Urinary biomarkers predict advanced acute kidney injury after cardiovascular surgery

This study was conducted by the biomarker investigation group from the National Taiwan University Study Group on Acute Renal Failure (NSARF) (see Additional file 1), using a multicenter, prospectively constructed database of AKI [22‐26]. Patients undergoing cardiovascular surgery (including coronary bypass, valvular operations and aortic aneurysm repair) between August 2009 and December 2014, were enrolled prospectively from a tertiary center in northern Taiwan and two regional hospitals in central and northern Taiwan. Patients with the following conditions were excluded: those who were younger than 18 years of age, those who were diagnosed with AKI under the KDIGO definitions during index hospitalization before surgery, those who had undergone renal replacement therapy, those who had a history of nephrectomy or renal transplantation and estimated glomerular filtration rate (eGFR) < 30 mL/1.73 m² at the time of ICU enrollment. The enrollee was required to have a baseline serum creatinine measurement, defined as 7–180 days prior to the index hospital admission.

This study was approved by the research ethics review board of National Taiwan University Hospital (201105040RC) along with established written informed consent. This research was carried out in accordance with the approved guidelines. Written informed consent was obtained from all participants before inclusion. This study was conducted in accordance with the Declaration of Helsinki.

Medical records of study participants were prospectively reviewed to retrieve hospitalization data, including baseline demographic characteristics (Table 1), intervention procedures, and comorbidity status. Creatinine level with the accompanying eGFR (through the Modification of Diet in Renal Disease (MDRD) Study equation formula) was used for eGFR in this study. Among adults, the MDRD Study equation provides a clinically useful estimate of GFR in patients with stable kidney status [27]; urine output was recorded at each time point after surgery as detailed in the study protocol. Postoperative inotropic agent use was quantified as per inotropic equivalents (see Additional file 1) [23, 25, 28]. Surgery-related parameters included surgical methods, aortic clamping time and cardiopulmonary bypass time. Lengths of total admission and ICU admission were also recorded. Disease severity was evaluated as post-surgical SOFA score. The clinical classifications of Liano’s score [19] and Cleveland Clinic Foundation Acute Renal Failure Scoring System (CCF ARF score) [20] were also assessed to examine the risk of postoperative renal failure. Clinical and demographic characteristics of these patients were collected at each site on a case report form, which was sent to the coordinating center for entry into the NSARF database [29‐31].

Fresh urine samples were obtained in the ICU at 0, 3, 6, 12 and 24 h after completion of surgery. Other laboratory examinations were performed as indicated clinically. Urine specimens were collected by a standardized procedure, centrifuged within 1 h and the sediments discarded. The urine samples, collected in separate polypropylene tubes containing sodium azide, were stored at − 80 °C until required. Each specimen was centrifuged (800 g at 4 °C for 5 min) and the supernatant was collected for ELISA.

The urinary HJV, KIM-1 and NGAL levels were measured by a human hemojuvelin ELISA kit (USCN Life Science, Inc., Wuhan, China), human urinary TIM-1/KIM-1/HAVCR Quantikine ELISA kit (R&D Systems, USA) and a human lipocalin-2/NGAL ELISA kit (R&D Systems), respectively. All of the results were expressed in nanograms per milliliter. The lower limit of detection for HJV, KIM-1 and NGAL was 0.156, 0.046 and 0.2 ng/mL, respectively. The urinary α-GST and π-GST levels were determined using Human Alpha and Pi GST EIA Test Kits (EKF Diagnostics). The results were expressed in micrograms per liter. The inter-assay and intra-assay coefficient of variation for α-GST was 6.3% and 2.7%, respectively, and for π-GST, it was 8.6% and 3.1%, respectively. The lower limit of detection for α-GST and for π-GST was 0.3 μg/L. Assays were completed as described by the manufacturer’s protocol, and each measurement was performed in duplicate. Urinary creatinine levels were measured using the Jaffe assay, with standardization to isotope dilution mass spectrometry (IDMS)-traceable reference. Technicians performing the biomarker measurements were blinded to each patient’s clinical information. All biomarkers were measured from frozen aliquots that did not undergo any additional freeze–thaw cycles.

The clinical primary endpoint was defined as the development of advanced AKI (stage 2 or 3, specified by the KDIGO criteria [32]), with both urine and creatinine criteria applied. The secondary outcomes were 90-day mortality after hospital discharge and composite outcomes defined as 90-day mortality after hospital discharge or advanced AKI, whichever occurred earlier.

All analyses were performed using SPSS software, version 20 (IBM, Armonk, NY, USA), R software, version 3.2.2 (Free Software Foundation, Inc., Boston, MA, USA), and MedCalc Statistical Software, version 15.11.3 (MedCalc Software bvba, Ostend, Belgium; https://​www.​medcalc.​org; 2015). The two-sample t test or Mann–Whitney rank sum test was used as appropriate to compare continuous variables; for categorical variables, the chi-square (χ2) or Fisher’s exact test was applied. Friedman two-way analysis of variance (ANOVA) was used to assess the overall difference in HJV, KIM-1, NGAL, α-GST and π-GST between the “no AKI or stage 1 AKI” and the “stage 2 or 3 AKI” groups at 0, 3, 6, 12 and 24 h after cardiovascular surgery [33]. We normalized biomarker levels with urine creatinine concentrations and analyzed them at each time point [34]. We fitted logistic regression models to examine the association between each marker (urinary NGAL, HJV, KIM-1, α-GST and π-GST) and advanced AKI, and generated an area under the receiver-operating characteristic (ROC) curve to assess the predictive accuracy of each marker. Power analysis was conducted based on the prior knowledge that the ratio of cases to controls was 4:1. We required 105 patients (21 with severe AKI and 84 without severe AKI) to achieve power of 0.8, with type I error of 0.05. This was based on the preliminary knowledge that the discriminatory power of urinary HJV to predict severe AKI was 0.7 [35]. We used a generalized additive model (GAM) (with spline) incorporating the subject-specific (longitudinal) random effects and adjusted for other clinical parameters to predict the outcomes [36, 37]. Simple and multiple generalized additive models (GAMs) were fitted to detect nonlinear effects of continuous covariates and identify appropriate cutoff point(s) for discretizing continuous covariates, if necessary, during the stepwise variable selection procedure. We defined the optimal cutoff value as odd equals to zero [38].

To assess the additive prediction ability of each biomarker compared with traditional clinical predictors, we identified clinical risk prediction models (including Liano’s score, CCF ARF score and SOFA score) that were based on area under the curve (AUC) analysis and then added each biomarker individually to this clinical model. We estimated the risk score function using a logistic regression model including clinical and biomarker variables as covariates. The Hosmer-Lemeshow logistic regression model test for goodness of fit was used to assess the calibration between the current model and the expected model. Finally, we calculated the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) to estimate overall improvement in reclassification with urinary biomarkers adding to clinical variables [39]. We reclassified the patients who developed advanced AKI using 0–12%, 12–30% and > 30% for the risk categories. A p value <0.05 was considered significant.

Hierarchical clustering analysis was performed using the cluster program and the results were visualized using the Treeview program [40]. The biomarkers were arranged in such a way that the most similar expression profiles were placed next to each other. In the color scheme, strong positive staining is indicated as a red cube, weak positive staining as a black cube and negative staining as a green cube. Absence of staining data is indicated with a grey cube.

A total of 149 patients who underwent cardiovascular surgery were enrolled in this study, with subsequent time-varying sample collection. There were 46 patients who had AKI episodes as defined by KDIGO criteria (30.9%), with 28 (18.8%) having stage 1 AKI, 11 (7.4%) stage 2 AKI, and 7 (4.7%) progressing to stage 3 AKI: 6 patients progressed to dialysis-dependent AKI in the advanced AKI group. The 90-day mortality rate in this cohort was 10.7%. The clinical characteristics of patients with and without advanced AKI are described in Table 1. Patients who developed advanced AKI had lower hemoglobin and albumin levels, were administered higher inotropic equivalents and had higher SOFA scores. However, there were no statistical differences between patients with and without advanced AKI with respect to age, gender, body mass index (BMI), comorbidities, baseline kidney function, baseline chronic kidney disease (eGFR between 30 and 60 ml/min/1.73 m²) status and the Liano and CCF ARF disease severity scores. Clinical outcomes were worse for patients in the advanced AKI group when they had longer ICU stays and a higher 90-day mortality rate, even after discharge.

We compared urinary biomarker levels between patients with and without advanced AKI (patients without AKI or patients with stage 1 AKI): the normalized urine biomarker data were transformed logarithmically to avoid the interference of numbers in the extreme and to approximate normal distribution in the analyses (Fig. 1). At 3 h post-surgery, normalized uHJV (p = 0.006), and uKIM-1 (p = 0.019) levels in the patients with advanced AKI were significantly higher than those without advanced AKI. Moreover, considerably higher levels of uHJV and uKIM-1 were observed at 6 and 12 h post-surgery. However, there were no significant differences in the uNGAL, uα-GST and uπ-GST levels between the patients with and without advanced AKI at these sequential time points. All biomarkers analyzed by Friedman two-way ANOVA (HJV, KIM-1, NGAL, α-GST and π-GST) were significantly different (all p <  0.05) between the no AKI/stage 1 AKI and the stage 2/3 AKI groups at each of the time points of 0, 3, 6, 12 and 24 h after cardiovascular surgery.

Among the patients who subsequently developed advanced AKI, the, uHJV biomarker had the highest AUC values, particularly at 12 h post-surgery. Following normalization with urinary creatinine concentration, the predictive ability of several urinary biomarkers was improved, especially normalized uHJV at 12 h post-surgery (Table 2). When compared to serum creatinine, normalized uHJV and uKIM-1 had better performance for predicting advanced AKI at 3 and 6 h post-surgery (Additional file 1: Table S1 and Fig. 1). The combined panel of normalized uHJV and uKIM-1 at 3 h post-surgery had the largest AUC (AUC = 0.898, 95% CI 0.80–0.96) compared to any of the other two or three biomarker combinations (Table 3 and Fig. 2). This combination had a positive predictive value of 42%, negative predictive value of 100%, sensitivity of 100% and specificity of 70%. The GAM plot was generated thereafter and demonstrated positive correlation between increased normalized uHJV and uKIM-1 at 3 h post-surgery and the risk of developing advanced AKI. The cutoff value 2.346 ng/mL for normalized uHJV and 0.047 ng/mL for normalized uKIM-1 performed best to predict advanced AKI (Additional file 1: Figure S2).

In combining the predictive ability of novel biomarkers and various clinical scoring systems for AKI, we found that the greatest AUC was observed with the combination of Liano’s AKI score and the panel of normalized uHJV and uKIM-1 at 3 h after surgery, with an AUC of 0.931 to predict advanced AKI. This panel had a greater AUC value than other combinations of biomarkers at 3, 6, 12, or 24 h after surgery (Table 4 and Fig. 2).

We assessed reclassification and evaluated any improvement in detection of advanced AKI events. The combination of the creatinine-normalized uHJV and uKIM-1 biomarker panel with the Liano’s score led to a significant increase in risk stratification (total NRI = 1.149; p < 0.001). The majority of this effect came from those with advanced AKI events (NRI event = 0.538; p < 0.001), whereas the NRI non-event was 0.610 (95% CI, p < 0.001) (NRI reclassification table and proportion was shown in Additional file 1: Table S2). Simultaneously, the total integrated discrimination improvement (IDI) was significant at 0.383 (95% CI, 0.25–0.52; p < 0.001). The combination of normalized uHJV and uKIM-1 with the Liano’s score was also significant for NRI and IDI reclassification in forecasting advanced AKI at 6, 12, and 24 h post-surgery (Table 5).

The combined panel of normalized uHJV and uNGAL at 3 h post-surgery had strong predictive ability for 90-day mortality rate after hospital discharge (AUC = 0.896) and the best predictive ability for composite outcomes (AUC = 0.905). The predictive ability for composite outcomes further improved after combination with Liano’s score (AUC = 0.943) (Additional file 1: Table S3). Given the potential for collinearity between some of the biomarkers, we performed an unsupervised cluster analysis to determine the relationship of the biomarkers to advanced AKI (Fig. 3). The unsupervised clustering resulted in groupings similar to functional groupings for some of the biomarkers. For instance, uHJV and uKIM-1 in the group associated with proximal tubular dysfunction were clustered in the near group, nearest to “advanced AKI”, and may explain their strong predictive ability for advanced AKI. Overall, the correlation within each group was stronger for biomarkers that had better predictive ability.