Background
Acute kidney injury (AKI) is a common and serious complication after cardiac surgery. Using standardized definitions of AKI based primarily on an increased serum creatinine (SCr), 10% to 40% of patients undergoing cardiac surgery develop AKI [
1‐
7]. AKI after cardiac surgery is associated with increased short-term and long-term mortality, increases in length of ICU and hospital stay, ventilator days, cost of hospitalization, and risk of developing chronic kidney disease (CKD) and end-stage renal disease (ESRD) [
8‐
11]. Staging AKI according to RIFLE (Risk, Injury, Failure, Loss of kidney function, and End-stage kidney disease), AKIN (Acute Kidney Injury Network), and/or KDIGO (Kidney Disease Improving Global Outcomes) criteria showed that even mild forms of AKI not requiring dialysis are associated with increased morbidity, mortality, and risk of CKD [
8,
12‐
15]. The diagnosis of AKI primarily depends on an increase in SCr concentration, which typically does not occur until 24 h to 72 h after injury [
16]. The delay in diagnosis until injury contributes to the failure in human trials to reproduce successful interventions of experimental animal models [
17,
18]. Without effective treatment of AKI, clinical management focuses on prevention and risk management.
Due to the importance of post-cardiac surgery AKI and influence on patient morbidity and prognosis, it is critical that patients and providers have a realistic pre-surgical understanding of AKI risk. Preoperative risk stratification for AKI after cardiac surgery is necessary for clinical decision making, for pre- and intra-operative treatment to minimize the risk of AKI, and to identify high-risk patients for clinical trials. A model developed at the Cleveland Clinic, using a combination of laboratory (including SCr) and clinical findings [
19], was reported to best predict cardiac surgery-related AKI [
3,
20]. That model, however, was developed to predict the risk of AKI requiring dialysis, and its ability to predict AKI of less severity is more limited [
21]. Kiers et al. [
3] reported an area under receiver operating curve (AUROC) of 0.75 for AKI-Risk and 0.81 for AKI-Injury, compared to an AUROC of 0.93 for AKI requiring dialysis.
A number of urine and blood biomarkers, including neutrophil gelatinase-associated lipocalin (NGAL), interleukin-18, cystatin C, and kidney injury molecule-1 (KIM-1), increase before SCr, improving the early diagnosis of AKI [
4,
22‐
26]. Addition of pre-operative cystatin C in place of SCr to the clinical AKI risk assessment was reported to modestly improve risk stratification [
7]. Identification of new biomarkers that replace or enhance current clinical risk stratification is needed to allow clinicians to apply appropriate preventive measures and to design clinical trials to identify effective therapies. The purpose of the present study was to identify preoperative candidate urine biomarkers in patients undergoing cardiac surgery that, alone or in combination with the current clinical scoring tool, would improve prediction of AKI. The data indicate that addition of preoperative urine HRG or complement factor B to the clinical scoring tools may improve the accuracy of prediction of AKI after open-heart surgery [
19,
27].
Discussion
Patients who develop AKI after cardiac surgery exhibit prolonged hospital stay, increased short-term and long-term mortality, and an increased risk of CKD [
9]. As no effective treatment for established AKI exists, current clinical management focuses on risk factor assessment and prophylaxis. Sensitive and specific prediction of the risk of developing AKI is critical to identifying patients in whom the risk of AKI outweighs the benefits of surgery or in whom aggressive preoperative risk-reduction management is appropriate. The combination of preoperative laboratory and clinical evaluation was shown to be the best predictor of AKI whether or not dialysis was required [
3]. That evaluation was highly predictive of AKI requiring dialysis with an AUROC of 0.93, however, the AUROC for AKI not requiring dialysis was only 0.75. As even milder forms of AKI are associated with worse short-term and long-term outcomes, improved risk assessment for AKI not requiring dialysis is needed. The current study tested the hypothesis that preoperative prognostic urinary biomarkers of post-surgical AKI
R could be identified using a proteomic approach and would be valuable adjunctive biomarkers for risk assessment. Six recognized diagnostic markers of AKI
R were detected, of which five (KIM1, NGAL, L-FABP, IGFBP7 and S100A8/S100A9) were not significantly different in patients developing AKI and one (AGT) was significantly different. AGT and four additional proteins (C3, CFB, HRG, IGFBP3), selected using rank ordering of abundance differences and pathways analysis, were studied by ELISA on the entire sample set. Two proteins in preoperative urine samples, HRG (AUROC 0.79) and CFB (AUROC 0.75), performed as well as the risk factor score (AUROC 0.73) and preoperative SCr (AUROC 0.79).
A multivariate model for the prediction of AKI
R performed better than any single factor measured. The addition of HRG or CFB to the risk factor score significantly improved the AUROC to 0.90 and 0.89, respectively. Differences in preoperative SCr observed between patients with and without AKI were corrected for in the model building process and did not explain the contribution of the newly discovered biomarkers to the prediction of AKI
R. Despite the sensitivity of the CFB ELISA being at or near the measured concentration of nearly one-half of the samples analyzed, CFB still performed well in the multivariate model. CFB should not be ruled out as an important predictor of AKI
R and the development of a more sensitive assay could benefit this biomarker. The contribution of HRG and CFB to the improvement in the prediction of AKI
R (event) or no AKI
R (non-event) was evaluated using logistic regression and two statistical tests, NRI and IDI, developed specifically for the evaluation of potential biomarkers. NRI calculates the contribution using the binary values of 0 and 1 based on group membership (event, non-event) and IDI calculates the contribution based on probability 0.0 to 1.0 of the event occurring [
38]. NRI and IDI were applied when factors were identified using logistic regression and were used to identify where prediction was improved (event, non-event, both) and as such are additive to the information displayed in the c-statistic. HRG and CFB influenced both the prediction of event (IDI only) and non-event (IDI only). The combined prediction (event + non-event) was also significant for CFB using both NRI and IDI. Use of HRG in the prediction of AKI
R resulted in an improvement in sensitivity over risk factor score from 0.38 to 0.63, a 25% improvement. There was improvement in 1-specificity from 0.23 to 0.15, an 8% improvement. Use of CFB in the prediction of AKI
R resulted in an improvement in sensitivity over risk factor score from 0.38 to 0.64, a 26% improvement, and an improvement in 1-specificity from 0.23 to 0.15, an 8% improvement.
The primary definition of AKI (AKI
R) in the current study was more restrictive that those proposed using RIFLE, AKIN, or KDIGO criteria. We initially used a more restrictive definition, to reduce over diagnosis of AKI due to changes in fluid balance after cardiac surgery potentially leading to misleading changes in SCr or urine output [
28]. Based on the improvement in predictive capability with CFB and HRG for AKI
R, we re-analyzed predictive capability using the KDIGO criteria for AKI. That analysis showed that HRG and CFB continued to significantly predict development of AKI, although the prediction was less sensitive than for AKI
R. Clinical risk factor score and preoperative SCr failed to predict AKI
KDIGO. It is not possible to determine whether there was increased misdiagnosis of AKI using KDIGO criteria or if the predictors of AKI are less reliable for very mild cases.
CFB is a C3-convertase involved in alternative complement pathway activation and amplification [
42,
43]. Genetic deletion of CFB or administration of anti-CFB monoclonal antibodies significantly impaired development of AKI in mice subjected to ischemia/reperfusion injury [
44‐
46]. That reduction in CFB also significantly reduced the deposition of C3b on tubular epithelial cells and accumulation of neutrophils in the renal interstitium. Renal tubular cells showed increased CFB production in mice subjected to cecal ligation and puncture model of microbial sepsis and in cultured human proximal tubular cells stimulated with toll-like receptor agonists [
46,
47]. Thus, increased CFB in the urine of patients undergoing cardiac surgery may identify those patients with underlying tubular cells CFB production that predisposes to complement-mediated tubular cell injury during surgery.
HRG is an abundant plasma glycoprotein with a multidomain structure that allows the molecule to interact with many ligands, including the complement components C1q, factor H, C8, C4, and C3 [
48]. In addition to binding to a number of complement components, HRG was reported to inhibit complement factor D-mediated cleavage of CFB [
49]. Although no association of HRG with AKI has been reported previously, the multiple protein-protein interactions with HRG regulates formation of immune complexes, removal of apoptotic cells, microbial invasion, cell adhesion, angiogenesis, coagulation, and progression of tumor growth [
48].
The strengths of our study included a clear and restrictive definition of AKI using pre-operative and multiple post-operative serum and urine data to define the patient populations studied. A second strength of this study was utilization of high-sensitivity, high-mass accuracy proteomic methods to address the novel hypothesis that the pre-operative urine proteome was associated with post-cardiac surgery AKI. Importantly, confirmation of two candidate risk biomarkers using ELISA, an orthogonal method, was performed on all subjects. Those candidate biomarkers significantly enhanced the value of the clinical risk score for milder forms of AKI prediction. The clinical benefits of enhanced AKI prediction include: (1) improved application of prophylactic measures to a high risk population, (2) improved clinical assessment of the risk-to-benefit ratio of surgery, and (3) better patient cohort design for studies investigating AKI management and treatment.
Our study has some limitations. First, our data are specific to patients at higher risk for AKI who underwent cardiac surgery, and may not generalize as well to other patient populations. Additionally, using the risk factor score to focus our study on patients at higher risk skewed the scoring range in our population. This could confound the comparison of the risk factor score with urinary biomarkers. However, the AUROC for the risk factor score in our study (0.73) was similar to that previously reported in similar group of patients developing AKI not requiring dialysis (0.75) [
3]. As patients in our study had AKI of mild severity, our urinary biomarkers may not improve risk prediction in patients with AKI requiring dialysis. Second, we did not study other biomarkers, except SCr, that have been associated with AKI following cardiac surgery. A previous study showed that pre-surgical serum cystatin C levels had a stronger and a more linear association with AKI risk than pre-surgical SCr [
7]. A reduced urine uromodulin to SCr ratio was reported to be associated with an increased risk of AKI after cardiac surgery upon univariate, but not multivariate, analysis [
50]. In neither report was the risk assessment determined by adding cystatin C or uromodulin values to the standard clinical risk assessment. Third, we did not have a validation set for our study, so confirmation will require further investigation. A larger cohort of unselected patients should generate a more powerful evaluation of whether HRG and CFB can improve risk discrimination for AKI.