Acute kidney injury (AKI) is a common clinical event with severe consequences. In the United States, >1 million hospitalized patients yearly receive a diagnosis of AKI, and its incidence is increasing (1- 4). AKI is associated with a 25% to 80% risk of in-hospital death (4- 6) and has been implicated in the pathogenesis of chronic kidney disease (CKD) (7).

The risk, injury, failure, loss of function, end-stage renal disease (RIFLE) and the Acute Kidney Injury Network definitions of AKI are based on changes in both the levels of serum creatinine (sCr) and urinary output (8- 9). These definitions may be problematic, however, when applied to patients in the emergency department (ED) because the baseline sCr may be unknown, and placement of a urinary catheter may not be indicated. Furthermore, subclinical AKI may fail to display diagnostic changes in sCr level despite evidence of nephron damage (10). Additionally, the sCr level may not adequately reflect the severity of kidney injury because sCr kinetics are altered by age, sex, muscle mass, nutritional status, and medications. Also, the RIFLE and Acute Kidney Injury Network definitions do not consider the presence of structural nephron damage (intrinsic AKI [iAKI]), despite its documented association with poor clinical outcomes. In fact, sCr levels may meet these criteria in the absence of nephron damage as a result of altered hemodynamics (pre-renal acute kidney injury [pAKI]) (5,11- 16).

Some of these shortcomings may be addressed by novel kidney injury biomarkers that are released into the urine after cellular injury. Currently, the most promising candidates include urinary neutrophil gelatinase–associated lipocalin (uNGAL), urinary kidney injury molecule (uKIM)-1, urinary interleukin (uIL)-18, urinary liver fatty acid binding protein (uL-FABP), and urinary cystatin C (uCysC) (17- 18). Their diagnostic and prognostic performance has been evaluated but only in single-center studies, usually involving subtypes of AKI in selected clinical settings. As a result, it is unclear: 1) how the biomarkers behave in large, heterogeneous populations; 2) which biomarker successfully discriminates iAKI from either pAKI or CKD, at the time of patient presentation; and 3) whether combinations of biomarkers have greater diagnostic accuracy than single markers. Urinary biomarkers may provide a strategy to improve the diagnosis of iAKI and predict its severity and clinical outcomes beyond currently available tests (16). Along similar lines, a recent meta-analysis suggested that uNGAL identifies a substantial population with subclinical AKI who had escaped detection by sCr measurements but who were at increased risk of a poor clinical outcome (19).

To address the comparative utility of urinary biomarkers to diagnose iAKI and to predict the hospital course, we conducted this multicenter, prospective, observational study in a heterogeneous population of ED patients.

The study was conducted at 3 centers: 1) Allen Hospital of New York-Presbyterian Hospital, an inner-city community hospital including a designated chest pain and stroke center; 2) Staten Island University Hospital, a community hospital including a designated trauma center, chest pain center, and stroke center; and 3) Helios Clinics, Berlin-Buch, a community hospital including a designated trauma center, chest pain center, and stroke center. Recruiting study personnel, during their work hours, enrolled all available patients older than 18 years of age irrespective of their condition who were in the process of admission to the hospital from the ED (September 2008 to March 2009). Patients who had <24 h of follow-up or were on long-term renal replacement therapy were excluded. One urine sample was collected in the ED and medical records were accessed. Treating physicians were neither aware of urinary biomarker levels, nor did our study hospitals routinely determine serum cystatin C levels. The study was approved by each site's institutional review board and performed in compliance with the Health Insurance Portability and Accountability Act at all study sites. Informed consent was obtained from all participants.

This observational study was intended to derive diagnostic test characteristics of urinary biomarkers to predict the development of inpatient iAKI. Secondary analyses included severity of AKI, duration of AKI, and a composite clinical outcome of dialysis initiation or mortality. Baseline sCr was determined by review of the previous 12 months of records, or, if unavailable, baseline sCr was assumed from the lowest recorded sCr level during the hospital course. Estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease formula (20).

Diagnostic categorization was performed by adjudicators who were blinded to urinary biomarker levels. A priori–defined algorithms assigned patients to 1 of 4 renal diagnoses (normal kidney function, stable CKD, pAKI, iAKI). Patients were labeled unclassified when ambiguity occurred or when disagreement among 2 or more investigators could not be resolved by re-evaluation of clinical data.

Definitions of diagnostic categories were applied as follows. In normal kidney function (n = 730), patients met the following criteria: 1) baseline eGFR ≥60 ml/min/1.73 m²; 2) failure to meet minimal RIFLE criteria for AKI (8); 3) absence of fluctuations in sCr level during the first 3 days of hospitalization (≥0.3 mg/dl when baseline sCr was ≥1.0 mg/dl, or ≥0.2 mg/dl when the baseline sCr was ≤1.0 mg/dl); and 4) absence of recent exposures to stimuli that typically cause iAKI (e.g., shock requiring vasopressors, positive blood cultures, systemic inflammatory response syndrome or sepsis, nephrolithiasis, recent chemotherapy, nephrotoxins, rhabdomyolysis, glomerulonephritis, interstitial nephritis, vasculitis, pre-eclampsia, multiple myeloma, thrombotic microangiopathy). Patients who met criteria 1 and 2, but did not meet criterion 3 or 4 were labeled unclassified.

In stable CKD (n = 154), patients had a baseline eGFR <60 ml/min/1.73 m² and met criteria 2 to 4 as defined in normal kidney function. Patients who met criteria 1 and 2, but did not meet criterion 3 or 4 were placed in the unclassified category.

In pAKI (n = 254), patients met the following criteria: 1) minimal RIFLE sCr criteria for AKI (1.5-fold increase in sCr level or 25% decrease in eGFR from baseline; urine output criteria were not considered due to difficulty in obtaining accurate measurements in the ED); 2) normalized values below the RIFLE-Risk threshold within three days; 3) historical and/or clinical data suggesting decreased kidney perfusion, but no exposure to stimuli, which induce iAKI (see previously); and 4) response to measures to restore renal perfusion, such as fluid application or diuretic withdrawal. Patients who met criteria 1 and 2, but did not meet criterion 3 or 4 were labeled unclassified.

In iAKI (n = 96), patients met the following criteria: 1) minimal RIFLE sCr criteria for AKI; 2) sCr level failed to normalize below the RIFLE R threshold by 3 days after admission; and 3) patients had evidence of exposure to stimuli, which induce iAKI (see previously). Patients who met criteria 1 and 2, but did not meet criterion 3 or who were exposed to factors that may have changed the creatinine course subsequent to inclusion (e.g., contrast administration in hospital) were labeled unclassified.

Urine samples were centrifuged (12,000 rpm for 10 min) and stored at −80°C within 12 h after patient enrollment. uNGAL, uIL-18, uKIM-1, and uCysC were measured by the ARCHITECT platform (Abbott Laboratories, Abbott Park, Illinois) (21). These assays used a chemiluminescent microparticle immunoassay using a noncompetitive, 2-antianalyte antibody sandwich. The assays include a microparticle reagent prepared by covalently attaching an antianalyte antibody to paramagnetic particles and a conjugate reagent prepared by labeling a second antianalyte antibody with acridinium. The calibrators for the uNGAL, uIL-18, and uKIM-1 assays were recombinant proteins, and the calibrators for the uCysC assay were prepared from human urine. The highest calibrator for each assay was 1,500 ng/ml, 1 ng/ml, 10 ng/ml, and 2500 ng/ml for uNGAL, uIL-18, uKIM-1, and CysC, respectively. Specimens were diluted to read within the calibration curve. Coefficients of variation were 3.0% for uNGAL at a 385 ng/ml (21), 2.5% for uKIM-1 at 5.8 ng/ml (Abbott Laboratories), 2.2% for uIL-18 at 0.048 pg/ml (Abbott Laboratories), 1.8% for uCysC at 350 ng/ml (Abbott Laboratories), and similar at other cut points. uL-FABP was measured using a sandwich-type enzyme-linked immunosorbent assay kit (CMIC Co., Ltd., Tokyo, Japan). The coefficient of variation was 6.8% for uL-FABP at 13 ng/ml (22). Monomeric uNGAL (23 to 26 kDa) was measured by immunoblots, which were prepared with nonreducing 4% to 15% gradient polyacrylamide gels (Bio-Rad, Hercules, California) using standards (0.3 to 3 ng) of human recombinant NGAL and NGAL antibody (AntibodyShop, Copenhagen, Denmark). The coefficient of variation was <5% at different cut points (16). sCr was assayed at each hospital by the Jaffe reaction, calibrated traceable to isotope dilution mass spectrometry.

Statistical analyses were performed with SAS version 9.2 (SAS Institute, Cary, North Carolina) and SPSS versions 16.0 to 19.0 (SPSS, Chicago, Illinois). Sample sizes were estimated based on previous data (16) and were calculated to detect within each study center differences in biomarker performance among patients with iAKI and others (for details, see the Online Appendix). Continuous variables were log-transformed and presented as geometric means and SDs, where appropriate. Comparisons between 2 conditions were performed using Student's t test; comparisons of 3 or more conditions were performed using analysis of variance and a post hoc Tukey test. Categorical variables were compared by a chi-square test. The null hypothesis was rejected at p < 0.05. Biomarker diagnostic test characteristics for iAKI were determined by area under the receiver-operating characteristic curve (AUC-ROC) analysis. For clinical outcomes, univariate logistic regression was used to identify independent predictors, including demographic parameters (age, sex, race), comorbidities (congestive heart failure, vascular disease, hypertension, diabetes, human immunodeficiency virus, CKD), admission parameters (heart rate, blood pressure, shock index), kidney function parameters, and study site. Predictors significantly associated with clinical outcomes from univariate models were entered into a multiple logistic regression model, and backward selection techniques were used to determine a baseline risk model. Biomarkers were then sequentially entered into the baseline model, including an adjustment for sCr to determine independent relationships among the biomarkers and the composite clinical outcome. Integrated discrimination improvement (IDI) and net reclassification improvement (NRI) were calculated according to Pencina et al. (23).

We enrolled 1,635 adult patients presenting to 3 EDs. Site-specific race and ethnic composition included Berlin (100% white); Staten Island University Hospital (7% Hispanic, 9% black, 79% white, 5% other), and Allen Hospital of New York-Presbyterian Hospital (52% Hispanic, 21% black, 27% white). A total of 1,234 (75.5%) patients could be assigned to 1 of 4 diagnostic categories (normal kidney function, stable CKD, pAKI, iAKI); 401 patients remained unclassified (Figure 1). Baseline characteristics by diagnostic and outcome groups are presented in (Table 1). The primary etiologies of iAKI were hypotension (34%), urinary obstruction (29%), sepsis (22%), glomerulonephritis or vasculitis (6%), hepatorenal syndrome (2.1%), and rhabdomyolysis (2.1%); causes of iAKI present at a lower prevalence (1%) included contrast nephropathy, acute interstitial nephritis (biopsy proven), scleroderma crisis, and multiple myeloma. Seventy-two patients (4.4%) experienced the composite outcome of dialysis initiation or death. Causes of death (n = 56) included sepsis (n = 22; 39%), metastatic cancer (n = 9; 16%), heart failure (n = 7; 13%), liver failure (n = 5; 9%), stroke or intracranial bleed (n = 5; 9%), respiratory failure (n = 3; 5%), and gastrointestinal bleed (n = 1; 2%). iAKI was strongly associated with adverse in-hospital outcomes, including death, dialysis initiation, and intensive care unit admission (Table 1).

Most patients with iAKI had already reached their peak RIFLE class on admission (n = 62; 64.6%). Accordingly, both the presenting sCr level and its change from baseline were significantly higher in iAKI compared with patients without iAKI ((Table 1), Figure 2). sCr level at presentation and the ratio of presenting-to-baseline sCr level highly discriminated iAKI patients (AUC-ROC: 0.91 [95% confidence interval: 0.87 to 0.94] and 0.89 [95% confidence interval: 0.84 to 0.94], respectively) (Table 2). However, presenting sCr level was not predictive of iAKI when its level was in the middle tertile of its range (0.9 to 1.2 mg/dl), nor could it distinguish those patients whose sCr level continued to increase after admission (>1.5-fold) from the rest of the cohort. Furthermore, only 63.5% of the patients had documented baseline sCr values at the time of presentation to the ED, highlighting the difficulties of interpreting single measurements of sCr in triage.

Urinary biomarker levels differed among patients with iAKI and any other diagnosis (by t test: p < 0.001 for all biomarkers). In addition, subset comparisons showed that urinary biomarker levels were significantly elevated in patients with iAKI compared with each of the other diagnostic categories (pAKI, stable CKD, or normal kidney function by 1-way analysis of variance and post-hoc Tukey test) (Figure 2). AUC-ROC analyses indicated good discriminatory ability for uNGAL (AUC-ROC: 0.81), fair discriminatory ability for uKIM-1 and uL-FABP (AUC-ROC: 0.71 and 0.70, respectively), and poor discriminatory ability for uCysC and uIL-18 (AUC-ROC: 0.65 and 0.64, respectively) ((6), Table 2) to distinguish iAKI from other diagnoses. The AUC-ROC of uNGAL was significantly greater than that of other urinary markers (p < 0.001 each) (Table 2). When the monomeric form of uNGAL was quantified by immunoblot, we found that it markedly correlated with the standardized clinical platform and performed similarly in AUC-ROC analyses ((6) and 6). Also, AUC-ROCs remained similar when biomarker levels were standardized for urinary creatinine concentrations (data not shown). Sensitivity, specificity, predictive values, and likelihood ratios for diagnostic cutoffs at the 60th and 75th percentiles for each biomarker are shown in (Table 2). The suggested cutoff value of uNGAL (104 ng/ml) was similar to a previously proposed cutoff (16) and meanwhile was validated in an independent cohort (24).

We performed sensitivity analyses by selecting subsets of patients based on study site, ethnicity, sex, infection of the urinary tract, and the co-prevalence of CKD or RIFLE-AKI. The AUC-ROCs for the diagnosis of iAKI remained consistent within these subsets (6).

Given that iAKI is associated with longer durations and greater severity of azotemia compared with pAKI (16,25), we performed secondary analyses to assess the relationship between biomarker levels and the duration and severity of AKI. To evaluate the duration of AKI, the entire cohort was categorized into 3 subgroups: no AKI, transient AKI (defined as RIFLE-AKI that resolved by 72 h), and sustained AKI (RIFLE-AKI that persisted for ≥72 h). Only uNGAL and uCysC levels were significantly higher in patients with sustained AKI compared with patients with transient AKI episodes (6). Conversely, uKIM-1, uL-FABP, and uIL-18 were not significantly different in sustained and transient AKI.

We next stratified the entire cohort by peak RIFLE severity class (within 7 days from inclusion) (6). uNGAL levels progressively and significantly increased in parallel with RIFLE severity class. In contrast, although the other biomarkers were higher in RIFLE-AKI than in non-AKI, these markers did not strictly parallel RIFLE class. Accordingly, the AUC-ROC for uNGAL progressively increased for the prediction of RIFLE-Risk, RIFLE-Injury, and RIFLE-Failure, whereas progressive increases were absent or less pronounced for other biomarkers (6). Together, these data indicated that uNGAL was the most powerful indicator of severity and duration of AKI.

Patients with iAKI experienced higher rates of a composite clinical outcome of death or dialysis during hospitalization (Table 1). We used multiple logistic regression to construct a conventional baseline prediction model that was adjusted for sCr level at inclusion (Table 3, model 1). We then added each individual biomarker to the baseline model (Table 3, models 2 to 6) and found that each biomarker added significantly to the predictive ability of the baseline model. In addition, 2 measures of the overall performance of the regression model, R² and AUC-ROC, improved when each biomarker was added to the baseline model. We then used stepwise backward selection techniques to identify combinations of biomarkers that independently contributed to the prediction of the composite outcome. We found that uNGAL and uKIM-1 independently added to a combined prediction model (Table 3, model 7). However, R² and AUC-ROC of the combined model increased only slightly compared with the single biomarker models.

To examine the incremental utility of urinary biomarkers and their combinations in more detail, we calculated the IDI, comparing the biomarker-aided models (models 2 to 5) with the sCr-based baseline model (model 1) (23). This method compares the risk estimates derived from each model in patients with outcomes (events) and in patients without outcomes (nonevents). A biomarker-aided risk estimate will achieve a positive IDI compared with a conventional risk estimate, if patients with events are assigned higher risk estimates and if patients without events are assigned lower risk estimates. The data showed that uNGAL- and uKIM-1–assisted models achieved significant IDIs compared with the baseline model (Table 3). The uNGAL-assisted model also achieved a significant IDI compared with models using uL-FABP, uIL-18, or CysC (models 4, 5, and 6) (p < 0.05 each). A triple model of sCr, uNGAL, and uKIM-1 (model 7) did not achieve a significant IDI compared with the double model with sCr and uNGAL (Table 3). These data indicated that uNGAL and uKIM-1 individually improved risk stratification when combined with sCr level, whereas a combination of uNGAL and uKIM-1 did not further improve the predictive ability.

To estimate the NRI facilitated by uNGAL and uKIM-1, we defined 3 risk categories (<2%, 2% to 15%, and >15%) of experiencing the composite outcome within our ED population, which were based on the expected rates of in-hospital mortality and in-hospital dialysis initiation (16). Next, we assigned each patient to one of these risk classes based on either the biomarker-assisted models (model 2 or 3) or the conventional model (model 1). The net number of patients with a classification improvement after uNGAL-stratified assessment was 120 (7.8%) among patients without events and 13 (18.3%) among patients with events (Table 4). The net number of patients with a classification improvement after uKIM-1–stratified assessment was 204 (13.4%) among patients without events and 7 (10.4%) among patients with events (Table 4). Hence, the introduction of uNGAL and uKIM-1 facilitated NRI of 26.1% and 23.8%, respectively.

To translate these findings into a diagnostic strategy, we risk stratified patients by sCr level and used uNGAL or uKIM-1 to further subdivide these categories. Based on 75th percentile cutoffs, we separated patients into sCr⁺ (sCr ≥1.4 mg/dl) and sCr⁻ (sCr <1.4 mg/dl) at the time of inclusion. We then subdivided sCr⁺ and sCr⁻ patients into biomarker positive (biomarker ≥ cutoff at 75th percentile) or biomarker negative (biomarker < cutoff at 75th percentile). Event rates within sCr⁺or sCr⁻ patients were substantially different depending on whether they were biomarker positive or biomarker negative (Figure 3). In particular, approximately 15% of the population had a low sCr level, but high biomarker levels, placing them at low risk by conventional stratification, but at an increased risk on application of biomarker-aided stratification.

This is the first multicenter study to comprehensively compare the diagnostic and predictive abilities of urinary biomarkers. Our goals were to evaluate their ability to: 1) distinguish iAKI from pAKI, stable CKD, and normal kidney function; and 2) facilitate a prospective risk assessment regarding a requirement for dialysis or the death of the patient during subsequent hospitalization.

Every patient who entered the ED and was subsequently hospitalized for >24 h was included in the study. Although the exclusion of patients hospitalized for <24 h may have introduced some bias in favor of sicker patients compared with the inclusion of all comers, derivation of the test characteristics of biomarkers compared with standard clinical information necessitated a reasonable period in which to accumulate follow-up data.

Our study was limited by the absence of a diagnostic gold standard of iAKI. We addressed this limitation by establishing a standardized adjudication procedure to define iAKI, which was based on sCr dynamics, the etiology of AKI, and the response to therapy considering kidney physiology in addition to AKI pathogenesis (16). Using this strict approach, we could analyze three fourths of the cohort without ambiguity, but because we could not assign a definitive diagnosis to approximately one fourth, we performed adjudication-independent secondary analyses across the entire cohort and found these approaches to be complementary: the utility to diagnose iAKI in a strictly defined cohort paralleled the biomarkers' ability to predict the intensity and duration of AKI in all comers. For instance, uNGAL performed significantly better than the other biomarkers in diagnosing iAKI, and consistently uNGAL displayed a closer association with severity and duration of AKI compared with the other biomarkers. Importantly, uNGAL was progressively more effective in predicting increasing RIFLE classes, a finding consistent with previous studies (26- 27).

It is noteworthy that both the presenting sCr level and its change from baseline highly discriminated iAKI patients from other diagnostic groups. This may in part be related to the fact that sCr level was a major determinant of the diagnostic adjudication procedure itself and that most patients had already achieved their peak RIFLE severity class at presentation to the ED. This also implies that urinary biomarkers will be most useful when sCr dynamics are unknown or when the sCr level is in the middle of its range.

Our study confirmed the known association of the type of AKI with the clinical outcome. Although <4% of patients with normal kidney function, stable CKD, or pAKI experienced the composite outcome of in-hospital mortality or the requirement to initiate in-hospital hemodialysis, 33.5% of the patients with iAKI experienced this outcome. Several smaller studies linked high urinary biomarker levels with these unfavorable clinical events (16,28- 31), and here we confirmed this association. However, our large sample size enabled us to characterize the contribution of each biomarker; uNGAL and uKIM-1 were the most accurate predictors of subsequent clinical events, and each markedly improved risk assessment when combined with the conventional sCr-assisted prediction models, as determined by IDI and NRI. In contrast, the combination of uNGAL and uKIM-1 did not further improve risk classification, nor were we able to find evidence of the superiority of 1 of the 2 markers in head-to-head comparisons.

Besides the issues discussed above, there are additional limitations to this study. Even with the biomarker-aided improvement in risk stratification, we cannot assess its potential implications for clinical management without a prospective randomized trial. Also, most of the biomarkers were assayed using single clinical assays, except for uNGAL, where we were able to determine the molecular identity of the AKI-specific NGAL monomer using immunoblots and correlating these findings with an established clinical platform. However, we were unable to achieve a comparable molecular assessment of the other biomarkers. For each biomarker, the diagnostic and prognostic utility may be different when other clinical platforms are used.

In summary, we characterized the performance of different urinary biomarkers obtained on patient entry to the hospital to diagnose iAKI (as defined by strict criteria) and to determine its severity and clinical sequelae. Our analysis prospectively validated the concept that the addition of urinary biomarkers and their interpretation together with sCr levels identified patients at risk who otherwise would have been missed during triage.

The authors thank Stanley Hazen, MD, PhD, and Cleveland Clinic for their assistance performing ARCHITECT uNGAL and uIL-18 testing. They are grateful for the work of Maria Saccio, RN, and Elysha R. Grauman, PA-C, at Staten Island University Hospital. The authors appreciate the technical and logistical support of Frank Quinn and Frank Grenier at Abbott.

For supplementary Methods and Results sections, tables, and figures, please see the online version of this article.