The predictive value of creatinine clearance for mortality in patients undergoing revascularization

Background Renal function plays a significant role in the prognosis and management of patients with multi-vessel coronary artery disease (CAD) referred for revascularization. Current data lack precise risk stratification using estimated glomerular filtration rate (eGFR) and creatinine clearance. Methods This prospective study includes a three-year follow-up of 1112 consecutive patients with multi-vessel CAD enrolled in the 22 hospitals in Israel that perform coronary angiography. Results The Mayo formula yielded the highest mean eGFR (90 ± 26 mL/min per 1.73m2) and chronic kidney disease-epidemiology collaboration (CKD-EPI) the lowest (76 ± 24 mL/min per 1.73m2). Consequently, the Mayo formula classified more patients (56%) as having normal renal function. There was a significant and strong correlation between the values obtained from all five formulas using Cockcroft-Gault as the reference formula: Mayo: r = 0.80, p < 0.001; CKD-EPI: r = 0.87, p < 0.001; modification of diet in renal disease (MDRD): r = 0.84, p < 0.001; inulin clearance-based: r = 0.99, p < 0.001). Multivariable analysis demonstrated that decreased renal function is an independent predictor of 3-year mortality in all five formulas, with risk increasing by 15–25% for each 10-unit decrease in eGFR. Despite the similarities between the formulas, the ability to predict mortality was highest in the Mayo formula and lowest in MDRD. Conclusions Our data suggest that while the Mayo formula is not currently recommended by any nephrology guidelines, it may be an alternative formula to predict mortality among patients with multivessel CAD, including to the widely used MDRD formula. Supplementary Information The online version contains supplementary material available at 10.1186/s13019-021-01502-1.

There are a number of equations used for GFR estimation and CKD staging. For decades, the Cockcroft-Gault (CG) equation was the most commonly used method for estimating kidney function. Recently, however, due to inherent limitations of the CG, several newer equations have been developed [6][7][8]. The Modification of Diet in Renal Disease (MDRD) and, more recently, the CKD Epidemiology Collaboration (CKD-EPI) equation, are the most extensively used of the newer formulas. The Mayo Clinic quadratic equation [9] and the inulin clearance-based (IB) eGFR equation [10] were developed in an attempt to better estimate GFR in patients with preserved kidney function.
Current literature fails in providing adequate data regarding the ability of these five formulas to predict outcomes in patients with multivessel CAD undergoing either percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG). We aim to evaluate the performance of all five eGFR formulas in predicting 3-year all-cause mortality in a 'real-world' cohort of patients with multivessel CAD.

Study design and population
Patients were enrolled and prospectively followed-up in the Multivessel CAD Israeli Registry, details of which have been previously reported [11,12]. In brief, this registry includes 1112 consecutive adult patients from the 22 hospitals in Israel that perform coronary angiography. All patients had multivessel CAD (Fig. S1). The study was approved by the Institutional Review Board of each of the participating centers and all patients provided informed consent.

Statistical Analysis
Data are presented as mean ± standard deviation. Categorical variables are given as frequencies and percentages. A chi-square test was used for comparison of categorical variables between different renal function groups. Student's t-test was performed for comparison of continuous variables.
The correlation between CG, MDRD, Mayo, IB and CKD-EPI formulas was tested using Pearson's correlation coefficient. We performed a Bland and Altman analysis to assess the agreement between values derived from each of the other formulas and the values obtained from the CG formula.
Survival analysis was done using the Kaplan-Meier method, and statistical differences between predefined renal dysfunction groups were tested using the logrank test. Multivariate Cox proportional hazard modeling was used to identify factors associated with mortality risk at follow-up. Candidate factors appear in Table 2.
To evaluate the ability of the formulas to predict all-cause mortality, we estimated receiver operating characteristic (ROC) curves and area under the curve a × (SCr/b) c × (0.993) age Where the variable "a" takes on the following values based on race and sex: • Black: women = 166; men = 163 • White/other: women = 144; men = 141 The variable "b" takes on the following values based on sex: • Women = 0.7 • Men = 0.9 The variable "c" takes on the following values based on sex and creatinine measurement: • Women:  (AUC) with 95% confidence interval (CI) using corresponding logistic models.
Hosmer-Lemeshow goodness-of-fit test was used to assess model suitability. Furthermore, to predict the benefit incurred by the addition of a GFR formula to a baseline model of mortality prediction, we estimated net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Using binary logistic regression, we computed predicted risk for 3-year mortality from a baseline model without GFR (age, diabetes, history of congestive heart failure, atrial fibrillation, previous stroke, SYNTAX score, chronic obstructive pulmonary disease, hypertension, sex) and a similar model that included GFR (for each formula separately). For calculation of the NRI, rescaled individual predicted risk from baseline and GFR models were compared in three pre-specified risk thresholds; low (< 10%), intermediate (10-20%) and high risk (> = 20%). IDI and relative IDI were computed similarly to NRI, albeit without pre-specified risk thresholds.
Statistical significance was assumed when the null hypothesis could be rejected at P < 0.05. All P-values are the results of two-sided tests. Statistical analyses were conducted using R (version 3.4.1).

Baseline study population characteristics
The mean age of the cohort was 66 ± 11 years, with a male majority of 80%. Among the 1112 patients with multivessel CAD included in the registry, 126 (11.3%)  (14) 156 (20) Others (%) 0 (0) 0 (0) 13 (2) eGFR Estimated glomerular filtration rate, SD Standard deviation, PCI Percutaneous coronary intervention, COPD Chronic obstruction pulmonary disease, CHF Congestive heart failure, CVA Cerebrovascular accident, TIA Transient ischemic attack, ACE-I Angiotensin converting enzyme inhibitor had significant renal impairment (GFR ≤60 mL/min per 1.73m 2 ) according to all five formulas (MDRD, Mayo, CKD-EPI, CG, and IB formulas), and 783 (70.4%) had non-significant renal impairment (GFR > 60 mL/min per 1.73m 2 ) according to all five formulas. Of the remaining 203 patients (18.3%), the eGFR was discordant and shifted between < 60 mL/min per 1.73m 2 and > 60 mL/min per 1.73m 2 depending on the formula used. "Discordant eGFR" was used when a patient was considered to have normal renal function by at least one formula and abnormal renal function by at least one formula. Patients with normal renal function tended to be younger with fewer cardiovascular risk factors such as diabetes, hypertension, history of stroke, history of congestive heart failure, atrial fibrillation and lower SYNTAX scores ( Table 2). There were 5 patients on hemodialysis prior to intervention, all were in the significant renal impairment group, and all underwent revascularization by CABG. CABG was used as the revascularization strategy in 50% of the significant renal impairment group, 51% of the discordant group and 54% of the non-significant renal impairment group. Compared with PCI patients, those who underwent CABG were more likely to be male (83% vs. 77%, p = 0.013), had a higher frequency of diabetes (49% vs. 42%, p = 0.026), and were more likely to have had a prior stroke (12% vs. 7%, p = 0.014). Those who underwent CABG also had higher mean SYNTAX scores (27 ± 9 vs. 18 ± 8, p < 0.001), reflective of more complex CAD. However, PCItreated patients were more likely to have had prior PCI (40% vs. 30%, p = 0.001), prior myocardial infarction (32% vs. 25%, p = 0.007) and a history of congestive heart failure (12% vs. 8%, p = 0.046), than those treated with CABG.

Estimated glomerular filtration rate
With all five formulas, only a minority of patients had severe renal impairment or kidney failure. The prevalence of patients with renal failure (eGFR < 60 ml/min/ 1.73m 2 ) at baseline was 12.9% by the Mayo equation, 20.9% by MDRD, 22.8% by CG, 22.9% by IB and 24.6% by the CKD-EPI equation.
The mean eGFR values on admission were in the mild renal dysfunction range for both CABG and PCI patients with no differences in eGFR between the two groups (Fig. S2). For both CABG and PCI patients, the Mayo formula yielded the highest mean value (89.8 ± 27.1 and 90.6 ± 25.4 mL/min per 1.73m 2 , p = 0.623; respectively) and CKD-EPI the lowest (75.4 ± 24.7 and 76.6 ± 23.4 mL/ min per 1.73m 2 , p = 0.402; respectively). Notably, the Mayo included more patients with normal renal function (N = 625, 56%) and thus fewer with mild (N = 343, 31%) and moderate renal dysfunction (N = 103, 9%) (Fig. S2, Table 3). This finding was consistent in both the CABG and PCI subgroups.
In order to determine whether the CG method, as the reference formula, and the Mayo, CKD-EPI, MDRD, and IB formulas yielded the same results, we calculated the correlation coefficient, and found a strong correlation between the CG formula and each of the other four formulas (Mayo: r = 0.80, p < 0.001; CKD-EPI: r = 0.87, p < 0.001; MDRD: r = 0.84, p < 0.001; IB: r = 0.99, p < 0.001). Furthermore, in order to assess the agreement between the values obtained from the CG formula and those from each of the other formulas we used the Bland and Altman analyses, which showed good agreement with all formulas used (Fig. 1a-d). The mean ± SD of the eGFR difference between the CG formula (reference) and the Mayo, CKD-EPI, MDRD, and IB formulas were: 2 ± 21.7, 12.2 ± 19.2, 5.4 ± 19.9, and 1.8 ± 4.1 mL/min per 1.73m 2 , respectively.

Mortality by renal function and post-procedural acute kidney injury
Post-procedural acute kidney injury was more prevalent in the CABG than the PCI group (5.8% vs. 0.9%, p < 0.001), and more in the eGFR ≤60 mL/min per 1.73m 2 than the discordant and eGFR > 60 groups (9.7% vs. 3.9% vs. 1.7%, p < 0.001).
Patients with eGFR ≤60 mL/min per 1.73m 2 by all five methods had significantly higher 30-day, 1-year and 3-  . 2a and Fig. S3). Sub-analysis among CABG and PCI separately revealed similar result (Fig. 2b-c). There were no significant differences in 3-year mortality rates between patients who underwent either CABG or PCI in all three eGFR categories (consistently low eGFR 25% vs. 29%, log-rank p = 0.879; discordant eGFR 11% vs. 14%, logrank p = 0.714; consistently high eGFR 3% vs. 6%, logrank p = 0.121; respectively). The overall 3-year mortality rates were not significantly different between the CABG and PCI groups (7.4% vs. 9.7%, log-rank p = 0.294). However, the severity of renal impairment correlated with increased mortality (Fig. 3). Multivariable analysis demonstrated that worse renal function is an independent predictor of 3-year mortality for all five formulas (Table 4). Mortality risk was increased by 16-28% for each 10-unit decrease in eGFR, using all five formulas. Other independent predictors of 3-year mortality were older age, chronic obstructive pulmonary disease, previous stroke, history of atrial fibrillation and diabetes. The ability of the five formulas to predict 3-year mortality risk was highest with the Mayo (AUC 0.78 [0.73-0.83]) and lowest with the MDRD formula (AUC 0.75 [0.70-0.80]) (p = 0.004). The p-value of Hosmer-Lemeshow test was > 0.05 for all models, indicating that they were suitable. The NRI results showed that addition of a GFR formula to the baseline model correctly reclassified approximately one-sixth of patients to a higher predicted risk group (Table 5). IDI analysis demonstrated that, when added to the basic logistic regression model, each of the five formulas improved mortality risk prediction. Among the five formulas, the Mayo had the highest additive effect on mortality prediction (Mayo: rIDI = 26.4%, p = 0.001; CG: rIDI = 16.6%, p = 0.009; IB: rIDI = 15.2%, p = 0.012; CKD-EPI: rIDI = 22.8%, p = 0.003; and MDRD: rIDI = 11.7%, p = 0.035) ( Table 5).

Discussion
This prospective multicenter national registry study describes several cardinal findings regarding evaluation of renal function in patients with multivessel CAD who are referred for revascularization in a 'real-world' setting. We have shown that: (1) patients with multivessel CAD and significant renal dysfunction demonstrate increased 3-year mortality; (2) despite the significant and strong correlation between eGFR values using all five formulas, the proportion of patients categorized into the different renal function groups varied considerably, suggesting a significant clinical impact of the GFR formula used; (3) mortality increased as renal function worsened in all renal function categories; (4) patients whose renal dysfunction status shifted from significant to nonsignificant or vice versa using the different formulas, had mortality rates that were intermediate between those with and without significant renal dysfunction. This suggests that the discordance between the formulas may result from differences in clinically relevant parameters, such as age and weight; and (5) although the MDRD formula is used extensively in clinical practice, it is less accurate in predicting mortality relative to other formulas in patients with multivessel CAD.
The 2011 European Society of Cardiology (ESC) guidelines recommended the MDRD and GC formulas for assessment of renal function in CAD patients [14]. However, according to more recent 2015 ESC guidelines, there is no recommended formula [15]. The CKD-EPI formula is considered the gold standard for eGFR by the 2012 Kidney Disease Improving Global Outcomes guidelines (KDIGO) [16]. In this study, the rate of renal impairment varied from 13 to 25% depending on the formula used, with the choice of formula influencing patient stratification into different renal function categories. We showed that despite this variability in eGFR using different formulas, renal dysfunction remained an independent predictor for mortality with all five formulas. Nevertheless, the absolute mortality rates varied for each renal function category using the different formulas.
Numerous studies have compared the five eGFR formulas outlining important observations. Lin et al. found that the MDRD is more accurate for eGFR calculation in healthy or mildly impaired renal function adults compared with the CG formula [17]. Matsushita et al. showed that the CKD-EPI equation classified fewer individuals as having renal dysfunction and more accurately categorized the risk for mortality than did the MDRD equation [18]. In contrast, however, Carter suggested that the CKD-EPI formula may actually over-estimate renal dysfunction prevalence in elderly patients and showed scarce differences between the CKD-EPI and MDRD formulas [19]. Willems et al. showed that the CG and MDRD formulas effectively predicted mortality, whereas the CKD-EPI formula did not [20]. A recent study by Jo et al.   [22]. Our analysis showed that the Mayo formula offered the most accurate mortality prediction in CAD patients undergoing revascularization, thus confirming the results of previous studies [22]. This finding can be explained by the coefficients and variables included in the equations. As both older age and lower body weight are well-known predictors of outcome in patients with CAD, it follows that a formula accounting for these factors would more accurately predict mortality. While the Mayo formula was found to accurately predict mortality in our cohort, it is less widely used, has not been validated in diverse populations and is not currently recommended by any nephrology practice guidelines. Furthermore, previous reports have shown that the Mayo equation proved inaccurate in type-2 diabetic patients with hyperfiltration [23] or normal renal function [24] and in the very elderly [25,26].
Interestingly, we have shown that in a 'real-world' setting the choice of GFR, as reflected by the different categories (consistently low, discordant and consistently high eGFR), did not affect decision-making regarding the revascularization approach. Indeed, we did not detect any significant difference in the 3-year all-cause mortality by the different revascularization strategies.

Limitations
This study was based on a national prospective observational registry of patients admitted for multivessel CAD. As such, bias in treatment decision cannot be excluded. Another limitation is the lack of available data on renal dysfunction duration. While our results were adjusted for possible confounding variables, residual confounding cannot be excluded, and the lack of adjustment for variables not captured in the registry may represent a limitation. Data regarding other outcomes such as recurrent myocardial infarction, recurrent revascularization and renal function during the follow-up period were unavailable. Analysis of cardiac events could reinforce the conclusion that significant renal dysfunction upon admission is associated with late cardiac events. The covariates included in the model were: age, gender, diabetes, congestive heart failure, history of CVA/TIA, SYNTAX score, atrial fibrillation, hypertension, COPD and post-procedure acute kidney injury