Author + information
- Received April 22, 2013
- Revision received August 17, 2013
- Accepted September 11, 2013
- Published online February 4, 2014.
- Femi Philip, MD∗,
- Heather L. Gornik, MD∗,
- Jeevanantham Rajeswaran, PhD†,
- Eugene H. Blackstone, MD∗,† and
- Mehdi H. Shishehbor, DO, MPH, PhD∗∗ ()
- ∗Department of Cardiovascular Medicine, Heart & Vascular Institute, Cleveland Clinic, Cleveland, Ohio
- †Department of Quantitative Health Sciences, Heart & Vascular Institute, Cleveland Clinic, Cleveland, Ohio
- ↵∗Reprint requests and correspondence:
Dr. Mehdi H. Shishehbor, Heart & Vascular Institute, Cleveland Clinic, 9500 Euclid Avenue, J3-05, Cleveland, Ohio 44195.
Objectives The aim of this study was to assess the impact of atherosclerotic renal artery stenosis (ARAS) on outcomes after open-heart surgery (OHS).
Background Acute kidney injury after OHS portends significant morbidity and mortality.
Methods Data from all adult patients undergoing OHS from January 2000 to April 2010 who underwent renal duplex ultrasound were prospectively collected. ARAS was severe (60% to 99% stenosis) if peak systolic velocity was >200 cm/s. The associations between ARAS and post-operative reduction in glomerular filtration rate (GFR), need for renal replacement therapy, length of stay, and overall short-term and long-term mortality (up to 8 years) were tested using multivariate time-to-event adjusted analysis.
Results A total of 714 patients were evaluated, with a mean age of 67 ± 12 years (63% men) and a mean GFR of 52 ± 25.9 ml/min/1.73 m2. A total of 206 (29%) had ARAS; of these, 79% (n = 163) had unilateral and 21% (n = 43) had bilateral ARAS. ARAS was associated with peripheral artery disease (p = 0.004) and lower high-density lipoprotein levels (p = 0.04). Patients with advanced age (p = 0.01) and descending aorta grafting (p = 0.004) had significant post-operative reductions in GFR. Adjusted models showed a nonsignificant trend between ARAS and reduction in GFR (p = 0.09). ARAS was not associated with need for renal replacement therapy (p = 0.4), longer length of stay (p = 0.7), or mortality (p = 0.7), but low pre-operative GFR was a strong predictor of long-term mortality.
Conclusions ARAS does not appear to be associated with post-operative change in GFR, need for hemodialysis, longer length of stay, or mortality in patients undergoing OHS.
Acute kidney injury (AKI) after open-heart surgery (OHS) portends significant morbidity and mortality (1–3). Depending on the definitions used, post-operative AKI occurs in 3% to 30% of patients, and AKI requiring renal replacement therapy (RRT) develops in 1% to 5% of patients (1,2). The latter group has a mortality of 60% compared with an overall mortality of 2% to 8% after OHS (3). Patients with AKI have a 4-fold increased risk for short-term and long-term death. In fact, increases in serum creatinine have an exponential association with risk for 30-day mortality (<0.5 mg/dl from baseline associated with a 3-fold risk and >0.5 mg/dl increase associated with an 18-fold increased risk) (3). The underlying pathophysiology of AKI after cardiac surgery is renal ischemic injury due to a variety of factors, including intraoperative hypotension and complications that impair renal perfusion or lead to atheroembolic or thromboembolic events to the kidneys (2).
Atherosclerotic renal artery stenosis (ARAS) is common and ranges in incidence from 25% to 38% in patients with established atherosclerotic vascular disease (4). However, the impact of ARAS on post-operative glomerular filtration rate (GFR) and clinical outcomes is unknown. We sought to study the impact of ARAS on post-operative change in GFR, AKI with need for RRT, length of stay, and overall short-term and long-term mortality.
Using the Cardiovascular Information Registry, we identified 37,000 consecutive patients age >18 years, undergoing first-time OHS from January 2000 to April 2010. This registry contains detailed demographic, clinical, pathologic, operative, and outcome variables on all patients undergoing cardiac surgery at Cleveland Clinic, abstracted from clinical records concurrent with patient care. The Cardiovascular Information Registry follow-up information was supplemented using the Social Security Death Index for mortality data. This database was cross-referenced with a vascular ultrasound database to identify 714 patients who had undergone renal vascular ultrasound in the 90 days before or after OHS. We collected follow-up information at 30, 60, and 90 days and yearly. These registries are approved for use in research by the institutional review board.
GFR was calculated using the Modification of Diet in Renal Disease study formula: GFR (ml/min/1.73 m2) = 186.3 × serum creatinine (mg/dl) − 1.154 × age (years) − 0.203 (× 1.212 if black or × 0.742 if female) (5). Renal artery stenosis was defined using the renal ultrasound criteria measuring peak systolic velocity (PSV) and end-diastolic velocity and the renal/aortic resistive index (RAR). The renal ultrasound definitions used were 0% to 59% (PSV <200 cm/s and RAR <3.5), 60% to 99% (PSV ≥200 cm/s and RAR ≥3.5), >80% (PSV ≥200 cm/s, RAR ≥3.5, and end-diastolic volume >150 cm/s), and occluded (100%) in the absence of flow (6).
Primary endpoints included pre-discharge GFR and the degree of change in GFR over time. Secondary endpoints included in-hospital morbidy and mortality, length of stay, and long-term all-cause mortality.
Simple descriptive statistics were used to summarize the data. Continuous variables are presented as mean ± SD and as 15th, 50th (median), and 85th percentiles. Categorical data are described using frequencies and percents, and comparisons were made using Wilcoxon rank sum tests or chi-square tests. Transformations of scale were performed on continuous variables to meet statistical model assumptions, and the results of regression models are presented with their coefficients rather than odds or hazard ratios.
A number of variables examined in the multivariate analyses had missing values (ranging from 0% to 22%). A 5-fold Markov-chain Monte Carlo imputation technique was used to impute the missing values (7). Estimates of regression coefficients and their variance covariance obtained for each of the imputed datasets were then combined to yield the final regression estimates and p values using PROC MIANALYZE in SAS version 9.0 (SAS Institute Inc., Cary, North Carolina).
Factors associated with graded ARAS and need for RRT were identified using multivariate cumulative and binary logistic regression models using PROC LOGISTIC in SAS. Variables affecting length of stay were identified using multivariate linear regression using PROC REG in SAS. Temporal trend analysis and factors influencing change in pre-discharge GFR were analyzed using longitudinal regression model P-1 (PROC MIXED in SAS) (8).
Survival was assessed nonparametrically using the Kaplan-Meier method and parametrically using a multiphase hazard mode. Variable selection (with a p-value criterion for retention of variables in the model) used bootstrap bagging (bootstrap aggregation) (9,10). This was a 4-step process. First, a new dataset was created by randomly selecting patients with replacement from the original dataset. Second, risk factors were identified using an automated forward stepwise selection. Third, the results of the variable selection were stored. These 3 steps were repeated 1,000 times. Finally, the frequency of occurrence of variables related to group membership was ascertained and indicated the reliability of each variable (bootstrap aggregation step).
Random forest analysis (RFA)
Random forest regression was performed as a confirmatory sensitivity analysis for the impact of renal artery stenosis on operative length of stay and need for dialysis in 714 patients using 1,000 trees. Using this machine learning tool, we were able to identify and depict the relationships of selected baseline variables with the binary (need for RRT) and continuous (length of stay) variables, as shown in Figure 1. The classification or regression tree method constructs a tree by recursive binary partitioning of the data into regions that are increasingly homogenous with respect to the binary or continuous response variables (10–12). The classification accuracy was improved by aggregating the results of many trees, each grown from a “bootstrap” dataset formed by random sampling of the data. We constructed 1,000 trees for random forest classification modeling and used 20 variables to split each node using the randomForest package in R (R Foundation for Statistical Computing, Vienna, Austria) (11). There were no pre-specified assumptions regarding variables, and randomization was introduced into this model by both random bootstrap sampling of patients from the original cohort and random sampling of variables for each tree branch. The random forest approach is preferable when it is necessary to depict a complex relationship (linear or nonlinear) between a predictor and an outcome variable. It is also used to identify the complex interaction effect (if any) among predictors and an outcome.
A power analysis was performed for the endpoint of need for dialysis on the basis of a sample size of 584 patients with 12% requiring dialysis using multivariate logistic regression. The calculated power of this logistic regression to detect a significant difference was >90%.
The demographic, pre-operative laboratory, and operative characteristics are listed in Table 1. In this study of mostly men with normal renal function, there was a high incidence of risk factors for atherosclerosis and established atherosclerotic vascular disease. The most common OHS performed was coronary artery bypass graft surgery, followed by aortic valve replacement or repair. The use of intraoperative support devices and blood products was low. None of the patients in this analysis underwent renal artery balloon angioplasty, stent implantation, or bypass surgery before OHS. In addition, none of the patients with ARAS underwent renal nuclear scanning before OHS.
Prevalence of ARAS and risk factors for ARAS
The distribution of renal ultrasound studies relative to the date of OHS is shown in Online Figure 1. Of the total patients, 29% (n = 206) had ARAS; of these, 79% (n = 163) had unilateral and 21% (n = 43) had bilateral renal artery stenosis (60% to 99%), as shown in Table 1. The unadjusted comparisons of patient and procedural characteristics that were associated with a higher grade of ARAS were prior peripheral artery disease, carotid artery disease, older age, lower body mass index, and need for coronary artery bypass grafting. On multivariate analysis, the independent predictors of a higher grade of ARAS (no renal artery stenosis vs. unilateral vs. bilateral renal artery stenosis) were a history of prior peripheral artery disease (p = 0.004), female sex (p = 0.0005), presence of low high-density lipoprotein (p = 0.04), low body mass index (p = 0.0002), and older age (p = 0.008), as shown in Table 2. In addition, on multivariate analysis, the predictors of bilateral ARAS versus unilateral ARAS were history of peripheral artery disease, female sex, and older age.
Association between ARAS and GFR
An unadjusted temporal trend analysis showed that post-operative GFR changed significantly over time. There was no difference in the average GFR or minimal GFR in patients with no ARAS, unilateral ARAS, or bilateral ARAS (Online Table 1). The unadjusted graded effect of the presence of ARAS on post-operative GFR showed a nonsignificant association (p = 0.07). A late separation was noted between ARAS and GFR; however, this was not statistically significant (Online Figs. 2a and 2b). On multivariate analysis, patients with lower GFR were significantly older, had hypertension, had diabetes, and had endocarditis. In addition, patients who underwent aortic surgery or coronary artery bypass grafting without the use of internal mammary artery grafts were more likely to have lower post-operative GFR, as shown in Table 3. Further subgroup analysis confirmed no association between ARAS and change in GFR, even when the groups with no ARAS and unilateral ARAS were combined and compared with the group with bilateral ARAS (p = 0.40).
Multivariate predictors of the need for RRT
Patients with lower platelet counts, histories of diabetes requiring insulin, need for left ventricular support, and any aortic surgery had a higher likelihood of requiring RRT in the post-operative period, as shown in Table 3. When ARAS was considered a binary variable (yes or no) or as an ordinal variable, it was not associated with need for RRT. A confirmatory RFA was performed with the top 3 predictors (pre-operative platelet count, pre-operative hematocrit, and body surface area) in conjunction with the study variable (presence or absence of ARAS) on the probability of needing RRT (Online Figs. 3a and 3b). These figures do not show a noticeable effect of the presence of ARAS on the probability of needing RRT. Furthermore, when we combined the groups with no ARAS and unilateral ARAS and compared them with the group with bilateral ARAS, we found no association with need for RRT (p = 0.50).
ARAS and its association with length of stay
Female sex, need for prior ventricular assist device, and any aortic surgery were all associated with longer length of stay, as shown in Table 3. However, in the risk-adjusted models, the presence of ARAS as a binary variable or as an ordinal variable (graded effect) did not influence length of stay. Additionally, RFA using the 3 strongest predictors of increased length of stay (isolated coronary artery bypass grafting, need for a ventricular assist device, and cholesterol levels) found no effect of ARAS on operative length of stay (Online Fig. 4). Furthermore, when the groups with no ARAS and unilateral ARAS were combined and compared with the group with bilateral ARAS, we found no association with operative length of stay (p = 0.60).
Multivariate predictors of all-cause death
The median follow-up duration was 3 years, with a total of 2,501 patient-years available for the analysis. A multiphase hazard analysis yielded 2 different overlapping risk periods (phases), an early peaking risk for death (approximately up to 1 year) followed by a constant risk for death (beyond 1 year). The parametric estimates of survival for this group of patients was 95% at 1 month, 83% at 6 months, 79% at 1 year, 66% at 3 years, 55% at 5 years, and 42% at 8 years. On multivariate analysis, the risk factors for early death included older age, diabetes, congestive heart failure, lower hematocrit, and any tricuspid valve or aortic procedure, as shown in Table 4. The risk factors for late death included older age, history of congestive heart failure, and lower GFR, as shown in Table 4. The presence of ARAS or the graded level of ARAS did not appear to be associated with risk for death (early or late death) (Figs. 2a and 2b, Table 4). Furthermore, when the groups with no ARAS and unilateral ARAS were combined and compared with the group with bilateral ARAS group, we found no association with all-cause death (p > 0.60).
This study adds to the limited body of evidence documenting the impact of ARAS on renal dysfunction and mortality after OHS (13). In addition, it builds on these findings by using a more reliable estimate of renal function (GFR), examines a variety of outcomes in addition to operative mortality, and assesses the impact of ARAS on long-term mortality. We demonstrate there is no association between the presence of ARAS and significant change in GFR or lowest GFR after OHS. Additionally, there is no association between ARAS and mortality, length of intensive care unit stay, or need for RRT.
Renal dysfunction after OHS and ARAS may share renal ischemia as a common pathophysiological mechanism (14). The mechanism of renal injury may be different with AKI in OHS (as seen in hypotension), affecting glomerular function while chronic renal hypoperfusion (seen in ARAS) results in intrarenal compensatory mechanisms that preserve glomerular perfusion at the expense of parenchymal ischemia (15). Autopsy studies of patients with ARAS have shown predominant tubulointerstitial atrophy (parenchymal injury) with glomerular sparing, suggesting relative preservation of glomerular reserve and capacity to tolerate hypoperfusion (16). Furthermore, deterioration of renal function in the presence of ARAS may not reflect “true ischemia” under normal conditions, because blood flow to the kidney is far in excess of metabolic needs (17). Moderate reductions in blood flow, as in ARAS, may not be the sole or even the major contributor to reduced renal function (18). The view that reduction in renal blood flow directly results in ischemic damage may be overly simplistic. Additionally, there does not appear to be a tight relationship between renal artery diameter and loss of GFR (19). In studies of moderate renal artery stenosis (50% to 70%), GFR ranged from 30 to 38 ml/min and was unrelated to renal artery diameter (20). Furthermore, the time to end-stage renal disease in patients with ARAS was the same irrespective of the degree of renal artery stenosis, and paradoxically, the Cox proportional hazard analysis of risk for developing end-stage renal disease indicated that moderate ARAS > 50% was more of a risk (relative risk [RR]: 3.39) than severe ARAS (relative risk: 0.95) (19). Corollaries to these observations are noted in the clinical trials assessing endovascular treatment for ARAS for preservation of renal function. These studies would, in theory, serve to attenuate the impact of renal artery stenosis on renal function. Among others, the 2 most contemporary prospective studies using angioplasty and stent implantation were the Stent Placement and Blood Pressure and Lipid-Lowering for the Prevention of Progression of Renal Dysfunction Caused by Atherosclerotic Ostial Stenosis of the Renal Artery and Angioplasty and Stenting for Renal Artery Lesion trials, which showed no clear association between endovascular intervention and decline in GFR (despite some limitations) (21–23). Our study is in keeping with these randomized trials, suggesting that treatment for ARAS does not prevent renal dysfunction.
The impact of an atherosclerotic lesion may relate to its effects not only on perfusion pressure but also on the duration of the hemodynamic insult, effects on kidney parenchyma due to hypertensive nephrosclerosis, and recurrent atheroembolic insults (23). Despite strategies to minimize renal dysfunction, we note that among others, the need for pre-operative hemodynamic support and aortic surgery was associated with increased risk for RRT (24). These factors are minimized in the current planning for contemporary OHS and post-operative management. In our study, independent of baseline GFR, ARAS was associated with a marginal decline in the rate of recovery of GFR in the latter stages after OHS on the temporal trend analysis. This may reflect a delayed capacity for renal recovery in response to ischemia due to the presence of ARAS.
We also show that the presence of ARAS does not affect short-term or long-term mortality, but a low GFR (<50 ml/min/1.73 m2) is a predictor of long-term mortality (>1 year). These findings are in contrast with those of prior studies that showed significantly higher mortality in patients with ARAS (25,26). The majority of studies testing the association between ARAS and overall mortality were limited by small sample sizes and design issues (retrospective and case-control studies). The largest study used Medicare claims data (n = 1,085,250) from 1999 to 2001 and showed that patients with ARAS had significantly higher rates of atherosclerotic disease, with a higher rate of triple ischemic endpoints (myocardial infarction, stroke, and death) (27,28). However, this study was retrospective, observational, and cross-sectional by design, making these data open to confounding factors. In addition, these observations were made in an era without contemporary medical therapy, making them of limited clinical relevance. We similarly show that patients with ARAS have a higher burden of established atherosclerotic disease, require more aortocoronary grafts, and tend to require more hemodynamic support, but we do not show any association with mortality. We note that long-term mortality is associated with baseline GFR and not the presence of ARAS (29).
We also used RFA to test the interaction of various variables in affecting our outcomes. There are 4 advantages to using RFA. First, the RFA method is intuitive because important variables to predict an endpoint can be identified by inspecting the tree trunks and simplified in a figure plotting the minimal depth of a variable from the tree trunk. Second, RFA do not require analysts to know in advance the relationship (i.e., linear, nonlinear) of a variable over time or to choose the best equation to transform nonlinear covariates. Third, the complex interactions among multiple variables can be easily understood with RFA. Finally, the overall accuracy of an RFA model is at least comparable to standard methodologies (10). RFA is a well-known and highly used machine learning method and has been used successfully in several applied settings, including staging for esophageal cancer (12).
First, it was observational in nature, and despite adjustment for confounders in the multivariate analysis, we can only investigate associations and cannot infer causality. Second, there were no differences in baseline GFR between the 2 groups, but the 2 groups studied fall into the category of chronic kidney disease, which may mask the true effect of ARAS on renal function and outcomes. Third, we allowed renal ultrasound studies up to 90 days after OHS; this could have produced selection bias. However, only 15% of renal ultrasound studies were performed after OHS, and a subgroup analysis excluding these patients did not reveal an association between ARAS and the endpoints of ARAS, length of stay, or need for RRT. Fourth, given the moderate sample size, the lack of an association between ARAS and operative and long-term outcomes after OHS may be due to inadequate power. Fifth, renal artery ultrasound may not be the best method to determine the clinical significance of ARAS, and novel tools such as direct pressure assessment may better determine the clinical significance of the ARAS in this setting (30).
We have shown no association between the presence of ARAS and change in GFR or lower GFR, need for RRT, length of intensive care unit stay, or mortality. ARAS found during ultrasound examination of the renal arteries in the post-operative period after OHS should be managed conservatively. What is unclear is the best management strategy with bilateral ARAS or severe unilateral ARAS in the highest risk category, as in patients undergoing aortic surgery or in those needing hemodynamic support.
The authors thank Kathryn Brock for her editorial assistance.
Dr. Shishehbor is a consultant and an educator for Medtronic, Abbott Vascular, Bayer, Spectranetics, Cordis, and Bard but has waived all compensations. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- acute kidney injury
- atherosclerotic renal artery stenosis
- glomerular filtration rate
- open-heart surgery
- peak systolic velocity
- renal/aortic resistive index
- random forest analysis
- renal replacement therapy
- Received April 22, 2013.
- Revision received August 17, 2013.
- Accepted September 11, 2013.
- American College of Cardiology Foundation
- Mangano C.M.,
- Diamondstone L.S.,
- Ramsay J.G.,
- Aggarwal A.,
- Herskowitz A.,
- Mangano D.T.,
- for the Multicenter Study of Perioperative Ischemia Research Group
- Rosner M.H.,
- Okusa M.D.
- Lassnigg A.,
- Schmidlin D.,
- Mouhieddine M.,
- et al.
- Rubin DB
- Diggle P.J.,
- Heagerty P.J.,
- Liang K.Y.,
- Zeger S.L.
- Ishwaran H.,
- Blackstone E.H.,
- Apperson-Hansen C.,
- Rice T.W.
- Yeboah E.D.,
- Petrie A.,
- Pead J.L.
- Keddis M.T.,
- Garovic V.D.,
- Bailey K.R.,
- Wood C.M.,
- Raissian Y.,
- Grande J.P.
- Cheung C.M.,
- Wright J.R.,
- Shurrab A.E.,
- et al.
- Weinberg M.D.,
- Olin J.W.
- Parolari A.,
- Pesce L.L.,
- Pacini D.,
- et al.
- Conlon P.J.,
- Athirakul K.,
- Kovalik E.,
- et al.
- Drieghe B.,
- Madaric J.,
- Sarno C.