Biomarkers in acute kidney injury settings to predict interventions and outcomes: the MARKISIO study

Abstract

Introduction

Predicting the need for renal replacement therapy (RRT) in acute kidney injury (AKI) remains challenging. The utility of biomarkers was explored during previous studies which were biased as RRT indications relied on clinician opinion rather than evidence. Those studies preceded trials that clarified RRT initiation criteria. We aimed to assess biomarkers in predicting criteria for RRT initiation in severe AKI patients.

Material and methods

This is an ancillary study of the AKIKI2 trial. Patients with severe AKI (stage 3) receiving invasive mechanical ventilation and/or vasopressors were included. Blood and urine samples were collected within 12 h after the occurrence of severe AKI when feasible, depending on the availability of trained research staff and appropriate sample storage infrastructure. The primary endpoint was the onset of precise criteria for RRT initiation within 72 h after severe AKI. We analyzed routine serum biomarkers (pH, serum potassium, serum creatinine) and novel urinary and serum biomarkers (CCL14, KIM1, nicotinamide and its metabolites, cDPP3, plasma proenkephalin A 119-159).

Results

Among the 256 patients, 101 (39%) met at least one criterion for RRT initiation or died within 72 h. No biomarker demonstrated satisfactory predictive performance for the primary endpoint. No novel biomarker was significantly associated with the occurrence of MAKE₆₀. In multivariable analysis, ‘SAPSIII’ and ‘Serum potassium level at D0’ were significantly associated with the occurrence of MAKE₆₀.

Conclusion

Neither routine nor novel biomarkers demonstrated conclusive predictive accuracy for the need for RRT in severe AKI patients. Given evidence-based criteria for initiating RRT, the tested biomarkers may not effectively guide RRT initiation.

Introduction

Acute Kidney Injury (AKI) in critically ill patientsis associated with high morbidity and mortality, particularly when renal replacement therapy (RRT) is required [1].

Reliable tools for predicting the need for RRT in this patient population are currently lacking. Biomarkers were claimed to be valid for achieving this objective [2]. In a recent prospective observational study (RUBY), a panel of novel candidate biomarkers was tested to identify persistent AKI [3]. However, that study did not address the question of whether any of these biomarkers are useful in identifying patients who will require RRT. The vast majority of studies addressing this issue did not clearly define the indications for RRT [4]. This may be a major limitation as they actually assessed how biomarkers predict practitioner’s decision regarding RRT initiation, a decision that can vary significantly within a country, a hospital or even among medical teams [5]. Finally, these studies were conducted before the publication of four large randomized controlled trials (RCTs) (AKIKI, IDEAL-ICU, STARRT-AKI, and AKIKI2) that provided clarity on RRT indications in critically ill patients [6,7,8,9].

Given these developments, it is of utmost interest to assess the ability of biomarkers to predict the onset of RRT initiation criteria in patients with severe AKI based on these recent studies. Such predictions may help clinicians anticipate clinical deterioration, initiate timely monitoring or preparation for RRT, and ultimately support individualized decision-making or patient stratification in future interventional trials. To achieve this goal, we relied on a biobank prospectively collected during the observational stage of the AKIKI2 study.

Methods

Study design and endpoints

The MARKISIO study is an ancillary study of AKIKI2 [6], a multicenter randomized controlled trial conducted in 39 ICUs in France from May 2018 to December 2019 that compared two delayed RRT initiation strategies in critically ill patients: a standard delayed strategy and a more delayed strategy. Adult patients hospitalized in the ICU with severe AKI (defined as Kidney Disease: Improving Global Outcomes [KDIGO] stage 3) [10] who were receiving (or had received for this episode) invasive mechanical ventilation and/or vasopressors were first included in an observational stage (prior to the randomization stage). A standard delayed RRT initiation strategy, similar to that used in the AKIKI trial [7], was applied during the observational stage. With this strategy, RRT was withheld until occurrence of one of the following criteria: oliguria/anuria for more than 72 h or blood urea nitrogen level above 112 mg/dl (serum urea level of 40 mmol/L). Once such criteria were present, patients were then randomized to immediate RRT initiation (standard group) or to a more delayed strategy in which RRT was withheld until BUN level reached a value of 140 mg/dl, regardless of the duration of oliguria/anuria.

In investigating centers that agreed to participate in the biobanking, blood and urine samples were prospectively collected within 12 h after the diagnosis of stage 3 AKI (according to KDIGO criteria), which marked inclusion into the observational stage of the AKIKI2 trial.

The study protocol, including the biobank collection, was approved by the competent French legal authority (Comité de Protection des Personnes de Sud-Est V) for all participating centers. All analyses were performed in accordance with the guidelines of the International Conference on Harmonization and Good Clinical Practice.

The primary endpoint was the occurrence of a criteria for RRT initiation within the 72 h after severe AKI onset. Those criteria were: urgent indications (severe hyperkalemia, severe metabolic acidosis, severe pulmonary edema), or oliguria/anuria (urine output < 0,3 mL/kg/h or < 500 mL/day) for more than 72 h, or blood urea nitrogen level of 112 mg/dL or more (serum urea concentration of 40 mmol/L or more) (details are provided in a supplementary appendix, Table S1). To avoid overlooking competitive risk issues, patients who died within 72 h of enrollment were also considered endpoint positive.

The secondary endpoint was the composite of RRT dependency or absence of renal recovery or death 60 days after the inclusion defining major adverse kidney events at 60 days (MAKE₆₀). The absence of recovery was defined as ≥ 25% loss in estimated glomerular filtration rate (eGFR). The estimated GFR was calculated from the serum creatinine levels using the MDRD equation.

Sample and data collection

Blood and urine samples were collected at the time of inclusion in the observational stage and then centrifuged at the investigating center. The resultant plasma, serum, and urine supernatants were quickly frozen and stored at − 80 °C. Shipments from investigating centers to central biobank (located in Bichat hospital, Paris, France) were done every three months by a secure transportation company in insulated boxes cooled with dry ice. The samples were thawed immediately before analysis.

All patients were followed for up to 60 days after enrollment. The data collection included demographics, comorbidities, exposure to nephrotoxic agents, SAPS III score on ICU admission, clinical and laboratory data throughout the ICU stay, occurrence of RRT initiation criteria, renal function recovery, and survival 60 days after inclusion.

Biomarker testing

We focused on a set of urinary and serum biomarkers chosen for their promising nature and their relevance to distinct pathophysiological pathways implicated in AKI and renal recovery. We refer to them as “novel biomarkers.” We also collected more traditional biomarkers routinely monitored (pH, serum potassium and serum creatinine) and refer to them as “routine biomarkers.” These routine biomarkers were assessed locally as part of standard patient care.

Novel Biomarkers were measured blinded to the clinical data:

• C–C motif chemokine ligand 14 (CCL14) and Kidney Injury Molecule 1 (KIM1) were measured in urine samples at INSERM UMRS 1155 (Tenon Hospital, Paris, France) using commercial kits from ABCAM. For interpretation purposes, KIM1 levels were normalized to urine creatinine levels.

• Nicotinamide and its metabolites N-methyl-2-pyridone-5-carboxamide (N2PY) and Methylnicotinamide (MNM) in plasma samples were simultaneously quantified by Liquid Chromatography coupled to tandem Mass Spectrometry (LC–MS/MS) method in the pharmacology unit of Jean Verdier hospital (Bondy, France) (more details are provided in supplementary appendix).

• Circulating dipeptidyl peptidase (cDPP3) concentration was measured in ethylenediaminetetraacetic acid (EDTA) plasma using a luminescence immunoassay (4TEEN4 Pharmaceuticals GmbH, Hennigsdorf, Germany) [11]. Proenkephalin A 119-159 was measured in ethylenediaminetetraacetic acid (EDTA) plasma using Sphingotest® penKid® luminescence immunoassay (Sphingotec GmbH, Hennigsdorf, Germany) [12, 13]. cDPP3 and Proenkephalin A 119–159 (penKid) measurements were performed by 4TEEN4 Pharmaceuticals GmbH and Sphingotec GmbH, respectively, blinded to the clinical data.

Statistical analyses

Continuous variables were tested for normality using the Shapiro–Wilk test. A sample t-test was used for normally distributed data, whereas the Mann–Whitney U test was applied for non-normally distributed data. Categorical data were compared using the chi-squared test or Fisher’s exact test, depending on the sample size and distribution of the variables.

The predictive potential of the biomarkers for the primary endpoint was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). The bootstrap method was used to calculate the confidence intervals of the AUC. In an exploratory analysis, we evaluated the predictive potential of all biomarkers across three distinct subpopulations characterized by specific causes of AKI: sepsis, rhabdomyolysis, and nephrotoxicity (Fig. S2, Table S3).

A logistic regression model was performed to identify independent predictors of both the primary outcome and MAKE60. We applied backward elimination, starting from a full model that included all clinically plausible variables—either supported by the literature or associated with the outcome in univariate analysis. Variables were retained based on statistical significance and improvement of model fit, while ensuring clinical coherence. The log-linearity assumption was tested for all continuous variables, and variables were dichotomized if the assumption was violated. This approach aimed to prevent overfitting while identifying the most relevant predictors. Notably, fewer variables were retained in the MAKE60 model due to the strict selection threshold.

Missing data were handled using multiple imputation by chained equations (MICE). Twenty imputed datasets were generated, using predictive mean matching for continuous variables and logistic regression for binary variables. All variables used in the regression models were included in the imputation process, and pooled estimates were obtained using Rubin’s rules. All biomarkers were assessed as continuous variables. No dichotomization was performed post hoc, as we aimed to avoid arbitrary thresholding and preserve the integrity of the analysis. All statistical analyses were conducted using R 3.5.1 and Python 3.9. Specifically, the ROCR package in R and scikit-learn library in Python were utilized. Statistical significance was defined as a two-sided p value of < 0.05.

Results

Patients and outcomes

Among the 767 patients included in the observational stage of AKIKI2, 261 were enrolled in the MARKISIO study, of whom 256 (98%) were available for the analysis of the primary endpoint (Fig. 1). Among these patients, 113 (44%) met the primary endpoint, which included 101 patients (39%) had at least one RRT initiation criteria (urgent indication for 75 patients, oligo-anuria for more than 72 h or serum urea > 112 mg/dl for 26 patients) and 12 died within 72 h after inclusion. Patient characteristics according to “onset of RRT initiation criteria within 72 h” (the primary endpoint) are presented in Table 1. Features did not differ in age, coexisting conditions (chronic kidney disease, congestive heart failure, hypertension, and diabetes mellitus), and severity scores (SAPS III and SOFA). Acute Respiratory Distress Syndrom (ARDS) was more common in patients who required RRT within 72 h than in the rest of patients (40.7% vs 23.8% respectively, p < 0.01). The primary causes of AKI were shock and/or sepsis. At the end of follow-up (60 days after inclusion), 117 patients (45.7%) had died. Among the 139 survivors, 27 (10.5% of the overall population) did not fully recover renal function, defined as ≥ 25% reduction in eGFR compared to baseline. No patient was RRT-dependent at day 60. Thus, the MAKE60 composite outcome occurred in 144 patients (56.2%), mostly driven by mortality, followed by persistent renal dysfunction.

Fig. 1

Patient flow diagram. RRT: renal replacement therapy

Table 1 Baseline patients characteristics^a

Biomarker performances for the primary outcome

Neither routine biomarkers nor any novel biomarker demonstrated satisfactory performance for predicting the need for RRT initiation within 72 h after severe AKI (Fig. 2) in the whole patient population. The AUC for prediction of the need for RRT within 72 h ranged from 0.58 to 0.60 for routine biomarkers and from 0.51 to 0.65 for novel biomarkers. Urinary CCL14 demonstrated a modest predictive performance (AUC = 0.65; 95%CI = 0.58 to 0.71).

Fig. 2

Area under the ROC curve (AUC) for prediction of the need for RRT within 72 h. Routine biomarkers refer to pH, serum potassium concentration and serum creatinine concentration (A). Novel urinary biomarkers refer to CCL14 and KIM1 (B). Novel plasma biomarkers refer to DPP3 and PenKid (C). Nicotinamide and its metabolites are also considered novel biomarkers but are presented separately (D). RRT: renal replacement therapy

An exploratory analysis of the subpopulations according to the cause of AKI is presented in Fig. S2.

In multivariable analysis (logistic regression model using clinical and biological variables), the five variables significantly associated with the occurrence of the primary outcome, presented in order from the strongest to the weakest association, were: ‘pH min’, ‘ARDS’ ‘serum creatinine at D0’, ‘urinary CCL14’ and ‘DPP3’ (Table 2).

Table 2 Logistic regression for prediction of the need for RRT within 72 h

Biomarkers performances for the secondary outcome

In multivariable analysis (logistic regression model using clinical and biological variables), ‘SAPS III’ and ‘Serum potassium at D0’ were the only two variables significantly associated with the occurrence of MAKE₆₀ (Table 3). No novel biomarkers were found to be significantly associated with this outcome.

Table 3 Logistic regression for prediction of MAKE₆₀

Discussion

The MARKISIO study was designed to evaluate the ability of a panel of both routine and novel biomarkers to predict the need for RRT within 72 h following the onset of severe AKI. Among the 256 patients enrolled, 39% met at least one criterion for RRT initiation according to precise criteria or died within 72 h post-inclusion and 56.2% presented a major adverse kidney event 60 days after inclusion. No biomarker achieved an acceptable predictive performance for these outcomes.

Previous studies, which were gathered in a meta-analysis published in 2018, have reported the potential utility of various biomarkers in predicting the initiation of RRT in patients with AKI [4]. In contrast, our findings do not confirm these results. This discrepancy can be explained at least in part by the more precise definition of the criteria for initiation. Indeed, we took advantage of the results of RCTs that were not available at the time when the vast majority of studies on biomarkers were issued. These recent RCTs allowed the use of well-defined and reproducible RRT initiation criteria that are now considered by many as the standard of care [14, 15]. Among these criteria, the persistence of oliguria or anuria for more than 72 h deserves particular attention. This threshold was associated with increased mortality in the AKIKI2 trial and is therefore now considered a clinically relevant and evidence-based indication for initiating RRT [6]. Its inclusion reflects a move toward standardized, outcome-driven criteria rather than arbitrary thresholds. In contrast, previous studies assessed the utility of biomarkers to predict the physician’s decision to initiate RRT; decisions varied widely from one country to another, from one hospital to another, and sometimes within the same team [5]. This is supported by the findings of a systematic review and meta-analysis published in 2018 by Klein et al., which included more than 40 studies on biomarkers for predicting RRT in AKI [4]. The authors noted that most of these studies, conducted prior to the publication of recent large RCTs, failed to define standardized criteria for RRT initiation. Indeed, they relied primarily on clinician judgment which increased the risk of bias or confounding by indication. In addition, an international survey by Legrand et al., published in 2013, further illustrated the wide variability in clinical practices regarding RRT initiation in ICU, with decisions often based on subjective or inconsistent criteria such as serum creatinine thresholds or azotemia [5]. Historically, there has been a tendency to initiate RRT based on the perceived severity of a patient condition rather than on objective criteria (such as metabolic complications of AKI). This approach not only muddies the predictive power of biomarkers but also creates a circular logic.

The KDIGO guidelines [10] illustrate this paradox by defining stage 3 AKI as “initiation of RRT.” This creates a self-perpetuating cycle in which the severity of a patient condition (Stage 3 AKI) is partly defined by the decision to initiate RRT, which, in turn, has historically been influenced by the perceived severity of patient condition. This feeds back into itself, obscuring both the predictive utility of biomarkers and objective assessment of a patient’s need for RRT. This kind of aporia likely explains the consistent efficacy demonstrated by the majority of biomarkers in predicting the specific clinical endpoint of RRT initiation when this initiation is mainly based on clinician perception and not on precise and objective criteria. They were probably primarily assessing the overall severity which is relatively straightforward to evaluate.

The biomarkers chosen for this study were selected based on their potential relevance to the diverse pathophysiological mechanisms implicated in AKI and the potential necessity for RRT. Specifically, urinary KIM1 was included for its presumed role in both kidney injury and recovery. The process of kidney cellular regeneration after toxic or ischemic injury involves undifferentiated proximal tubular epithelial cells that disperse onto denuded areas of the basement membrane and reconstitute a continuous epithelial layer [16]. The KIM1 protein is expressed in these dedifferentiated cells and may play a role in their adhesion to the basement membrane and dispersion onto damaged areas [16, 17]. Nicotinamide and its metabolites, (N2PY and Methylnicotinamide) were examined for their involvement in metabolic pathways directly impacting renal function recovery. In the context of AKI, marked diminutions in nicotinamide adenine dinucleotide (NAD +) concentration, for which nicotinamide is a precursor have an adverse impact on energy synthesis, thereby undermining the primary renal function of selective solute filtration. In contrast, elevated NAD + levels may confer resilience to renal tubules when exposed to a range of acute stressors, making this mechanism one of the most promising directions for therapeutic intervention. Indeed, investigation of this pathway is highly relevant in this particular setting [18,19,20]. The marker cDPP3 was assessed for its role in the regulation role in Renin Angiotensin System (RAS) known for its implication in glomerular hemodynamics renal inflammation and fibrotic process [21,22,23]. PenKid was assessed for its previously described potential ability to detect early AKI and renal recovery [24, 25]. Lastly, we evaluated urinary CCL14, given that recent research has suggested its potential as a predictor of persistent AKI and, consequently, lack of renal recovery [3, 26].

In contrast to our study, RUBY [3] reported that urinary CCL14 might be a promising marker for persistent AKI and subsequent renal non-recovery. This was further supported by an external validation study by Bagshaw et al. [26]. In the RUBY study, the primary outcome was persistent stage 3 AKI within 72 h. This primary outcome was met in more than 50% of cases because RRT was initiated despite the absence of pre-specified RRT initiation criteria. This created a circular reasoning, i.e. the clinician to start RRT influenced the primary outcome y definition. A recent study by Meersch et al. [27] reported good performance of urinary CCL14 for predicting absolute indications for RRT in a postsurgical population. However, this single-center study combined biomarkers with the furosemide stress test, used different endpoints, and did not address death as a competing event. These pitfalls and the many differences in term of methodology, population, and endpoint definition may explain the discrepancies with our findings. Although urinary Neutrophil Gelatinase-Associated Lipocalin (NGAL) has been widely claimed a potential biomarker for AKI, we consciously chose not to include it in our panel for this study. This decision was supported by the disappointing performance of NGAL in previous studies, including the ELAIN study on the timing of RRT initiation, where this biomarker failed to discriminate between varying stages of AKI severity. Indeed, it disqualified only 0.5% of screened patients from randomization in the ELAIN trial [28]. Similarly, the STARRT-AKI study, the largest study in this field to date which initially considered NGAL, opted not to include it in the main study because of its very poor performance in its pilot phase [8, 29].

One of the principal strengths of our study lies in its methodological rigor, informed by the latest evidence from randomized controlled trials on RRT timing [7, 8] [6, 9]. This allowed us to apply recognized criteria for initiating RRT, thereby reducing the variability often observed in clinical practice. Additionally, the diversity of biomarkers assessed enhances the comprehensiveness of our findings. However, several limitations should be acknowledged. The most significant constraint is the relatively small sample size, which limits the statistical power and generalizability of our results. It is worth noting, however, that this limitation is not unique to our study; the RUBY trial and many other studies on biomarkers also operated under the same constraint of limited sample sizes. Furthermore, the cross-sectional nature of our biomarker measurements could be seen as a limitation, as it provides only a snapshot of a dynamic physiological process. This issue is also present in most other studies, including the RUBY study, due to the practical challenges of obtaining longitudinal biomarker data in a clinical setting. We also could not evaluate the NephroCheck biomarkers (Insulin-like Growth Factor Binding Protein 7 [IGFBP-7] and Tissue Inhibitor of Metalloproteinase 2 [TIMP-2]) or perform the NephroCheck test itself due to the unavailability of kits from the manufacturer, which may have affected the comprehensiveness of our biomarker assessment. Although alternative techniques such as ELISA could theoretically have been employed, our intent was to preserve methodological consistency and clinical applicability by relying on the clinically validated NephroCheck platform. Furthermore, recent evidence suggests that in patients already presenting with severe AKI, the incremental predictive value of [TIMP-2][IGFBP7] may be limited. For instance, in the recent study by Palmowski et al., this biomarker combination was only useful when applied in conjunction with functional testing (i.e., the furosemide stress test) for predictive enrichment in sepsis-associated AKI [30].

In summary, we found that no biomarker demonstrated significant predictive accuracy for the need for RRT within 72 h after severe AKI onset in ICU patients.

Conclusion

This ancillary study of the AKIKI2 trial found that within the context of precise RRT initiation criteria, biomarkers may not be suitable for guiding RRT initiation nor informing renal prognosis in patients with severe AKI potentially requiring RRT.