrol. 2009 Aug;

20(8):1823–32.

12 Mårtensson J, Bell M, Oldner A, Xu S, Venge P, Martling CR. Neutrophil gelatinase-associated lipocalin in adult septic patients with and without acute kidney injury. Intensive Care Med. 2010 Aug;36(8):1333–40.

13 Metzger J, Kirsch T, Schiffer E, Ulger P, Men- tes E, Brand K, et al. Urinary excretion of twenty peptides forms an early and accurate

diagnostic pattern of acute kidney injury.

Kidney Int. 2010 Dec;78(12):1252–62.

14 de Geus HR, Fortrie G, Betjes MG, van Schaik RH, Groeneveld AB. Time of injury affects urinary biomarker predictive values for acute kidney injury in critically ill, non- septic patients. BMC Nephrol. 2013 Dec;

14(1):273.

15 Parikh CR, Coca SG, Thiessen-Philbrook H, Shlipak MG, Koyner JL, Wang Z, et al.;

TRIBE-AKI Consortium. Postoperative biomarkers predict acute kidney injury and poor outcomes after adult cardiac surgery. J Am Soc Nephrol. 2011 Sep;22(9):1748–57.

16 Di Somma S, Zampini G, Vetrone F, Soto- Ruiz KM, Magrini L, Cardelli P, et al. Opinion paper on utility of point-of-care biomarkers in the emergency department pathways decision making. Clin Chem Lab Med. 2014 Oct;

52(10):1401–7.

17 Group KA. KDIGO clinical practice guideline for acute kidney injury. Kidney Int Suppl.

2012;2:1–138.

18 Knaus WA, Draper EA, Wagner DP, Zim- merman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985 Oct;13(10):818–29.

19 Ralib AM, Pickering JW, Shaw GM, Devara- jan P, Edelstein CL, Bonventre JV, et al. Test characteristics of urinary biomarkers depend on quantitation method in acute kidney injury. J Am Soc Nephrol. 2012 Feb;23(2):322–

20 Chawla LS, Bellomo R, Bihorac A, Goldstein 33.

SL, Siew ED, Bagshaw SM, et al.; Acute Dis- ease Quality Initiative Workgroup 16. Acute kidney disease and renal recovery: consensus report of the Acute Disease Quality Initiative (ADQI) 16 Workgroup. Nat Rev Nephrol.

2017 Apr;13(4):241–57.

21 Long TE, Helgadottir S, Helgason D, Sigurds- son GH, Gudbjartsson T, Palsson R, et al.

Postoperative Acute Kidney Injury: Focus on Renal Recovery Definitions, Kidney Disease Progression and Survival. Am J Nephrol.

2019;49(3):175–85.

22 Endre ZH, Pickering JW, Walker RJ, Devara- jan P, Edelstein CL, Bonventre JV, et al. Im- proved performance of urinary biomarkers of acute kidney injury in the critically ill by stratification for injury duration and baseline renal function. Kidney Int. 2011 May;79(10):

1119–30.

23 Liangos O, Tighiouart H, Perianayagam MC, Kolyada A, Han WK, Wald R, et al. Compara- tive analysis of urinary biomarkers for early detection of acute kidney injury following cardiopulmonary bypass. Biomarkers. 2009 Sep;14(6):423–31.

24 Murray PT, Mehta RL, Shaw A, Ronco C, En- dre Z, Kellum JA, et al.; ADQI 10 workgroup.

Potential use of biomarkers in acute kidney injury: report and summary of recommenda-

tions from the 10th Acute Dialysis Quality Initiative consensus conference. Kidney Int.

2014 Mar;85(3):513–21.

25 Ho J, Tangri N, Komenda P, Kaushal A, Sood M, Brar R, et al. Urinary, Plasma, and Serum Biomarkers’ Utility for Predicting Acute Kid- ney Injury Associated With Cardiac Surgery in Adults: A Meta-analysis. Am J Kidney Dis.

2015 Dec;66(6):993–1005.

26 Mishra J, Dent C, Tarabishi R, Mitsnefes MM, Ma Q, Kelly C, et al. Neutrophil gelatinase- associated lipocalin (NGAL) as a biomarker for acute renal injury after cardiac surgery.

Lancet. 2005 Apr;365(9466):1231–8.

27 Ralib AM, Pickering JW, Shaw GM, Than MP, George PM, Endre ZH. The clinical utility win- dow for acute kidney injury biomarkers in the critically ill. Crit Care. 2014 Nov;18(6):601.

28 Molitoris BA. Measuring glomerular filtration rate in the intensive care unit: no substi- tutes please. Crit Care. 2013 Sep;17(5):181.

29 Liu KD, Thompson BT, Ancukiewicz M, Ste- ingrub JS, Douglas IS, Matthay MA, et al.;

National Institutes of Health National Heart, Lung, and Blood Institute Acute Respiratory Distress Syndrome Network. Acute kidney injury in patients with acute lung injury: impact of fluid accumulation on classification of acute kidney injury and associated outcomes. Crit Care Med. 2011 Dec;39(12):

2665–71.

30 Mitsnefes MM, Kathman TS, Mishra J, Kartal J, Khoury PR, Nickolas TL, et al. Serum neutrophil gelatinase-associated lipocalin as a marker of renal function in children with chronic kidney disease. Pediatr Nephrol. 2007 Jan;22(1):101–8.

31 McIlroy DR, Wagener G, Lee HT. Neutrophil gelatinase-associated lipocalin and acute kidney injury after cardiac surgery: the effect of baseline renal function on diagnostic performance. Clin J Am Soc Nephrol. 2010 Feb;5(2):

211–9.

32 Puthucheary ZA, Rawal J, McPhail M, Con- nolly B, Ratnayake G, Chan P, et al. Acute skeletal muscle wasting in critical illness.

JAMA. 2013 Oct;310(15):1591–600.

33 Yang Y, Chen L, Yam Y, Achenbach S, Al- Mallah M, Berman DS, et al. A clinical model to identify patients with high-risk coronary artery disease. JACC Cardiovasc Imaging.

2015 Apr;8(4):427–34.

34 Gunnerson KJ, Shaw AD, Chawla LS, Bihorac A, Al-Khafaji A, Kashani K, et al.; Sapphire Topaz investigators. TIMP2•IGFBP7 biomarker panel accurately predicts acute kidney injury in high-risk surgical patients. J Trauma Acute Care Surg. 2016 Feb;80(2):243–9.

35 Basu RK, Wang Y, Wong HR, Chawla LS, Wheeler DS, Goldstein SL. Incorporation of biomarkers with the renal angina index for prediction of severe AKI in critically ill children.

Clin J Am Soc Nephrol. 2014 Apr;9(4):654–62.