Biomarker Enrichment in Sepsis associated Acute Kidney Injury: Finding High Risk Patients in the ICU

Abstract

Background:

Sepsis associated Acute kidney injury (AKI) is a leading comorbidity in admissions to the intensive care unit. While a gold standard definition exists, it remains imperfect and does not allow for the timely identification of patients in the setting of critical illness. This review will discuss the use of biochemical and electronic biomarkers to allow for prognostic and predictive enrichment of patients with sepsis associated AKI over and above the use of serum creatinine and urine output.

Summary:

Current data suggest that several biomarkers are capable of identifying patients with sepsis at risk for the development of severe AKI and other associated morbidity. This review discusses this data and these biomarkers in the setting of sub-phenotyping and endotyping sepsis associated AKI. While not all these tests are widely available and some require further validation, in the near future we anticipate several new tools to help nephrologists and other providers better care for patients with sepsis associated AKI.

Acute kidney injury (AKI) is a frequent clinical syndrome in the intensive care unit (ICU) setting, with up to 50% of all ICU admissions having a diagnosis of AKI [1-3]. In the ICU AKI is associated with worse outcomes, including increased inpatient and long-term post-discharge mortality as well as increased morbidity such as continued dialysis requirements, increased length of stay and the development of post-AKI chronic kidney disease [4].

The international consensus definition of AKI by the Kidney Disease Improving Global Outcomes (KDIGO) group is based on changes in urine output (UOP) and serum creatinine (SCr) changes (Table 1). This review will focus on the emerging and established data regarding traditional and novel biomarkers of AKI in the setting of sepsis. However, the definition of AKI has varied across disciplines and has change several times over the last 20 years. This definition evolution supports the need for further classification of AKI, however, the more recent trials discussed throughout this review all use the KDIGO definition. KDIGO defines AKI with serum creatinine via both absolute rise, percentage rise over pre-defined time periods and through decreases in urine output over different pre-defined time periods.

Table 1.

Kidney Disease Improving Global Outcomes AKI Staging

	Creatinine	Urine Output
Stage 1	-Increase in creatinine ≥0.3 mg/dl or 1.5 to 1.9x baseline	<0.5mL/kg/hour for 6 to 12 hours
Stage 2	-Increase in creatinine to 2.0-2.9x baseline	<0.5mL/kg/hour for 12 to 24 hours
Stage 3	-Increase in serum to ≥ 3x baseline OR -Increase in serum creatinine of ≥ 0.30mg/dl to ≥ 4.0mg/dL OR -RRT initiation	<0.3mL/kg/hour for 24 hours or anuric for 12 hours OR -RRT initiation

Prior to the publication of the KDIGO definitions AKI was defined via the RIFLE and the AKI-Network (AKIN) criteria. [5]RIFLE The “Risk, Injury, Failure, Loss of kidney function, and End stage Kidney disease(RIFLE) criteria were first proposed by the Acute Dialysis Quality Initiative (ADQI) group in 2004 to better classify AKI [6]. While the urine output portion of the RIFLE criteria are the same as the KDIGO. The RIFLE criteria only focused on percentage increases in serum creatinine, not absolute changes in creatinine and included percent changes in estimated glomerular filtration rate (eGFR). It was the AKIN criteria that added the concept of a 0.3 mg/dl increase over 48 hours to the definition of AKI, recognizing that in patients with advanced pre-existing kidney disease (e.g. baseline creatinine of 2.4 mg/dL) a 50% increase would be hard to achieve but that increasing creatinine to 2.8 (or higher) constituted a clinically meaningful loss of kidney function. Additionally, the AKIN criteria removed the use of eGFR, as AKI patients are not in steady-state and eGFR calculations do not accurately reflect true underlying kidney function. Lastly, AKIN added the use of renal replacement therapy (RRT) to stage 3 AKI. As you can see in Table 1 the KDIGO criteria serve as a melding of the RIFLE and AKIN criteria and remain the current gold standard for the epidemiologic definition of AKI.

This is not to say that other definitions of AKI do not exist, there are several out there in the AKI literature including worsening renal function (WRF) in the setting of acute decompensated heart failure (often defined as a 0.3 mg/dl increase) or contrast associated AKI (CA-AKI) (often defined as a 0.5 mg/dL absolute increase or a 25% relative increase over 72 hours). However, these definitions have been as widely validated as the RIFLE or KDIGO and are not applicable to the sepsis population. While patients with sepsis are often exposed to radio-contrast and at risk for CA-AKI the studies we discuss below will all define AKI based on the internationally recognized KDIGO consensus criteria.

Others have recently suggested that the KDIGO definition of AKI needs to move beyond serum creatinine and urine output [7] Recent data demonstrates that other biomarkers of kidney function have clinical utility and are highly associated with clinical outcomes in both those with increased risk of AKI before there are changes in SCr or UOP as well as those with the early stages of formally diagnosed KDIGO AKI [8-10]. Subclinical AKI has been defined as the evidence of kidney damage (as measured by increased biomarkers) without a rise in creatinine or change in urine output [11]. It is important to recognize that markers of tubular dysfunction do not correlate precisely with markers of eGFR.[12] In fact Bhatraju and colleagues have recently demonstrated that when compared to tubular secretion in “healthy adults” the tubular secretion in 170 critically ill patients with sepsis was altered and was only moderately correlated with serum creatinine and cystatin C (r=−0.51 and −0.53 respectively). Thus, tubular markers of injury may be able to augment our diagnostic and prognostics ability beyond those of serum creatinine and urine output. In this review we will discuss the current level of evidence to support the diagnosis of “Sub-clinical AKI” specifically in the setting of sepsis and septic shock and explore why this entity of “biomarker positive” AKI is no longer truly “subclincal” and should be thought of as standard AKI [13].

In addition to these emerging definitions, within broad clinical scenarios of AKI (e.g. sepsis or cardiac surgery associated) patients may present as several different unique phenotypes, each carrying a different prognosis in regards to severity and duration of AKI as well as different underlying biological features [7, 14]. A phenotype in reference to AKI (and elsewhere) refers to the clinical presentation of AKI in a specific setting, and by current definitions must be accompanied by changes in serum creatinine and/or urine output. Examples of phenotypes include sepsis-associated AKI, Cardiac surgery associated AKI, amongst others. This is separate from an endotype which is defined as variant of a given condition (like AKI) by distinct functional or pathobiological mechanism (e,g, hyperinflammation from sepsis / infection).

Sepsis associated AKI is defined as AKI defined by KDIGO criteria occurring within 48 hours to 7 days of the clinical onset of sepsis based on sepsis-3 criteria [15]. Sepsis is a dysregulated immune response resulting in both immune system up-regulation, compromise, and a variety of molecular changes systemically resulting in the clinical presentation of sepsis [15] The dysregulated host response includes changes to the RAAS[16, 17] Pathogen Associated Molecular Patterns (PAMPs) and Damage Associated Molecular Patterns (DAMPs) signaling [18]. These cytokine and neuro-hormonal based changes, which are beyond the scope of this review, occur on a continuum and range from changes to filtration (rise in creatinine) without tissue/tubular damage in the setting of time-limited hemodynamic AKI to tissue damage with rising biomarkers (for example, ATN)[19].

Within the realm of sepsis associated AKI there are several plausible phenotypes often within the same patient. In an older adult with septic shock requiring mechanical ventilation for their community acquired pneumonia (treated with broad spectrum antibiotics), is their AKI due to the ischemia/hypotension from the shock; the inflammatory response from the infection itself; the shift in hemodynamics from mechanical ventilation or unintended exposure to potential nephrotoxins such as piperacillin/tazobactam or vancomycin? As it currently stands, the tools to differentiate these clinical entities are imperfect; but as discussed below, sub-phenotyping AKI in the setting of sepsis via traditional (SCr and UOP) biomarkers and newer biomarkers remains an area of intense investigation [16, 8, 9] . The clinical prognostic value of identifying the unique AKI phenotpyes (and endotypes) is emerging and not yet fully established, However early data do suggest that tubular damage markers are associated with worse outcomes and can be highly correlated to the other molecular signatures of severe AKI [11, 20, 21] Figure 1

Figure 1 -.

This figure demonstrates the possible pathophysiology of sepsis associated AKI. These sepsis associated AKI endotypes are provided as examples, and may change over time as more data is presented. As a schema, the sub-phenotypes of AKI in sepsis are now described as both traditional KDIGO classification and as based on biomarker positive states. This figure which was previously published by the Acute disease quality initiative (ADQI) represents a proposed roadmap as well for future thought on sepsis-associated AKI. The description of tissue tolerance refers to a state where there is no significant change in biomarker concentrations. Tubular markers of damage or stress such as TIMP-2/IGFBP7, may serve a different role than serum creatinine which is a marker of glomerular filtration that indicate the AKI is primarily driven by changes in filtration. Sepsis associated AKI is defined here as AKI adjacent to sepsis, but not driven solely by sepsis; so it may include injury from nephrotoxic medications, hypotension, and volume depletion. -Adapted from [61]

The sub-phenotyping of AKI is a new paradigm and may point to further evaluation of patients with AKI using biomarkers to better understand that AKI can be both heterogenous and homogenous at the same time. Using our prior example, it would not be surprising that prior attempts to treat sepsis associated AKI have failed if we think that all patients with sepsis and AKI have been lumped together due to the inability to define a shared pathobiology. Prior AKI studies have used biomarkers for prognostic enrichment; defined as identifying a cohort at high risk for specific outcomes (e,g. need for RRT or inpatient mortality) however fewer have used them for predictive enrichment; identifying those with a shared underlying biology[22-24]. This is it not surprising; drugs that target repair of ischemic injury do not help patients who have a nephrotoxin or inflammatory mediated AKI and drugs that quiet the inflammation of AKI do not help repair ischemia. While this paper will focus on using biomarkers to sub-phenotype sepsis associated AKI, this novel concept may be able to be applied to more types of AKI using some of the same biochemical biomarkers (e.g. cardiac surgery associated AKI or cardio-renal AKI).

Additionally, there has been a recent movement away from traditional and novel biochemical biomarkers (blood and urine based) and using an “electronic biomarker” such as machine learning or other augmented intelligence generated tools., These “electronic biomarkers” often combine traditional biomarkers (SCr, UOP) with other novel and traditional markers and clinical features (e.g. vitals, intake/output status, standard labs like blood counts or hepatic function), to generate a risk assessment tool for the future development of AKI. These electronic biomarker-based risk scores are yet another way to identify high risk sepsis-AKI patients earlier. If validated in future investigations, they may help in separating out a heterogeneous AKI-risk population and identifying a homogenous subpopulation to provide predictive enrichment.

AKI Biomarkers

At this time, all of AKI is broken into KDIGO stages, without regard to the initial etiology, modifying factors or even non-SCr/UOP based biomarkers, and only recently is the addition of tubular biomarkers being introduced[7]. The APACHE II is a risk score designed and validated to predict mortality in ICU patients and calculating the score requires using both clinical (e.g. vitals) and laboratory (e.g. SCr) findings[25]. While it is not commonly calculated for clinical purposes, it is often used as part of study enrollment criteria or as a method to compare 2 distinct patient populations across the spectrum of critical illness. As a field we need tools that are not only useful for risk assessment but also practical to use at the bedside.

In this review we will cover the recent literature around the use of biomarkers for the prognostic and predictive enrichment in sepsis and more specifically sepsis associated AKI (Figure 2). There have been several biomarkers (biochemical and electronic) which have demonstrated promise in patient identification and risk stratification and others which have not. This review is not intended to be an exhaustive list but rather an opportunity for the reader to better understand the concepts around biomarker enrichment in the setting of sepsis associated AKI and its role in the setting of clinical trials and beyond.

Figure 2 -.

The figure demonstrates the way prognostic and predictive enrichment can be used to improve outcomes in ICU patients. Prognostic enrichment can identify those destined to have more severe AKI (e.g. death or need for RRT). In our example the majority of those at risk for RRT are those with Inflammatory and Ischemia induced AKI. Tools like this can be used to identify patients who may need enhanced kidney-focused care or specific interventions. While Predictive enrichment identifies those with a shared underlying pathobiology, perhaps to allow for enrollment in a clinical trial specifically ischemia induced AKI. Adapted and altered from [62]

Randomized Trials with Enrichment strategies for diagnosing AKI

Discussion of trials with biomarker enrichment will follow, with emphasis on biomarker performance, patient outcomes, and predictive models. The area under the receiver operator characteristic curve (AUC) provides insight into the accuracy of the tested diagnostic value as compared against the known outcome. An AUC of 1.0 is a perfect test while an AUC of 0.5 is equal to chance. In addition to AUC, hazard ratio (HR) and odds ratio (OR) will be discussed. Hazard ratios compare the hazard rate of the occurrence between the study population and the control population of a trial. An odds ratio is the ratio of the odds of an occurrence after exposure to a variable. The hazard ratio is used in prospective trials, while odds ratio is used in retrospective trials, and many of these biomarkers were initially identified and studied in retrospective studies, then applied to a prospective study, which why both HR and OR will be present.

Tissue inhibitor of metalloproteinase 2 and insulin like growth factor binding protein 7 (TIMP2*IGFBP7)

Sapphire/Topaz

Urinary tissue inhibitor of metalloproteinase 2 and insulin like growth factor binding protein 7 (TIMP2*IGFBP7) are markers of cell cycle arrest which were first described in a paper by Kashani and colleagues that collected and blood and urine samples in ICU patients at high risk for the development of severe (Stage 2 or 3) AKI. AKI risk was defined as the presence of AKI stage 1 or elevated Pulmonary SOFA, or Cardiovascular SOFA score. These urinary markers are clinically available in several countries and approved for severe AKI risk prediction. These markers are believed to be generated, in part, by the injured tubular cells specifically during episodes of kidney stress / tubular damage[9]. In a post-hoc analysis of the original discovery and validation cohorts (Sapphire cohort) and the subsequent follow up studies (Topaz) Honore and colleagues identified 138 patients from the Sapphire study and 101 patients from the Topaz study that had sepsis at ICU admission [26]. Within 12 hours, 192 patients did not develop Stage 2/3 AKI, while 40 patients developed severe AKI (Stage 2/3 AKI). The TIMP2/IGFBP7 levels were measured within 24 hours of admission. Baseline characteristics between severe AKI and no AKI were broadly similar on admission. TIMP-2*IGFBP7 was more accurate than serum creatinine at predicting the development of severe AKI. TIMP-2*IGFBP7 remained a significant predictor of severe sepsis associated AKI even after adjusting for APACHE, SOFA, serum creatinine, BMI and 24 hour inputs-outputs. The addition of TIMP-2*IGFBP7 to the clinical model increased the AUC 0.08 to 0.94 (0.90-0.98). An elevated TIMP-2*IGFBP7 value >0.3 was also associated with an odds ratio of 7.3 (3.1-17.5) for developing severe AKI.

ProCESS Trial and Utility of Biomarkers for Sepsis Associated AKI

The ProCESS (Protocolized Care for Early Septic Shock) trial was a prospective randomized controlled multi-center trial to assess 3 distinct resuscitation efforts (Early Goal Direct Therapy) in septic patients the emergency department (22). The original ProCESS trial failed to show a difference in patient outcomes with regard to its primary endpoint of 90-day mortality although there were some differences in fluid balance and AKI event rates across the 3 arms. In a post-hoc analyses of a subset of ProCESS patients the authors sought to assess how TIMP2*IGFBP7 biomarker levels changed before and after the initial 6-hour resuscitation protocol and examined how these TIMP2*IGFBP7 changes associated with patient outcomes (AKI and mortality). [20]

The author explored the incidence of elevated biomarkers across the cohort based on their KDIGO AKI staging. Where, as above a TIMP2*IGFBP7 level above 0.30 constituted a biomarker positive AKI state. In those with Stage 1 only 15% of patients were biomarker positive, this increased to 27% and 34.4% in Stage 2 and 3 AKI respectively [20]

In a separate analysis of the ProCESS cohort , Molinari, et al[20] investigated whether higher levels of TIMP2/IGFBP7 were associated with lower survival with the same stage of AKI. The primary outcome was survival at 30 days after enrollment. 999 patients were included from the original dataset. They found that biomarker positive patients, defined as TIMP2/IGFBP7 concentration greater than 0.30 in the setting of KDIGO defined AKI, had a lower survival 30 days with RR of 2.20( 95% CI, 1.02-4.72) in stage 1 AKI, stage 2 (RR, 1.53; 95% CI, 1.04-2.27), and stage 3 (RR, 1.61; 95% CI, 1.00-2.60). When no AKI was present, the level of TMP2-IGFBP7 was not correlated with changes in 30-day mortality. This is suggestive that TMP2-IGFBP7 concentrations in patients with sepsis and septic shock and AKI can be used to identify patients at higher risk for 30-day mortality and other adverse outcomes.

Fiorentino and colleagues also published data from ProCESS examining the ability of TIMP2*IGFBP7 to prognosticate the development of a composited primary endpoint of the risk for progression to stage 3 AKI, the receipt of dialysis, or death within 7 days. [9] The composite endpoint was reached in 113 (16%) patients. The TIMP2-IGFBP7 was measured at hour 0 and hour 6, (before and after the 6 hour protocolized resuscitation across the 3 arms of the study). For the purposes of their analyses, they called a TIMP2*IGFBP7 > 0.3 as positive. From this data, 4 subgroups emerged: 1.) negative TIMP2-IGFBP7 on admission/negative TIMP2-IGFBP7 hour 6; 2.) negative TIMP2-IGFBP7 on admission/positive TIMP2-IGFBP7 at hour 6; 3.) positive TMP2-IGFBP7 on admission and negative TMP2-IGFBP7 at hour 6; and 4.) positive TMP2-IGFBP7 on admission and TIMP2-IGFBP7 at hour 6. 8% of patients in group 1 reached the composite endpoint, 21.8% reached the composite endpoint in group 2, 9.8% reached the composite endpoint in group 3, and finally 23.8% of the patients in group 4 reached the composite endpoint.

The group 4 was at highest risk of developing the primary endpoint with elevated TIMP2-IGFBP7 before and after resuscitation. Those with an initially elevated TIMP2-IGFBP7 with an improved negative value at hour 6, demonstrated similar outcomes as patients in group 1,(negative at both timepoints) suggesting that the protocolized resuscitation improved / ameliorated the underlying septic pathophysiology was improved, Group 2, those who went from a negative to positive TIMP2*IGFBP7, demonstrated the opposite, with and increased risk of the composite endpoint compared to those in group 1 or 3 (Table 2).

Table 2-.

Change in TIMP2*IGFBP7 before and after resuscitation of septic shock

Pre-resuscitation Hour 0 Biomarker Status	Pre-resuscitation Hour 0 AKI status	Primary outcome achieved ratio; %
Positive	Positive	54/224; 24%
Negative	Positive	16/50; 32%
Positive	Negative	32/233; 13.7%
Negative	Negative	11/181 6.1%
Post-resuscitation Hour 6 Biomarker Status	Post-resuscitation Hour 6 AKI Status	Primary outcome achieved ratio; %
Positive	Positive	57/196; 29%
Negative	Positive	13/78; 16%
Positive	Negative	25/153; 16.3%
Negative	Negative	18/261; 6.9%

This is table summarizes data around the changes in TIMP2-IGFBP7 over the course of early resuscitation of sepsis in patients from the PROCESS trial. The Primary outcomes was the composite of development of Stage 3 AKI, receipt of RRT or death at 7 days Biomarker positive was defined as having a TIMP2*IGFBP7 > 0.3.

^1.Fiorentino M, Xu Z, Smith A, Singbartl K, Palevsky PM, Chawla LS, et al. Serial Measurement of Cell-cycle Arrest Biomarkers [TIMP-2]•[IGFBP7] and Risk for Progression to Death, Dialysis or Severe Acute Kidney Injury in Patients with Septic Shock. Am J Respir Crit Care Med. 2020.

In multiple analyses from separate cohort enrichment using TIMP2-IGFBP7 can predict with higher accuracy, those who will go on to develop severe AKI, the receipt of dialysis, or death. This enrichment is more impressive when you consider that TIMP2*IGFBP7 was able to frequently prognosticate these outcomes prior to the presence of any SCr or UOP-based AKI and prior to the start of the clinical resuscitation process. Thus, future therapeutic trials in the setting of sepsis and sepsis-associated AKI should consider utilizing this tool for enrollment/enrichment.

Proenkephalin

Proenchepalin (PENKID) is an endogenous opiod protein that has been extensively studied as a potential biomarker of kidney function in the ICU [27]. Penkid is freely filtered at the glomerulus and may serve as a marker for glomerular filtration with changes in PENKID occurring prior to those of serum creatinine or other functional markers of GFR. As with serum creatinine, plasma concentration of Penkid is strongly negatively associated with measured GFR, and should be considered a biomarker of function, not damage. Penkid has not been shown to be associated with other ongoing systemic effects such as inflammation.

The Kidney in Sepsis and Septic Shock Study (Kid-SSS)[8] study evaluated the ability of Penkid to detect the future development of AKI and other adverse events in patients admitted to the ICU with sepsis and septic shock. The primary end point of the Kid-SSS trial was the development of major adverse kidney events (MAKE) at day 7, defined as a composite of all-cause mortality, receipt of RRT, and persistent AKI at day 7 (elevated serum creatinine level from baseline by >1.5 fold or >0.3 mg/dl at day 7).

The Kid-SSS Study utilized data and samples from 2 prior sepsis / septic shock cohorts. The first was the Adrenomedullin and Outcome in Severe Sepsis and Septic Shock (AdrenOSS)[28] The AdrenOSS trial was a prospective, observational multinational study developed to assess the ability of adrenomedullin (a different biomarker) to predict 28-day mortality. Patients received standard ICU care and samples were collected at day 0, 1, and 2 of their ICU admission. This cohort served as the discovery cohort group to assess PENKID’s ability to predict MAKE-day7. The second septic shock and severe sepsis cohort in this study was the French and euRopean Outcome reGistry in Intensive Care Units[29]. FROG ICU was a prospective observational, multicenter cohort study of ICU survivors. The cohort included patients admitted to an ICU with invasive mechanical ventilation and/or vasoactive drug support. In both cohorts, AKI was defined using KDIGO guidelines and Penkid was measured at a central laboratory using standard techniques. Lowest detection limit is 5.5 pmol/L.

Penkid concentrations were assessed at ICU admission, and higher levels were associated with the future development of MAKE. Penkid was significantly elevated in patients who developed MAKE. Additionally, Penkid levels were found to be higher at ICU admission in other forms of AKI including persistent AKI which was defined KDIGO Stage 1 SCr-based AKI at day 7, RRT at day 7, or death within 7 days; and worsening renal function (WRF) was defined as elevated serum creatinine level from baseline by SCr based Stage 1 AKI at 48 hours post ICU admission (p<0.001 for both). As expected, Septic shock (as compared to just sepsis) was more frequent in patients who developed MAKE compared to those without MAKE, (71.6% vs 58.3%). Similarly, those who developed MAKE-7day outcomes had more incident CKD (6.8% vs 27.8%), hypertension, diabetes and were more likely to be in positive fluid balance compared to those who did not develop MAKE.

A subsequent analysis of over 2500 patients, from both cohorts, demonstrated that in the absence of clinical AKI (changes in serum creatinine and urine output) that elevations in PENKID were associated with increased adverse outcomes. Depret[11] and colleagues defined subclinical-AKI as a PENKID level above the normal range (>80 pmol/L) in the absence of KDIGO AKI. While this only occurred in roughly 6% of the combined cohorts, these patients had nearly a 2.5-fold increased risk of mortality compared to those with no AKI and no elevated PENKID. The hazard ratios were 2.4 (1.5-3.7) in the FROG-ICU cohort and 2.5 (1.1 to 5.9) in the ADRENOSS study when compared to those with no AKI. In patients with elevated eGFR or no AKI, elevated PENKID concentration above 80pmol/L alone was associated with a hazard ratio of 1.4 (1.1 to 1.8 95%CI) for mortality which increased to 1.6(1.3 to 1.8 95% CI) after adjusting for age, sex, creatinine, diuresis, and other co-morbidities. They additionally did analyses looking at the cohorts based on pre-enrollment GFR and demonstrated cutoffs for the ideal detection of 28-day mortality as well as demonstrating that penkid levels correlated with mortality in those with KDIGO defined AKI. However, all these retrospective analyses and cutoffs require future prospective validation.

More recently, Boutin, et al, investigated the urinary peptide signatures of patients with sub-clinical AKI from the FROG-ICU cohort. [30]. Defining subclinical AKI as no evidence of KDIGO criteria AKI, with elevated biomarkers (including cystatin c (urine and serum), neutrophil gelatinase-associated lipocalin NGAL (urine and serum), PENKID, liver fatty acid binding protein LFABP), and plasma Galectin-3 (pGal3). They compared outcomes subclinical AKI patients to those with traditional serum creatinine/urine output based KDIGO-AKI. The cohort had a total of 1885 participants with biomarker data, 1154 (61%) of whom did not have KDIGO AKI. Of those without AKI 346 patients (30%) were defined as having subclincal-AKI. In those with biomarker defined sub-clinical AKI the presence of elevated biomarkers varied with elevations across most biomarkers occurring in about one-third of patients ( pNGAL (32.4%), uNGAL (29.8%), uCystatin (31.5%), PENKID (38.9%), uLFABP (38.4%), and pGAL-3 33.5%). Only 9% of those with subclinical AKI tested positive for all biomarkers. The authors compared the urinary peptide signatures of those with subclinical AKI to those with and without KDIGO AKI to assess the presence of inflammation, hemolysis, and endothelial dysfunction. They specifically investigated peptides like alpha 1 antitrypsin and urinary albumin which have been previously shown to be associated with endothelial dysfunction and mortality in patients with KDIGO based AKI [31][32]. Subjects with no form of AKI (no biomarker or KDIGO AKI) did not have elevated urinary peptides, while the patients with sub-clinical AKI and AKI-KDIGO both had extremely similar peptides expression patterns. Additionally, mortality for those patients with sub-clinical AKI ranged from 32.6% to 46.8% across the individual biomarker positive sub-AKIs and mortality for AKI-KDIGO was 48%. PENKID diagnosed sub-AKI specifically was associated with a 42.9% mortality in this cohort. This data reaffirms that sub-clinical AKI is a clinically meaningful syndrome that may be missed by traditional markers; requires further investigation and should not be ignored by intensivists, nephrologists and other providers.

This idea that PENKID, or any biochemical biomarker, can be elevated even in the absence of changes in traditional AKI biomarkers is not new. ADQI and others first described this concept of biomarker positive and creatinine negative AKI over a decade ago[33, 34]. However, this PENKID data is specific to those with sepsis and septic shock and rather than relying on biomarkers of tubular damage, PENKID is providing information about GFR that is not being detected by “the gold standard” serum creatinine. Future studies need to demonstrate that intervening on those with an elevated PENKID in the absence of KDIGO AKI improves patient outcomes and mitigates the risk of morbidity and mortality. It is promising that using a tool like PENKID, can specifically identify a subset of septic shock patients destined for AKI and mortality earlier than current tools, perhaps allowing them to be enrolled in clinical trials or receive specific therapeutic interventions (e.g. AKI Care bundles).

Angiopoietin

Just as the PENKID data was developed as a secondary analysis, other enrichment biomarkers have been developed/validated through repurposing data. Recently, Bhatraju and colleagues have used data from the Vasopressin and Septic Shock Trial (VASST) to explore several new ways to identify phenotypes of sepsis associated AKI (14). Briefly the VASST[35] trial was a multicenter, randomized, double blind trial of norepinephrine versus vasopressin in septic shock patients. The primary endpoint was 28-day all-cause mortality and there was no difference across the 2 arms, 39.3% vs. 35.4% (RR 0.90; 95% CI 0.75-1.08; P=0.26). However, in their post-hoc secondary analysis of the VASST trial , Bhatraju et al[36] measured multiple biomarkers, including Angiopoietin 1 (Ang1), Angiopoietin 2 (Ang2), and interleukin 8 (IL-8) at study enrollment and determined their association with patient outcomes including response to vasoactive medication and development of AKI and 7-day non recovery of AKI. They compared these results from the VASST trials within 2 other cohorts of patients with shock (a cohort of septic shock patients admitted to Harborview Medical Center and a cohort from the Acute Respiratory Distress Syndrome (ARDS) genomic study.

Using the KIDGO AKI criteria in the Harborview and ARDS genomic trial cohorts (N=794 and 425 respectively) they used latent class analysis to identify two distinct phenotypes of sepsis associated AKI. They then sought to replicate these findings in 271 subjects with AKI from the VASST trial. They were able to validate these two phenotypes in the VASST data calling them AKI SP1 and AKI SP2.

There were significant differences across the presentation and outcomes of these to phenotypes. Subjects with SP2 were perhaps sicker with higher APACHE3 and SOFA scores on enrollment as well as increased risk of 7-day renal non-recovery. While subjects in SP1 experienced improved survival when treated with vasopressin with 90-day mortality of 27% compared to 46% in those not treated with norepinephrine (p=0.02), there was no vasopressin survival benefit seen in those with SP2.

While these clinical characteristics and outcomes differentiated these two phenotypes, the authors were also able to demonstrate these cohorts had different levels of biomarkers including angiopoietin 1 and 2 as well as interleukin-8 and soluble tumor necrosis factor receptor (sTNFR). SP2 was associated with higher ratio of Ang2/Ang1, 10 (7-16) vs SP1 2 (0.7-2.8) p <0.01. Additionally, SP2 had higher IL8, 106 pg/mL (49-345) vs SP1 23 pg/mL (13-47) p <0.01. The combination of these biomarkers was able to reliably identify these phenotypes in the discovery and validation cohorts with a c statistic using Ang2/Ang1 ratio and sTNFR was 0.98 (95% CI 0.97-0.99), for the discovery and 0.93 (95% CI 0.91-0.95) replication data set. Similarly the c statistic using Ang2/Ang1 ratio and IL 8 was 0.95,(95% CI 0.94-0.97) in the validation cohort and 0.92 (95% CI 0.89-0.94) in the validation.

The authors then followed up this study by investigating differences in single nucleotide polymorphisms (snps) within the Ang 1, Ang 2 and TNF receptor 1A genes. They demonstrated that a snp within the Ang2 gene were associated with a reduced risk of being in the SP2 phenotype and had lower Ang2 levels.[37] These plasma Ang-2 accounted for over 40% of the relative risk of developing the SP2 phenotype, but these data correlated the Ang2 genotype with Ang2 levels and risk of sepsis associated AKI need to be validated in other cohorts. However, taken together, the VASST -angiopoietin data demonstrated that in the setting of sepsis/septic shock unique phenotypes can be identified via using genetic testing, biochemical biomarkers, and advanced learning techniques. Future investigations will need to demonstrate if proactive testing /identification of patients with SP1 and pre-emptive treatment with vasopressin (over other vasoactive medications) leads to less sepsis associated AKI and improved outcomes.

It deserves noting that while not specific to sepsis and perhaps beyond the scope of this review, biochemical biomarkers are being used to phenotype other forms of AKI. Recently Moledina and colleagues identified and validated CXCL9- an interferon gamma induced chemokine as a urinary biomarker of acute interstitial nephritis (AIN) .[38] Through a series of preclinical and human studies they demonstrated CXCL9 levels were higher in those patients with biopsy proven AIN and that tissue had higher miRNA expression of the chemokine when AIN was present. Thus, as the molecular signature of other forms of AKI are increasing discovered it will allow for improved diagnostics around sepsis associated AKI.

Electronic Biomarkers

While the data around biochemical biomarkers in the setting sepsis and septic shock continues to show promise, others have turned their attention to artificial intelligence and advanced learning techniques to improve AKI patient identification and risk stratification.[39-41] Artificial intelligence allows for evaluation of big data across many patients, charts, and hospitals to establish models for predicting sepsis associated AKI.

Using data from several randomized clinical trials, Seymour and colleagues retrospectively analyzed datasets to derive and validate 4 distinct phenotypes in the setting of sepsis. [42] These phenotypes, which they call α, β,γ and δ were derived in data from 20,189 sepsis encounters from 16,552 unique patients and then subsequently validated in 43,086 sepsis encounters in 31,160 unique patients. These analyses included data from the aforementioned PROCESS trial but also from the PROWESS (Activated Protein C Worldwide Evaluation in Severe Sepsis)[43], ACCESS (A Controlled Comparison of Eritoran in Severe Sepsis)[44] and Genetic and Inflammatory Markers of Sepsis (GenIMS) [45] sepsis cohorts. Utilizing data from the these trials and the electronic health records they were able to demonstrate that there were unique patterns in the need for vaso-active medications, critical illness scores (SOFA) and degree/number of organ dysfunction across these phenotypes. [42]

Using biomarkers such as TIMP2*IGFBP7 and kidney injury molecule-1 they demonstrated different rates of AKI across these phenotypes with AKI being more common in the δ cohorts. This may have contributed to the δ phenotype having the higher rates of inpatient mortality (32%) compared to 10% across the entire cohort. [42] There is a wealth of data in these combined analysis as they had data on over 25 biochemical biomarkers and were able to look at several other outcomes besides renal injury/AKI: including inflammation, endothelial dysfunction and abnormal coagulation. While the identification these clinical phenotypes still needs to be operationalized clinically, this work hold promise to identify subsets of sepsis patients who may derive benefit from therapies directed at their specific sets of organ dysfunction.

Others have attempted to use big data sets to improve the risk stratification of patients at risk for sepsis associated AKI, this includes Nadkarni and colleagues who used data from the Medical Information Mart for Intensive Care (MIMIC) III database[40]. Using this retrospective data-warehouse of ICU patients from Boston, MA, USA, they identified 4,001 patients with sepsis-associated-AKI within the first 48 hours of ICU admission. They then used deep learning to identify 3 unique clinical sub-phenotypes of sepsis associated AKI, using vital signs and laboratory measurements from the first 48 hours along with co-morbidities (188 variables / features). As with the aforementioned phenotypes, they demonstrated significant differences in hepatic failure, heart disease and other organ dysfunction across these 3 phenotypes which also correlated with different outcomes, length of stay, need for dialysis and inpatient mortality.[40] Additionally, they performed manual chart review for 30 cases from each of the 3 phenotypes in order to estimate the predominant cause of AKI. There were significantly different number of acute tubular necrosis (ATN) and pre-renal azotemia cases across these 3 groups, with ATN predominating in the sicker phenotype (Phenotype 3).

Others have used advanced learning to look at the subset of patients with post-operative sepsis. Bihorac and colleagues developed a cohort of 243 surgical sepsis patients and utilized hierarchical clustering analysis combining clinical variables and 42 blood and urine biomarkers to identify phenotypic subgroups of sepsis. [46] They used a derivation cohort to describe 2 patient clusters of early organ dysfunction and a separate one for recovery. There were differences in several Inflammatory, renal, and endothelial biomarkers across these 2 clusters, including Interleukin-8, Tumor Necrosis Factor-alpha, cystatin c, serum creatinine, fluid overload, lactic acid, angiopoeitin 2.[46]

Not surprisingly the organ dysfunction cohort associated with higher APACHE scores, SOFA scores, higher chronic disease burden, and more septic shock. Additionally, this cluster was associated with higher 14- and 365-day mortality. [46] While this investigation was not specific to sepsis associated AKI, it represents another use of “big data” to phenotype sepsis patients to improve patient risk stratification and it requires large scale, independent validation.

Finally, while not yet specific to sepsis associated AKI, there are a several groups who have attempted to use artificial intelligence and machine learning techniques to predict the development of AKI[47, 39, 41]. In a prospective observational study of 5 ICUs, Flechet and colleagues developed and validated the AKIPredictor risk score which takes patient demographics, labs and vitals over the first 48 hours of ICU admission determines their risk of subsequently develop Stage 2 or 3 AKI.[39] Their model was dynamic and could be varied depending on whether you were evaluating a patient at the time of ICU admission or on Day 1 or 2. Using a cohort of 252 patients, 30 of whom developed Stage 2 or 3 AKI, their model provided an AUC of 0.80(0.69-0.92) for Stage 2 or 3 AKI based only on information at the time of ICU admission, while that increased to 0.95(0.89-1.0) after 24 hours of ICU admission. Importantly, only 10 percent of this population had sepsis, but it is not hard to imagine in the near future larger cohorts will be used to validate a similarly dynamic machine learning derived score specifically for those with sepsis associated AKI. [39] Table 3 summarizes the published literature around biomarkers and big data and their role in identification of patients at risk for sepsis associated AKI.

Table 3.

Novel Biomarkers for the early identification of patients at risk for sepsis associated AKI

Biomarker	Description	Investigations	Results
TIMP2*IGFBP7	Cell cycle arrest marker, measured in urine	Sapphire/Topaz (N=249 patients that had sepsis at ICU admission [26] ProCESS (n=999) -a post-hoc analyses of a RCT around a 6 hour early goal directed therapy trial[9] [20]	-Prediction of severe AKI in patients admitted to ICU with elevation of TIMP2*IGFBP7 AUC of 0.84 vs creatinine on admission AUC 0.73 [26] -Prediction model using TIMP2*IGFBP7 was compared to model without biomarker, and was able to predict severe AKI with an AUC improvement to 0.94 from 0.86 [9] -Lower survival at 30 days with RR of 2.20( 95% CI, 1.02-4.72) in stage 1 AKI, stage 2 (RR, 1.53; 95% CI, 1.04-2.27), and stage 3 (RR, 1.61; 95% CI, 1.00-2.60) with elevation of TIMP2*IGFBP7 in septic patients on admission to ICU [20] -Elevation of TIMP2*IGFBP7 before and after resuscitation associated with MAKE, please review table 2 for details [20]
Proenkephalin (PENKID)	Endogenous opioid protein; serves as a filtration marker	Kid-SSS FROG ICU AdrenOSS -Patients admitted to ICU on vasoactive medications from septic shock with or without AKI [11] [34] [6]	-Elevated level of PENKID (pmol/L) 179 vs 65 associated with MAKE without traditional AKI[8] -Elevated level associated with persistent AKI development [8] -elevated level in setting of sub-clinical AKI associated with mortality in ICU[11] -In absence of AKI, elevated PENKID associated with 2.5x increase in mortality vs those patients without elevated PENKID and no AKI[30]
Angiopoietin	Endothelial growth factor	- post-hoc secondary analysis of the VASST trial , in which multiple biomarkers and AKI related genotypes were explored [36, 17]	Elevated ratio of Ang2/Ang1 associated with sub-phenotype of AKI SP2 with high SOFA scores, APACHE scores, and increased risk of 7 day renal non-recovery [17] SP2-AKI associated with no improvement in mortality with vasopressin use, where-as SP1-AKI did see improvement in mortality [36] Identification of SP2-AKI using Ang2/Ang1 AUC: 0.93 which was more responsive to vasopressin therapy [36]
AI Big Data Analysis AKIPredictor (electronic)	Big Data analysis of large cohorts of sepsis patients Electronic model for prediction of AKI stage 2/3 over first 48 hours of ICU admission using demographics, labs, and vitals	Post-Hoc analyses of several large scale cohorts including PROCESS, [63]PROWESS [43], ACCESS [44] (GenIMS) [45] sepsis cohorts	-α, β,γ and δ AKI phenotypes were derived in data from 20,189 sepsis encounters from 16,552 unique patients and then subsequently validated in 43,086 sepsis encounters in 31,160 unique patients. [40] - Using TIMP2*IGFBP7 and kidney injury molecule-1 they demonstrated different rates of AKI across these phenotypes - AKI being more common in the δ phenotype which had higher rates of inpatient mortality (32%) compared to 10% across the entire cohort [40]

TIMP2*IGFBP7: Urinary tissue inhibitor of metalloproteinase 2 and insulin like growth factor binding protein 7

PENKID: Proenkephalin

Ang2/Ang1: Ratio of angiopoietin 2 : angiopoietin 1, specifically referring to the Angiopoietin section of the review. This ratio was found to be elevated in a subset of patients labelled SP2 which did not respond to vasopressin in septic shock.

MAKE: composite of all-cause mortality, receipt of RRT, and persistent AKI at day 7 (elevated serum creatinine level from baseline by >1.5 fold or >0.3 mg/dl at day 7).

Future Directions

Many nephrologists and other physicians may think that there is no benefit to the earlier detection of sepsis associated AKI (and AKI in general). However multiple studies across other settings of AKI have shown that earlier detection paired with kidney focused care can improve patient outcomes (e.g. development of severe AKI, shorter lengths of stay and lower inpatient mortality). [48-51] Admittedly not all studies have shown clinical benefits, however these studies have either been performed after established KDIGO AKI is present or not paired the later with a specific kidney focused intervention.[52, 53] Earlier detection, whether via risk scores or biomarkers, will allow for the further sub-phenotyping of sepsis associated AKI. Additionally, earlier diagnosis will provide an opportunity to optimize clinical AKI care, with several studies showing that in the setting of established AKI, many patients do not receive guideline based care.[54-56] These tools may provide the opportunity to identify patient who will likely benefit from the thoughtful implementation of labor intensive and more costly, kidney focused care. Future investigations should help clarify which sepsis populations (e.g. community acquired versus post-operative, or by anatomic site or by specific organisms (bacterial versus fungal) may derive a benefit from an early AKI care bundle. These investigations should not only seek to determine which segments of the bundle are most clinically impactful but also determine the relative cost-effectiveness.

Additionally, the use of biomarkers will allow for prognostic and predictive enrichment in future clinical trials. These tools can help ensure the correct patients will be enrolled in future prophylactic and therapeutic trials. Having confidence that a study is recruiting patients at higher risk for a specific outcome (e.g. need for RRT or inpatient mortality) may allow for a smaller study size depending on the desired endpoint and the potency of the intervention. It is our sincere expectation that understanding phenotypes, and the underlying biological differences in the endotypes of sepsis associated AKI will drive new therapy options and change AKI care moving forward.

It is important to acknowledge that much of the sub-phenotyping data has come from retrospective cohort studies. Future studies should seek to validate some of these findings in prospective cohorts to further inform future trials. Moving forward AKI trials should definitely include the concept sub-clinical AKI, using several of the tubular damage markers. Comparisons across those with and without biomarker elevations should continue to be investigated and be prespecified and powered for in all future therapeutic trials in the setting of sepsis associated AKI. Finally, endotyping patients (in AKI and elsewhere) is an emerging process field, with the example provided here of the angiopoietin cohort trial of Ang2/Ang1 ratio. While future trials should seek to determine if treating patients based on this ratio and/or their genotype will prospectively change outcomes it also sets the standard of utilizing pre-existing data and advanced learning techniques to detect differences in cohorts with sepsis and other forms of critical illness associated AKI[57, 17]

Future trials in sepsis associated AKI should be focused on the establishment of clinically meaningful, patient centered and reproducible end points. As above this should include the validation of retrospectively generated data. While there will be inherent differences in trial design and enrollment criteria based on whether a trial is prophylactic / preventive or therapeutic [58] . Biomarkers that can be used for predictive enrichment of a trial for dialysis requiring AKI may be different than those used for prognostic enrichment for a therapeutic intended to treat ischemic injury in the setting of sepsis associated AKI.

Interventional trials of sepsis management in critically ill patients will need to account for AKI based on the KIDGO definition as well as consider secondary endpoints based on changes in the novel biomarkers / sub-clinical AKI. If trials do not account, the potential confounding they may miss the opportunity to accurately determine outcomes in the setting of kidney injury. We stand at the dawn of a new era of clinical trials in the setting of AKI and more specifically sepsis associated AKI, biomarkers of AKI should be used to refine enrollment criteria as well as trigger future interventions. They must be used within their limitations, just as serum creatinine and urine output are imperfect, so are these novel tools. They need to be used in the correct population at the correct time for the correct endpoints/ outcomes. There is not a one-size fits all approach for these biomarkers. It is more than likely that different tools will be needed for different trial designs and perhaps in the future as displayed in Figure 1, prognostic enrichment tools will identify a cohort of ICU patients at high risk for dialysis requiring AKI in the setting of inflammatory injury from sepsis, and a separate group of predictive tools will determine which patients actually have inflammatory injury compared to ischemic injury.

Key messages:

Predictive and prognostic enrichment using both traditional biomarkers and novel biomarkers in the setting of sepsis can identify subsets of patients with either similar outcomes or similar pathophysiology respectively.

Novel biomarkers can identify kidney injury in patients without consensus definition AKI (e.g. changes in creatinine or urine output); and can predict other adverse outcomes (e.g. severe consensus definition AKI, inpatient mortality).

Finally, emerging artificial intelligence and machine learning derived risk models are able to predict sepsis associated AKI in critically ill patients using advanced learning techniques and several laboratory and vital sign measurements.