The clinical utility window for acute kidney injury biomarkers in the critically ill

Reliance on plasma creatinine delays acute kidney injury (AKI) diagnosis [1],[2]. Novel injury biomarkers may enable earlier diagnosis. However, biomarker performance also depends on the interval between insult and time of measurement [3],[4]. In the trial, early intervention in acute renal failure (EARLYARF) [5], triaging to intervention within 6 h of admission to the ICU by the urinary tubular enzymes, alkaline phosphatase (AP) and γ-Glutamyl Transpeptidase (GGT) resulted in many false negatives. This occurred because this combination of biomarkers had a short temporal profile with peak urinary concentrations before 12 h after injury and the window of opportunity for diagnosis had been exceeded in many subjects, despite rapid processing in the ICU. Apart from one study of plasma and urinary cystatin C [6] little is known about the temporal profiles of candidate AKI biomarkers of patients presenting to the emergency department (ED). While studies have examined biomarker diagnostic performance in the ED [7],[8], subsequent performance at later time points has not been assessed.

We evaluated the temporal profile of plasma creatinine and AKI biomarkers in high-risk critically ill patients, who presented soon after a probable hypoperfusion insult to the kidney. The temporal profiles of biomarker performance in AKI diagnosis and mortality prediction were investigated. We hypothesised that earlier measurement (in the ED) would enhance clinical utility compared with later measurement in the ICU.

The EDAKI study was a prospective observational study investigating the utility of plasma and urinary biomarkers from admission to the ED throughout the ICU stay, in patients at high risk of developing AKI. The study was approved by The Upper South A Regional Ethics Committee (URA/09/09/062) and registered under the Australian and New Zealand Clinical Trials Registry (ACTRN12610001012066) and adhered to the Declaration of Helsinki. Patients included were admitted to the ICU at Christchurch Hospital following cardiac arrest, sustained or prolonged hypotension or probable ruptured abdominal aortic aneurysm (AAA). Sustained hypotension was defined as systolic blood pressure <90 mmHg or mean arterial pressure <65 mmHg for 60 minutes after 1 L of intravenous fluid. Profound hypotension was defined as systolic blood pressure <60 mmHg of any duration. Exclusion criteria were: under 16 years old, moribund (not expected to survive 24 h), likely to be discharged within 24 h, drug overdose, treatment limitation order, absent urinary catheter within 4 h of ED admission, already on dialysis, longer than 4 h since admission to the ED, or patients having had inter-hospital transfer unless arrival in the ED occurred less than 3 h after the initiating event, and the patients had not already been admitted to the ICU. Informed consent was obtained from patients, or a family member. Here we present the primary analysis of this study. Partial results for some patients (after cardiac arrest) have been published previously to illustrate an analysis of combining volume and creatinine kinetics [9].

First urine samples were collected during urinary catheterisation in the ED. Routine plasma samples taken on presentation to the ED were retrieved for additional assay. Both plasma and urine samples were collected upon admission to the ICU, and at 4, 8, and 16 h post ICU admission, and at 2, 4, and 7 days. Plasma samples were assayed for creatinine, neutrophil gelatinase-associated lipocalin (NGAL) and cystatin C (CysC), and urine samples were assayed for alkaline phosphatase (AP), γ-Glutamyl transpeptidase (GGT), NGAL, α- and π-glutathione S-transferase (GST), CysC, and albumin. Plasma and urinary creatinine were measured by the Jaffe reaction on an Architect c8000 analyzer using Abbott reagents (Abbott Laboratories, Abbott Park, IL, USA). Urinary AP was measured by p-nitrophenol rate reactions and GGT by γ-Glutamyl-p-nitroanilide rate on the Architect c8000 analyzer using Abbott reagents (Abbott Laboratories, Abbott Park, IL, USA). Urinary α-GST and α-GST assays were performed by Argutus Medical using enzyme immunoassays (Argutus Medical Alpha GST EIA-BIO-91 and Pi GST EIA-BIO85, Argutus Medical, Dublin, Ireland). Plasma NGAL was analysed using the Triage® NGAL test (Biosite, San Diego, CA, USA). Urinary NGAL was analysed using The Architect Urine NGAL assay (Abbott Diagnostic, Wiesbaden, Germany). Plasma and urinary CysC were measured on a BNII nephelometer (Dade Behring, GmbH, Marburg, Germany) by particle-enhanced nephelometric immunoassay. For AKI outcomes biomarkers were not normalised to urinary creatinine as we have previously shown that for AKI outcomes normalisation reduces the area under the receiver operator characteristic curve (AUC) [10]. Conversely, we showed that for mortality and need for dialysis outcomes normalisation to urinary creatinine improves the AUC, therefore, for the prediction of mortality and dialysis outcomes, we normalised the urinary biomarkers to urinary creatinine.

For each patient, the time of commencing renal hypoperfusion was adjudicated from the time of the ambulance call (for cardiac arrest), or when symptoms of severe shock, such as mental obtundation or hypotension, were documented (for hypotension), or the time of onset of abdominal pain (for suspected ruptured AAA). Baseline plasma creatinine was determined from a chart review, according to the hierarchy: (1) pre-admission creatinine within 7 days to 1 year prior to hospital admission; (2) creatinine at 30-day follow up; (3) hospital discharge creatinine; and (4) creatinine value on admission. AKI was diagnosed using the plasma creatinine criteria, kidney disease improving global outcomes (KDIGO): an increase of more than 26.5 μmol//l (0.3 mg/dl) within 48 h, or a relative increase of more than 50% above baseline within 7 days of admission [11]. Urine output was not used to classify AKI because it was not measured during the early course of the disease in the ED and because of its poor prognostic ability for meaningful clinical outcomes [12],[13]. Severity of illness was assessed in each patient during the first 24 h of admission using the acute physiological and chronic health evaluation II (APACHE II) [14]. For the prediction of mortality and dialysis we used the maximum normalised urinary biomarker concentration within 24 h and we evaluated their AKI status during this time (AKI₂₄), using as baseline creatinine the pre-admission baseline creatinine where it was known, or the first creatinine measurement in the ED. In this way the prediction is based on clinically relevant variables.

Between 24 March 2010 and 29 February 2012, 109 patients were recruited. Fifteen patients were excluded because they were not admitted to the ICU. In the ICU, another 17 patients were excluded: 3 did not fit the inclusion criteria, 6 were not expected to survive 24 h, 1 underwent an inter-hospital transfer, 1 a drug overdose, 2 refused to give consent and 4 had no samples taken in the ICU because of clinical priorities (Figure 1). Thus, 77 patients were included in the analysis: 49 (64%) after cardiac arrest, 22 (29%) after sustained hypotension (14 secondary to septic shock), 5 (7%) with ruptured AAA, and one with profound hypotension. Baseline characteristics are presented in Table 1. A total of 45 patients (58%) developed AKI. There were no differences in the demographic profile or baseline co-morbidity between patients with and without AKI. The baseline creatinine used was the pre-admission creatinine in 43 patients, 30-day follow-up creatinine in 18, final creatinine on hospital discharge in 13, and admission creatinine in 3 patients. There was no difference in source of baseline creatinine between those with and without AKI. Amongst survivors, there was no significant difference in the duration of the ICU stay, hospital stay, or mechanical ventilation between those with and without AKI. Of 45 AKI patients, 28 were at severity-stage 1, 8 were stage 2 and 9 were stage 3. A total of 27 patients (35%) died in hospital and 9 (12%) needed dialysis: 31 patients (40%) developed the composite outcome of death or dialysis. The median (interquartile range) between onset of injury and the first ED sample was 0.9 h (0.6 h to 1.1 h) and between the first ED sample and first ICU sample was 1.0 h (0.6 h to 3.2 h).

Contrary to our hypothesis, earlier biomarker measurement in the ED did not improve prediction of AKI. Plasma cystatin C and NGAL and urinary NGAL all diagnosed AKI in the ED and maintained that diagnostic utility for at least 24 h. Urinary cystatin C and GGT were only diagnostic in the ICU, and in the case of GGT only for a short period of time. The urinary concentrations of GGT, AP, and GST decreased rapidly with little differentiation between those with and without AKI. This highlights that these biomarkers are markers of very early injury, and do not correlate well with change in glomerular filtration rate (GFR). Indeed, the main driving forces of their profiles are the reservoir of available biomarker, which may be released upon injury to the brush border [19]. Albumin decreased similarly with time, although perhaps less rapidly in patients with AKI. Urinary NGAL and to a lesser extent, plasma NGAL and CysC, illustrated distinct elevation in most AKI patients over that of non-AKI patients during the 24 h following ED presentation. For plasma CysC, this reflects the loss of GFR and shorter half-life compared with plasma creatinine. Both plasma and urinary NGAL concentrations are partly affected by loss of GFR as well as tubular injury resulting in upregulation of NGAL and release into both the urine and plasma [19],[20]. The differences in temporal profiles highlights the importance of choosing the right temporal window in which to measure these biomarkers. Peak urinary NGAL and albumin within 24 h of presentation were independently predictive of mortality after adjustment for AKI status and illness severity score (APACHE).

A low concentration of urinary α-GST in the ED and ICU was also diagnostic of AKI. This may reflect that α-GST was released even more rapidly than the other biomarkers and was depleted by the time of measurement in the ED. Urinary GGT, α-GST and α-GST were elevated in the ED for nearly all patients. Compared to the cut points for prediction of dialysis need or 30-day mortality determined from the EARLYARF trial [21] GGT was 2 to 4 times greater, α-GST 100 to 1,000 times greater, and α-GST 10 to 100 times greater, regardless of whether subjects had AKI or not, suggesting that these tubular enzymes are very sensitive to renal hypoperfusion, with early release even after mild injury. The significance of an increase in tubular enzymes remains to be tested in a larger patient cohort. As these assays are cheap and rapidly performed, their utility in detecting a milder form of injury remains a potential area for exploration.

Albumin excretion was high in patients without AKI, perhaps because of endothelial dysfunction secondary to inflammation, a known consequence and independent predictor of mortality in critically ill patients [22],[23]. Albumin remained increased longer in AKI, possibly due to proximal tubular injury, which reduces albumin reabsorption [24],[25].

Induced, biomarkers such as urinary NGAL, and freely filtered biomarkers such as CysC, increase following injury to the proximal and distal tubules. Urinary NGAL is expressed in the distal tubules and collecting duct. NGAL expression in the distal tubules is upregulated in AKI. NGAL is also absorbed into the circulation and filtered later by the glomerulus [26],[27]. Reabsorption of filtered NGAL along with other low molecular-weight proteins is impaired in the presence of proximal tubular injury, further increasing urinary excretion [28],[29]. Urinary CysC is freely filtered in the glomerulus, and similarly reabsorbed by the proximal tubule. In the presence of proximal tubule injury, reabsorption of CysC is similarly impaired, leading to increased urinary excretion [30]. These biomarkers are released early following insult, and their concentration remains high for a longer duration [4],[31]. Their utility in diagnosis of AKI and prediction of mortality has been documented in many studies across different population groups [32]-[36].

We suggest that because of the inherent limitations in using plasma creatinine as a gold standard, the performance of structural biomarkers is best assessed against a meaningful clinical outcome, such as mortality or dialysis. Based on our previous study urinary biomarkers normalised to urinary creatinine are most likely to be of prognostic value [10]. Normalised urinary NGAL and albumin were independently predictive of mortality or dialysis. A 10-fold increase in urinary NGAL or albumin increased the odds of mortality or dialysis by approximately 3-fold. Similarly, Siew et al. showed that urinary NGAL was independently predictive of mortality or dialysis in a study of 451 critically ill patients [36]. Albuminuria has also been shown to predict mortality in critically ill patients [22],[23].

Both urinary NGAL and albumin improved the clinical model incorporating the APACHE II score and AKI in the prediction of mortality or dialysis. Plasma CysC, and urinary α-GST improved the risk prediction model to a lesser extent than urinary NGAL. Of interest is how much the improvement of risk assessment is clinically important. The question remains as to whether the average increase in risk of 6% identified by addition of urinary NGAL in those who died or were dialysed (IDI_event), and the average reduction in risk of 4% in those who survived or were dialysis-free (IDI_non-event), is sufficient to warrant investment in this additional test? Will the indication of structural injury to the kidney change practice? The answer to these questions will only emerge as clinicians engage with the new biomarkers, and learn to use them in high-risk cohorts at the appropriate times.

Overall in this cohort of patients at high risk of AKI and with well-documented timing of renal hypoperfusion, plasma NGAL, CysC and urinary NGAL best diagnosed AKI in the ED and continued to perform best over 24 h. The diagnostic performance of six of nine candidate AKI biomarkers (plasma CysC and NGAL, urinary α-GST, α-GST, AP and albumin) were equivalent when measured in the ED or in the ICU. The remaining biomarkers (urinary NGAL, CysC, and GGT) better diagnosed AKI at the later time point on entry to the ICU. The evaluated biomarkers, GGT and AP, facilitated detection of tubular cell injury in the ED, but without apparent diagnostic utility. Peak urinary NGAL within 24 h of admission best predicted the composite outcome of mortality or dialysis requirement.

AMdR consented patients, collected data, undertook analysis, and wrote the first drafts of the manuscript. JWP co-designed the study, obtained grant funding, supervised AMdR, undertook analysis and co-wrote the manuscript. GS participated in the design, oversaw data collection in the ICU, and provided feedback on the manuscript. MT participated in the design, oversaw data collection in the ICU, and provided feedback on the manuscript. PG supervised the development of core biochemistry and AKI biomarker assays and provided logistical support for the research staff working on the samples, ZH co-designed the study, obtained grant funding, supervised AMdR, provided input to the analysis, and co-wrote the manuscript. All authors read and approved the final manuscript.