Adjudication of etiology of acute kidney injury: experience from the TRIBE-AKI multi-center study

Acute Kidney Injury (AKI), regardless of the definition employed, is a common complication following adult cardiac surgery [1, 2]. Increasing acceptance of standardized definitions of AKI amongst nephrologists and intensivists has advanced several aspects of AKI research yet the practical issue of what constitutes “true AKI” clinically, remains unsettled [3‐8]. Novel biomarkers have demonstrated promising associations with AKI [3, 4], leading to a resurgence in clinical trials in the setting of AKI [9]. However, given the suboptimal performance of some biomarkers, there is only limited data to support the role of accurate adjudication of AKI to improve the diagnostic and discriminative performance of traditional and modern markers of renal function [10].

Recent research has focused on augmenting serum creatinine’s limited ability to detect true renal tubular injury, rather than a temporary change in glomerular filtration (pre-renal azotemia, PRA) [11‐15]. Conceptually, PRA, defined as a transient decrease in the effective perfusion of the kidney, either from volume depletion or relative hypotension, should spare damage to the kidney/renal tubules in comparison to other intrinsic (intra-renal) causes of kidney injury [16, 17]. Some maintain that attempting to clinically distinguish PRA from intrinsic kidney injury is not feasible and likely unnecessary as all AKI lies on a spectrum [18]. A limited number of non-cardiac surgery studies have had varying success in attempting to adjudicate AKI events and tease out the ability of novel biomarkers to distinguish PRA from direct tubular injury [10, 14, 15].

Separating PRA from intrinsic forms of AKI, including its most severe form acute tubular necrosis (ATN), is crucial as future clinical trials will be designed and powered to treat those with evidence of cellular kidney death (ATN). Thus recruiting, enrolling and studying patients with ATN is of the utmost importance, ensuring that we are treating a cohort of patients at the highest risk for the most severe forms of AKI and the worst clinical outcomes. Enrolling patients who have PRA, and who will not develop ATN will not only, increase study costs and weaken findings but also potentially expose patients to a therapeutic that they will not benefit from and thus increase patient risk. Pragmatically, all AKI (PRA, ATN, obstruction, athero-embolic disease and other sources) are at their simplest defined in terms of a rise in serum creatinine and/or a decrease in urine output, thus differentiation utilizing these two traditional tools is difficult. Compounding this dilemma, prolonged PRA may progress across the AKI spectrum to overt ATN over a patient specific time course. Unfortunately the pathognomonic “muddy brown casts” of ATN are not always present early in the course of AKI or at the time of renal consultation, rendering ATN a diagnostic possibility even in the absence of these clinical findings [19]. Finally, several of the other tools (fractional excretion of sodium and urea) used to distinguish between PRA and ATN have been shown to be ineffective following cardiac surgery [11, 20].

In this nested prospective cohort sub-study of the Translational Research Investigating Biomarker Endpoints in AKI (TRIBE AKI) study, we established a framework for the AKI adjudication process as well as provide adjudication outcomes in adults who developed AKI [3]. Additionally, we investigated the association of several AKI biomarkers (traditional and modern) with the adjudicated AKI outcome.

We enrolled 478 of patients in this sub-study Three subjects were excluded due to violations of inclusion/exclusion criteria, leaving 475 subjects in the final analyses. Sixty-seven (14%) of these individuals developed AKI, defined by RIFLE “Risk”. Of those individuals who developed AKI, 6 (1.2% of the total, 9% of those with AKI) went on to require RRT. Of the 67 subjects with AKI, there was 100% agreement amongst the 3 adjudication panelists on only 13 cases (19%). In 11 (16%) subjects, there was complete agreement that ischemic ATN was the cause of AKI; while PRA was the adjudicated diagnosis in the remaining 2 cases (3.0% of the total). In none of the 67 cases did all 3 panelists agree that the cause of AKI was indeterminate (Table 1).

In the 56 (81%) remaining cases of AKI there was not a unanimous consensus; however, in 46 (69% of the total) cases, 2 adjudicators agreed on a diagnosis. The most common scenario was 2 reviewers diagnosing ATN with the third diagnosing PRA (25% of the total cases); a complete list of the adjudications can be found in Table 1. There were 10 (14.9%) cases in which all 3 panelists originally disagreed about the underlying diagnosis, with each panelist choosing separate options; PRA, ATN and Indeterminate. During an in-person meeting these 10 cases were eventually judged to be 1 case of PRA, 3 cases of ATN and 6 cases of indeterminate. The clinical characteristics of the adjudicated sub-groups (ATN, PRA and Indeterminate) are depicted in Table 2. Individuals adjudicated to have ATN were more likely to have a higher peak post-operative serum creatinine, lower urine output during the first 48 post-operative hours and a longer duration of AKI compared to those with PRA. Those judged to have PRA were more likely to have pre-operative exposure to radio-contrast.

Investigation into the diagnosis patterns of the 3 individual adjudicators demonstrated that despite receiving the same instruction set and diagnostic materials, there was tremendous variability across all 3 panelists (Table 3). The adjudications of our 3 panelists were statistically independent of each other (Fleiss’ Kappa = 0.046). However when looking at individual adjudicator pairings no pairing provided a Cohen’s Kappa < 0.05. (Adjudicator 1 and 2 Cohen kappa = 0.24; Adjudicator 1 and 3 Cohen kappa = 0.06; Adjudicator 2 and 3 Cohen kappa = 0.10).

Table 4 demonstrates the mean pre and post-operative biomarker values throughout the first 5 days for all subjects adjudicated to either ATN or PRA. No biomarker was significantly different across these 2 groups during the first 24 post-operative hours. There was a statistical difference between KIM-1 values at Day 2 but this effect was not sustained over subsequent post-operative days. Serum Creatinine and Serum Cystatin C were not statistically different until Day 3 with both of these effects being sustained over the next 2 post-operative days.

The development of AKI following cardiac surgery is extremely complex. Preoperative factors such as exposure to nephro-toxins (e.g. radio contrast from cardiac catheterization) and pre-existing conditions (congestive heart failure, diabetes mellitus and chronic obstructive lung disease) contribute and interact with operative factors such as emergent surgery, intra-operative hypotension and cardio-pulmonary bypass (CPB) pump time in determining AKI and patient outcomes. The interplay of these factors along with the patient’s post-operative course (e.g. mechanical ventilation, use of vaso-active medications, balloon pumps or diuretics) all may contribute to the development of AKI. As such determining the etiology of AKI while relying on the clinical course and changes in serum creatinine and urine output is therefore often a subjective and challenging task. As AKI research moves into an era of clinical trials, it will be increasingly important to ensure that all subjects enrolled with AKI have a similar etiology of renal dysfunction as interventions aimed at ischemia-induced ATN may not work in nephrotoxin or contrast related injury, let alone in the setting of functional PRA. Just as previous interventions have failed, in part, because they were initiated too late, future agents may not work if they are investigated across a broad spectrum of AKI [27]. Our initial experiences with the adjudication of cardiac surgery associated AKI demonstrate that while feasible, this process is not perfect.

It is extremely important that clinical definitions of AKI are harmonized beyond a simple elevation in serum creatinine or decrease in urine output. Our experience demonstrates that not all AKI following cardiac surgery is the same and that a clean, precise, one-hit phenotype likely does not exist. In the past cardiac surgery associated AKI was thought to be a homogenous clinical entity predominantly due to renal ischemia-reperfusion injury however over the last decade investigators have recognized that it is a multi-factorial process that involves ischemia-reperfusion, inflammation, variation systemic hemodynamics, athero-embolic disease and nephrotoxins [28, 29]. With so many factors to consider it is not surprising that there was a large degree of variation in the adjudicated diagnoses across our 3 panelists with all 3 adjudicators agreeing only 19% of the time. Their discrepant clinical decision making when possessing identical information is readily apparent in Table 3. In the end, 41 (61%) subjects were judged to have ATN while 13 (19%) had PRA with several of these cases requiring an in-person discussion in order to arrive at a final diagnosis. It is important to reiterate that the evaluation process was done retrospectively months after the hospitalization and that all three reviewers had access to the same clinical information, but not the novel biomarkers data. Thus, at most, only 60% of AKI subjects would have been appropriate for enrollment in a therapeutic study to treat or diminish the effects of ATN. This is an important issue, as future trials will seek to identify cohorts in the early post-operative period who are at risk for ATN while attempting to minimize the enrollment of those with functional forms of AKI. In our study, differences in glomerular filtration markers (Serum Creatinine and Cystatin C) were not apparent until post-operative day 3 (Table 4). Such a timeframe is likely outside of the therapeutic window for many novel treatments that require early access to the injured nephron. Relying solely on serum creatinine kinetics therefore is not a viable option for achieving diagnostic distinction.

Not surprisingly, given the poor agreement across our 3 adjudicators, no biomarker, consistently distinguished etiology of AKI within the first 48 post-operative hours. While there is limited data to support the notion that novel biomarkers have the ability to discriminate ATN from PRA [14, 15, 30]; our data does not corroborate this, but must be interpreted in light of the lack of consensus amongst the adjudicators. In the past it has been argued that elevation of urinary markers of renal tubular injury with concomitant increased in serum creatinine would indicate true ATN; however, given our inter-rater variability, we could not verify this [17]. Importantly, some of these past studies used specific time parameters to differentiate PRA from ATN while our adjudicators where free to determine a diagnosis regardless of AKI duration. However, despite this difference our ATN cases were of a significantly greater duration compared to PRA. While biomarkers were not included in our adjudication process, we endorse the idea that future research should investigate the utility of biomarkers in distinguishing ATN from other types of kidney injury in a variety of clinical settings.

As nephrologists advance AKI clinical research and re-engage the idea of prophylactic and therapeutic interventional trials we need to 1) recognize that while cardiac surgery produces a reliable rate of AKI, not all elevations in serum creatinine are pathophysiologically equivalent and 2) develop a reliable and consistent system to adjudicate AKI patient diagnoses. It would be unfounded to assume that individuals with PRA have the same risks (short and long term) as those with ATN; yet the published literature often lumps these two clinical entities together to encompass changes in serum creatinine [5, 6]. For decades cardiologists, intensivists and oncologists have refined their ability to define patient diagnoses, yet to date there has been little movement on this issue in the field of AKI. More recently clinical adjudication has begun to be relevant and appeared in the AKI literature [10, 14, 15, 30]. However, in these few cases the AKI in question was diverse and the adjudication sought to identify those subjects with PRA in order to determine how well novel biomarkers correlated with AKI outcomes. Additionally, these studies measured biomarkers at the time of clinical AKI or time of peak serum creatinine and our study measured biomarkers in the early post-operative period which may help explain the lack of association in our study.

Our study was limited in that we could only adjudicate 67 cases of AKI. Additionally, our adjudication process was limited by the lack of urinalysis microscopy data. We recommend that future studies should attempt to include some element of urine sediment analysis as microscopy reporting is increasingly standardized and protocol driven allowing for inter-study comparison [31‐33]. However, microscopy remains a user dependent tool that displays a tremendous amount of inter-physician variability [34]. Wald and colleagues demonstrated that despite the importance of urinalysis in the diagnosis of renal injury, nephrologists generally achieve only fair to moderate levels of agreement in the identification of important structures on urinalysis and that agreement between nephrologists was not enhanced based on their level of experience. Thus the inclusion of microscopy may have only further muddied our adjudication process given that the pathognomonic “muddy brown casts” are not always present at the time of microscopy even in cases of a known diagnosis of ATN [19]. Additionally, had microscopy been included in the early post-operative period (e.g. ICU arrival), it’s utility remains unclear as there is no published data on the ability of urine microscopy to predict patient outcomes following adult cardiac surgery.

For the past decade several other branches of medicine have succeeded in formalizing the process of patient event adjudication and utilized the results to help further medical research. Over the next decade, adjudication of AKI events will be crucial during the data collection and event analysis of both interventional and prophylactic AKI trials. Our study demonstrates that while adjudication is technically feasible, the interpretation of clinical data in the setting of cardiac surgery remains a challenge. Our experiences, while not perfect, will hopefully provide the building blocks of a formalized process that will allow for inter-study comparison of AKI events. Once a standardized format for adjudication is in place, the role of novel biomarkers in this process can be explored but, currently, such a role remains incipient. Further work is needed to ensure that in the setting of AKI following cardiac surgery, one nephrologist’s PRA is not another nephrologist’s ATN. Until the capacity to produce unified descriptions/definitions is in place, we are instead left with a definition of cardiac surgery associated ATN that is extremely subjective.

In this nested prospective cohort sub-study of the TRIBE AKI study 56% of those with AKI were adjudicated to have ATN (by at least 2 of the 3 reviewers). Demonstrating that over one-third of patients with AKI may not derive a benefit from interventions aimed at treating or repair tubular injury. Our experiences have led us to conclude that the etiology of AKI after cardiac surgery is multi-factorial and pure AKI etiologies, such as ATN and PRA may not exist. In our study biomarkers did not correlate with the adjudicated etiology of AKI; however the lack of agreement among the adjudicators impacted these results. Further investigation and formalization of the AKI adjudication process, and its interaction with AKI biomarkers, is needed in order to ensure equality in AKI etiologies in future critical care nephrology clinical trials.