Predictive value of plasma proenkephalin and neutrophil gelatinase-associated lipocalin in acute kidney injury and mortality in cardiogenic shock

The CardShock study (NCT01374867 at www.​clinicaltrials.​gov) was a European prospective, observational, multicenter and multinational study on cardiogenic shock. Patients were recruited between October 2010 and December 2012 from emergency departments, cardiac and intensive care units, as well as catheterization laboratories from nine tertiary hospitals in eight countries. For a detailed design and description of the study population please see Harjola et al. [18].

Consecutive patients older than 18 years were enrolled in the study within 6 h from identification of cardiogenic shock. Written informed consent was obtained from the patient or next of kin if the patient was unable to give consent. To be included in the study, the patients needed to have (1) an acute cardiac cause for the shock, (2) systolic blood pressure < 90 mmHg (after adequate fluid challenge) for 30 min or a need for vasopressor therapy to maintain systolic blood pressure > 90 mmHg, and (3) signs of hypoperfusion (one or several of the following: altered mental status, cold periphery, oliguria < 0.5 mL/kg/h for the previous 6 h, or blood lactate > 2 mmol/L). Exclusion criteria were shock caused by ongoing hemodynamically significant arrhythmia or shock after cardiac or non-cardiac surgery. Baseline characteristics and previous medical history were recorded. Biochemical and clinical findings as well as hemodynamic parameters were documented at detection of shock and at pre-specified time points up to 48 h after inclusion. Patients were treated according to local practice and treatment and procedures were registered. Local investigators were responsible for determining the etiology of cardiogenic shock. Acute coronary syndrome etiology was defined as shock caused by myocardial infarction (with or without ST-elevation). Echocardiography was performed per protocol at study inclusion. Coronary angiography was carried out in 128 patients, of which 95 (74%) were done before study inclusion, and 24 (19%) within 12 h after study inclusion.

Plasma samples were collected at baseline in 178/219 patients enrolled in the CardShock study. Additionally, serial blood samples were collected at 12 h, 24 h, 48 h (all ± 3 h) and at discharge from ICU (5–10 days) (see Additional file 1: Fig. 1 for a diagram). A baseline sample and at least one additional sample within 48 h were available for 154 patients, which constitute the study population in this report. After collection, the blood sample was centrifuged, and separated plasma was immediately frozen in aliquots and stored at − 80 °C. PENK concentrations in plasma were determined using the Sphingotest^® penKid immunoassay (SphingoTec GmbH, Hennigsdorf, Germany). Plasma NGAL concentrations were determined with a commercially available particle-enhanced turbidimetric immunoassay (BioPorto Diagnostics A/S, Hellerup, Denmark). Creatinine, C-reactive protein (CRP), high-sensitivity troponin T, N-terminal pro-B-type natriuretic peptide (NT-proBNP), alanine aminotransferase, alkaline phosphatase, total bilirubin and cystatin C were analyzed from the plasma samples using commercially available standard kits (Abbott Laboratories, Abbott Park, IL, USA for cystatin C, Roche Diagnostics, Basel, Switzerland for all other tests) at a central accredited laboratory (ISLAB, Kuopio, Finland). Arterial blood gas analysis (including arterial pH and lactate), hemoglobin and leucocytes were analyzed by local laboratories. Estimated glomerular filtration rate (eGFR) was calculated from creatinine values using the Chronic Kidney Disease Epidemiology Collaboration equation [19]. AKI was defined and staged according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria [4]. For AKI staging at baseline, a recently described staging which includes biomarker levels was used [5]. Urine output was recorded at 6, 12, 18, and 24 h and used for urine output-based definitions of AKI. Main outcomes investigated in this study were AKI defined by an increase creatinine of more than 26,5 μmol/L within 48 h of study inclusion (AKI_crea48h) and all-cause 90-day mortality. Assessment of AKI_crea48h was based on creatinine levels from baseline until 48 h, and the highest increase within this time was used for staging. AKI staging by urine output was categorized according to the lowest urine output for a time interval within the first 24 h. Subclinical AKI was defined as positive biomarker (either P-PENK or P-NGAL > cut-off) but no AKI_crea48h. Vital status of the patients during follow-up was determined through direct contact with the patient or next of kin, or through population and hospital registers. Patients with missing plasma samples were left out of the analysis. Two patients were lost to follow-up and were left out of the survival analyses. The study was approved by local ethics committees and conducted in accordance with the Declaration of Helsinki. Two published risk scores for cardiogenic shock, CardShock risk Score [18] and IABP SHOCK II score [20] were calculated for every patient to assess whether P-PENK or P-NGAL provided additive value in risk prediction.

Results are presented as number (n) and percentage (%), mean and standard deviation (SD), or median and interquartile range (IQR) as appropriate. Group comparisons were performed using Fisher’s exact test for categorical variables, and Student’s t test or Mann–Whitney U test for continuous variables, as appropriate. Associations between continuous variables were assessed using Spearman correlations. Youden index was used to select the cut-offs of P-PENK and P-NGAL used for AKI and mortality prediction. To determine variables independently associated with P-PENK and P-NGAL levels univariable general linear models were constructed using log-normalized P-PENK and P-NGAL levels as dependent variables. Differences in survival between groups were assessed comparing Kaplan–Meier survival curves using log-rank test. Cox regression models were used to assess associations of variables with AKI and mortality in uni- and multivariable models. Forward and backward selection of variables was used in multivariable Cox regression models to calculate likelihood ratios, using a significance of > 0.10 for elimination and a significance of < 0.05 for retention. The additive value of a variable in the multivariable Cox regression models was assessed using likelihood ratio test for nested models. The proportionality of hazards assumption was tested using log-minus-log plots. Hazard ratios (HR) for Cox regression analyses are shown with 95% confidence intervals (CIs). Discriminative capability of P-PENK and P-NGAL at different time points was assessed using the area under receiver-operating characteristics curve (AUC). The statistical significance of changes between different time points in P-PENK and P-NGAL levels was assessed using Wilcoxon signed ranks test. Differences in P-PENK, P-NGAL, and creatinine trajectories between different groups were assessed using linear mixed modeling. Two-sided p values < 0.05 were regarded as statistically significant. Data were analyzed using the SPSS statistical package, version 23 (IBM Corp, Armonk, NY, USA). STATA (StataCorp. 2019. Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC) was used to calculate c-statistics for the different Cox regression models.

Baseline characteristics of patients (n = 154) stratified by P-PENK and P-NGAL quartiles at baseline are presented in Table 1. The mean age of all patients in the cohort was 66 years and 26% of the patients were women. The main cause of cardiogenic shock was acute coronary syndrome, which accounted for 81% of the cases. The overall 90-day mortality was 38%. AKI was observed in 47/154 (31%) patients within 48 h of baseline. Most patients (n = 29) had AKI already within 12 h from baseline. Median baseline P-PENK (n = 152) was 105 (IQR 71–167) pmol/mL, and median baseline P-NGAL (n = 146) was 138 ng/mL (IQR 84–214). Patients with P-PENK > median and patients with P-NGAL > median levels had lower eGFR and hemoglobin at baseline, and had higher levels of creatinine, NT-proBNP, alkaline phosphatase and lactate. Oliguria prior to enrollment was more frequent in patients with biomarker levels > median and they had higher CardShock and IABP II Shock risk score points. Patients with PENK/NGAL > median had a higher incidence of AKI within 48 h and their mortality was higher. STARD flow diagrams of study participants are provided as Additional file 1: Fig. 2.

Figure 1 shows the trajectories of P-PENK, P-NGAL and creatinine between baseline and 48 h in patients with and without AKI_crea48h and Fig. 2 shows P-PENK and P-NGAL trajectories in survivors and nonsurvivors. In nonsurvivors, the increase from baseline to 24 h was statistically significant for P-NGAL (p = 0.003), but not for P-PENK (p = 0.58). There were statistically significant differences in both P-PENK and P-NGAL levels with respect to time and group, and an interaction between the groups with time was observed.

AUCs were calculated at each time point for AKI_crea48h and 90-day all-cause mortality to assess the discriminatory capabilities of both biomarkers (Additional file 1: Table 1). At baseline, the predictive value of P-PENK_0h and P-NGAL_0h for AKI_crea48h was only moderate (AUCs 0.621 and 0.664, respectively) (Additional file 1: Table 1). Compared to baseline measurements, biomarker values at later time points had better performance in prediction of AKI_crea48h. However, since most AKI cases occurred within 12 h from baseline, P-PENK and P-NGAL measurements in time points later than baseline were not considered for AKI prediction. The optimal biomarker cut-off value at baseline for predicting AKI_crea48h was found to be 84.8 pmol/mL for P-PENK_0h and 104 ng/mL for P-NGAL_0h. In univariable Cox regression analysis, both P-PENK_0h > 84.8 pmol/mL and P-NGAL_0h > 104 ng/mL were associated with the development of AKI (Table 2). Multivariable regression modeling identified high P-PENK_0h (adj HR 2.2 [95% CI 1.1–4.4] and P-NGAL_0h (adj HR 2.8 [95% CI 1.2–6.5]) as being independently associated with AKI_crea48h (Table 2). Table 3 shows renal outcomes and differences in interventions using these P-PENK_0h and P-NGAL_0h cut-offs. There were 51 (34%) patients, who fulfilled the criteria for AKI stage 1S. In patients with oliguria at baseline but without high PENK or NGAL (stage 1A) only 2/16 (13%) of patients developed AKIcrea48h, compared with 13/51 (26%) of patients in stage 1S and 33/63 (48%) in stage 1B (p = 0.001). Only 1/21 (5%) of patients in stage 0 developed AKIcrea48h. High P-PENK_0h was associated with very early AKI (< 24 h), whereas high P-NGAL_0h was also associated with AKI detected at 24–48 h (p < 0.05 for comparisons between subsets). Both high P-PENK_0h and high P-NGAL_0h were associated with AKI as defined by low urine output according to the KDIGO criteria, whereas only high P-NGAL_0h was associated with AKI as defined by an increase in Cystatin C. High P-NGAL_0h was also associated with the use of renal replacement therapy, whereas high P-PENK_0h was not. Figure 3 shows Kaplan–Meier curves for the development of AKI_crea48h for high P-PENK and high P-NGAL groups. Cross tabulation of high P-PENK and P-NGAL levels with respect to AKI_crea48h can be seen in Additional file 1: Table 3.

The relationships between AKI, 90-day mortality and high P-PENK_0h and/or P-NGAL_0h are depicted as a Venn diagram in Fig. 4. 90-day mortality differed significantly between AKI stages at admission (7% for stage 0, 19% for stage 1S, 5% for stage 1A and 68% for stage 1B, p < 0.001). Both P-PENK_0h > 84.8 pmol/mL and P-NGAL_0h > 104 ng/mL were associated with higher 90-day mortality (Table 3). Figure 5 shows the survival curves for patients with and without AKI_crea48h separated by baseline P-PENK (panel A) and baseline P-NGAL (panel B) higher (subclinical AKI) or lower than the optimal cut-off for AKI prediction (84.8 pmol/mL for P-PENK_0h and 104 ng/mL for P-NGAL_0h). Using the cut-offs predictive of AKI, P-NGAL_0h > 104 ng/mL was able to further stratify patients with and without AKI_crea48h further into high or low mortality risk groups (Fig. 5).

Although high levels of both P-PENK and P-NGAL were associated with higher mortality at all time points examined, the AUC for mortality was highest for both biomarkers measured at 24 h (Additional file 1: Table 2). The AUCs for 90-day mortality for CardShock risk score and IABP SHOCK II risk score at 24 h were 0.778 and 0.707.

The optimal cut-off for mortality was 105.7 pmol/mL for P-PENK_24h, and 151 ng/mL for P-NGAL_24h. Survival curves using these cut-offs showed that for patients with P-PENK_24h > 105.7 pmol/mL 90-day mortality was 68.2% compared with 17.4% for patients with P-PENK_24h < 105.7 pmol/mL. Similarly, 90-day mortality for patients with P-NGAL_24h > 151 ng/mL was 63.5% compared with 17.7% for patients with P-NGAL_24h < 151 ng/mL (p < 0.001 for all, Additional file 1: Fig. 2).

In univariable Cox regression, both P-PENK_24h > 105.7 pmol/mL and P-NGAL_24h > 151 ng/mL were strongly associated with mortality (Table 4). In multivariable analysis, the associations of both P-PENK_24h > 105.7 pmol/mL and P-NGAL_24h > 151 ng/mL with higher mortality were found to be independent of both CardShock risk score and IABP II Shock score. Further adjustment of the multivariable models with AKI_crea48h did not change the independent association of P-PENK_24h with 90-day all-cause mortality, whereas P-NGAL_24h was no longer significantly associated with 90-day mortality after adjusting for both CardShock risk score and AKI_crea48h. Cross tabulation of P-PENK_24h and P-NGAL_24h with respect to 90-day mortality can be seen in Additional file 1: Table 3.

Results from a Spearman correlation analysis with baseline variables and levels of P-PENK and P-NGAL are shown in Additional file 1: Table 2. Markers of renal function had moderate to strong correlations with both P-PENK and P-NGAL at baseline. P-PENK and P-NGAL were also intercorrelated. General linear model analysis showed that the biomarkers had strongest associations with eGFR (F-statistic 82.8 for P-PENK and 84.9 for P-NGAL; p < 0.001 for both) and log-normalized lactate levels at baseline (F-statistic 5.0; p = 0.03 for P-PENK and 16.4; p < 0.001 for P-NGAL). In addition, body mass index (F-statistic 6.0, p = 0.02) and alkaline phosphatase (F-statistic 5.0, p = 0.03 were independent associates of P-NGAL levels. eGFR alone accounted for 36% of the observed variance of P-PENK and 39% of P-NGAL variance.

Elevated levels of P-PENK and P-NGAL were both associated with the development of AKI_crea48h and had comparable discriminatory properties in AUC analysis. Interestingly, we found that AKI_crea48h occurred almost solely if oliguria (< 0.5 mL/kg/h for > 6 h) before study inclusion was combined with subclinical AKI at baseline. P-NGAL has been shown to predict AKI after cardiac surgery [21], in critically ill children [22] and adults [23] with suggested cut-offs ranging from 100 to 270 nmol/mL [24]. In a recent meta-analysis comparing plasma/serum NGAL, urine NGAL and serum cystatin C, plasma/serum NGAL was found to be the earliest marker predicting contrast-induced nephropathy [25]. Of the two markers investigated in the present study, only P-NGAL was associated with renal replacement therapy and AKI assessed by changes in cystatin C plasma concentrations, suggesting P-NGAL may be a more robust indicator of AKI and other renal outcomes. For most patients who developed AKI, it occurred very early, which makes it most likely related to the acute state of cardiogenic shock and not secondary to other events in-hospital. Thus, based on our findings, the ability to predict the occurrence and severity of AKI using biomarkers measured at presentation opens a narrow window of opportunity for targeted interventions.

Overall, levels of P-PENK and P-NGAL decreased from 0 to 48 h. In nonsurvivors and patients who developed AKI within 48 h the levels were higher at all time points for both P-PENK and P-NGAL compared to survivors and patients without AKI. While the absolute levels differed, the trajectories of P-PENK over time were similar in survivors and nonsurvivors. In contrast, for P-NGAL, the levels were increasing up to 48 h in nonsurvivors, but were stable or declined in survivors. These findings most likely represent differences in the production and kinetics of the two biomarkers. In a mouse renal ischemia–reperfusion injury model, P-NGAL expression has been shown to be upregulated in the proximal kidney tubular cells early in response to renal ischemia with a peak at 12 h post-ischemia [26], which could explain the initial increase observed for P-NGAL in nonsurvivors and patients who developed AKI and also had higher levels of lactate. P-PENK, on the other hand, is a marker of the endogenous opioid system activity, which could be higher in those with more severe clinical presentation, but be more stable over time. P-PENK could also cause a deterioration of renal function through cardiodepression [27], with high initial levels causing a more marked decline in kidney function. These pathophysiological mechanisms may be seen to be in concordance with the results of our study that patients with high levels of either P-PENK or P-NGAL were at significantly higher risk for mortality and for developing AKI within 48 h.

Both P-PENK and P-NGAL showed good discriminatory capabilities for 90-day mortality at all time points up to 48 h, with the highest AUCs for both occurring at 24 h. High levels of both P-PENK_24h and P-NGAL_24h were associated independently with 90-day mortality even after adjusting for the CardShock risk score, IABP II SHOCK risk score and AKI. In fact, the AUCs for both P-PENK_24h and P-NGAL_24h were similar to those of the risk scores alone. The independent association of P-PENK_24h and P-NGAL_24h with 90-day mortality suggests that they are not just markers for AKI, but may also contribute to other unidentified causes of increased mortality in the CardShock study population that are not included in the relevant risk scores. A recent study in critically ill patients suggests that high levels of P-PENK in patients without AKI may be a marker of subclinical AKI, which was associated with a risk of death close to patients with AKI [28]. Similarly in our study, patients with subclinical AKI (elevated biomarkers at baseline but without AKI_crea48h) had higher 90-day mortality. Interestingly, P-NGAL was also able to stratify patients with AKI_crea48h into high or low mortality risk groups. It could be hypothesized that this might be due to less severe AKI, where despite an increase in creatinine the renal injury is smaller. Using the recently described AKI staging using biomarkers on admission [5], we were also able to show that low urine output (< 0.5 mL/kg/h for > 6 h) before study inclusion was associated with worse outcomes only if combined with high baseline levels of P-PENK or P-NGAL.

Elevated P-NGAL may not be specific for acute kidney injury, as NGAL is also expressed at low levels in several human tissues, including lung, stomach, and colon, as well as neutrophils [14]. Serum NGAL levels have been shown to increase also with infection, inflammation and ischemia [29]. The association of NGAL with inflammation and ischemia might be a drawback in acute kidney injury prediction, but might help explain the high discriminatory capability in 90-day mortality in cardiogenic shock, where systemic inflammatory response syndrome and end-organ ischemia play an important role. P-PENK and P-NGAL could be useful in assessing mortality risk of cardiogenic shock patients at later time points for which few risk markers have so far been assessed [30]. However, although this study offers new insight to prediction of AKI and mortality in cardiogenic shock populations, overall data on biomarkers in this area are still very limited and more research is warranted. The results of our study should be validated in another cohort of cardiogenic shock patients to confirm our findings.

We did not have plasma samples available for all the CardShock study participants, and also were unable to measure P-NGAL for some of the patients in this study. However, this is one of the largest cohorts of biomarker studies in patients with cardiogenic shock and despite the general challenges with serial sampling in acute cardiac care, we consider the available samples to be representative of the cohort. One of the study limitations is the lack of data on creatinine prior to study inclusion. On the other hand, the European Renal Best Practice position statement recommends using baseline creatinine values instead of historical values or back-calculated value based on an assumed GFR of 75 mL/min/1.73 m² [31]. Although adjustments were made for several variables in multivariable analyses, there may have been other confounding factors we were unable to account for leading to an overestimation of the independent association of P-PENK and P-NGAL with AKI and mortality.