Author + information
- Received March 6, 2015
- Revision received April 12, 2015
- Accepted April 20, 2015
- Published online June 30, 2015.
- Jessica Parsh, MD∗,
- Milan Seth, MS†,
- Herbert Aronow, MD‡,
- Simon Dixon, MBChB§,
- Michael Heung, MD‖,
- Roxana Mehran, MD¶ and
- Hitinder S. Gurm, MD†,#∗ ()
- ∗Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan
- †Department of Internal Medicine, Division of Cardiovascular Medicine, University of Michigan, Ann Arbor, Michigan
- ‡Michigan Heart and Vascular Institute, St. Joseph Mercy Hospital, Ann Arbor, Michigan
- §Department of Cardiovascular Medicine, Beaumont Hospital, Royal Oak, Michigan
- ‖Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan
- ¶The Zena and Michael A. Wiener Cardiovascular Institute, Mount Sinai Medical Center, New York, New York
- #VA Ann Arbor Healthcare System, Ann Arbor, Michigan
- ↵∗Reprint requests and correspondence:
Dr. Hitinder S. Gurm, University of Michigan Cardiovascular Center, 2A394, 1500 East Medical Center Drive, Ann Arbor, Michigan 48109-5853.
Background Multiple equations exist to estimate glomerular filtration rate (GFR); however, there is no consensus on which is superior for risk classification in patients with chronic kidney disease (CKD) undergoing percutaneous coronary intervention (PCI).
Objectives The goals of this study were to identify which equation to estimate GFR is superior for predicting adverse outcomes after PCI and to examine how equation selection would impact drug-dosing recommendations.
Methods Estimated GFR (eGFR) was calculated with the Cockcroft-Gault, Modification of Diet in Renal Disease Study (MDRD), and Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations for 128,805 patients undergoing PCI in the state of Michigan. Agreement between patient pre-PCI eGFR estimates and resultant CKD stage classifications, their ability to discriminate post-procedural in-hospital clinical outcomes, and the impact of equation choice on dosing recommendations for commonly used antiplatelet and antithrombotic medications were investigated.
Results CKD-EPI best discriminated post-PCI mortality by receiver operator characteristic analysis. There was wide variability in eGFR, which persisted after grouping by CKD stages. Reclassification by CKD-EPI resulted in net reclassification index improvement for acute kidney injury and new requirement for dialysis. Equation choice affected drug-dosing recommendations, with the formulas agreeing for only 50.3%, 40.0%, and 34.3% of potentially impacted patients for eGFR cutoffs of <60, <50, and <30 ml/min/1.73 m2, respectively.
Conclusions Different eGFR equations result in CKD stage reclassification that has major clinical implications for predicting adverse outcomes after PCI and drug-dosing recommendations. Our results support the use of CKD-EPI for risk stratification among patients undergoing PCI.
For patients with acute coronary syndrome who undergo percutaneous coronary intervention (PCI), chronic kidney disease (CKD) and advanced age are associated with an increased risk of adverse outcomes, including in-hospital mortality, bleeding, and acute kidney injury (1–3). Estimated glomerular filtration rate (eGFR), the most common method used to diagnose and stage CKD, can be calculated by several equations, including the Cockcroft-Gault equation, the Modification of Diet in Renal Disease (MDRD) Study equation, and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation (4–6). The Berlin Initiative Study (BIS1) equation was developed recently in adults ≥70 years of age in an attempt to more accurately predict eGFR in this subgroup (7). The 2012 Kidney Disease Improving Global Outcomes (KDIGO) guidelines recommend that clinicians use CKD-EPI because of its superior accuracy (8). The Cockcroft-Gault equation is no longer recommended because it was developed before the standardization of creatinine assays and is less accurate (9); however, the preference for which estimate to use varies widely among institutions, and the Cockcroft-Gault equation is still often used in clinical practice and in pharmacokinetic studies (10).
Although others have shown that the use of different equations can result in CKD stage reclassification, it is unclear how this would impact prognostication for patients undergoing PCI. We investigated whether the use of different eGFR equations would result in different CKD staging of patients undergoing PCI and examined how reclassification correlates with risk of adverse events after PCI. We then extended our analysis to determine how differences in CKD classification would affect the dosing of commonly used antiplatelet and antithrombotic agents in the catheterization laboratory.
We performed a retrospective post-hoc analysis using data from the Blue Cross Blue Shield of Michigan Cardiovascular Consortium, a regional registry of all patients undergoing PCI at nonfederal hospitals in Michigan. A detailed outline of the registry has been described previously (11). To summarize, this is a prospective, multicenter, statewide registry of patients undergoing PCI at 47 participating centers. For our present study, consecutive patients undergoing emergent or elective PCI between January 2010 and March 2014 were included. We excluded all patients who required hemodialysis before PCI, those with incomplete data on serum creatinine levels before or after PCI, and those without both a height and weight recorded. Pre-procedural serum creatinine values were measured within 30 days before PCI, with the value closest to the time of PCI chosen as the baseline value. Peak post-procedural creatinine was defined as the highest value after PCI and before the next procedure or discharge.
Estimation of glomerular filtration rate
For our population of all-comers, we calculated eGFR using the Cockcroft-Gault, MDRD, and CKD-EPI equations. Each equation is provided with a summary of its developmental cohort in Table 1. The Cockcroft-Gault equation calculates creatinine clearance and not eGFR; however, its output has been compared in the literature with the eGFR of other equations, both with and without adjustment for body surface area (12,13). We performed our analysis with the unadjusted Cockcroft-Gault output and an additional analysis with Cockcroft-Gault adjusted for body surface area by normalizing the output per 1.73 m2 of body surface area (identical to the normalization of the glomerular filtration rate [GFR] measurement) (14). Additionally, because of unclear recommendations for body weight, estimation of eGFR by Cockcroft-Gault was calculated twice, with both actual and ideal body weight (15). For analysis of the subgroup of patients ≥70 years of age, eGFR was also calculated by the BIS1 equation.
The diagnostic accuracy of candidate eGFR estimates was evaluated with respect to 4 in-hospital clinical outcomes, including the primary endpoint of acute kidney injury (AKI) and secondary endpoints of new requirement for dialysis (NRD), in-hospital mortality, and receipt of transfusion. AKI was defined as a ≥0.5 mg/dl increase in absolute serum creatinine from the baseline pre-procedural value (16,17). NRD was defined as any new, unplanned need for dialysis after PCI. In-hospital mortality was defined as mortality attributable to any cause during the initial hospitalization after PCI. Receipt of transfusion was defined as the transfusion of whole blood or packed red blood cells from the time of PCI to before discharge or death.
Statistical analysis and data visualization
Scatterplots comparing the eGFRs were used to visualize disagreement between eGFR estimates at the patient level. Estimated eGFRs values were winsorized (censored) at 200 ml/min for convenience in graphic representation and to prevent the undue influence of large-value outliers on the analysis. Lin’s concordance correlation coefficient was used to assess agreement between estimates, with standard errors estimated with an assumption of asymptotic normality after Fisher Z-transformation (18). Receiver operating characteristic analysis was used to assess the diagnostic discrimination of candidate eGFR estimates with respect to in-hospital outcomes, with standard errors for the area under the receiving operating characteristic estimates obtained by the Delong method (19).
Patients were assigned to CKD stages based on each of the eGFR estimates with the use of KDIGO guidelines, with stage 1, 2, 3a, 3b, 4, and 5 defined by eGFR ≥90, 60 to 89, 45 to 59, 30 to 44, 15 to 29, and <15 ml/min/1.73 m2, respectively (8). Confusion matrices were used to determine the extent of agreement in CKD staging by different eGFR estimates. Agreement was assessed with Cohen’s kappa without weighting. Net reclassification index (NRI) statistics were used to compare the diagnostic accuracy of CKD stage estimates between the eGFR candidates with respect to in-hospital outcomes (20).
To assess the extent to which candidate eGFR estimates could potentially impact drug-dosing recommendations, we constructed Venn diagrams to compare how many patients would meet the U.S. Food and Drug Administration–recommended eGFR drug-dose–reduction cutoffs of <60, <50, or <30 ml/min/1.73 m2 for tirofiban, eptifibatide, and bivalirudin, respectively (21). R version 3.0.3 was used for this analysis (22).
Between 2010 and 2014, a total of 135,462 patients underwent PCI at institutions participating in the Blue Cross Blue Shield of Michigan Cardiovascular Consortium. Of these patients, 3,183 were missing pre-procedural serum creatinine values, 3,357 were undergoing hemodialysis before PCI, and 189 were missing 1 or more variables required to estimate GFR (age, weight, or height). A total of 6,657 patients met 1 or more exclusion criteria. The baseline demographic and clinical characteristics of the remaining 128,805 patients included in our overall cohort, as well as the elderly subgroup, are presented in Table 2. AKI occurred in 3.24% of patients, whereas NRD occurred in 0.35%. A total of 1,790 patients (1.39%) died before discharge. Transfusion was required for 2.98% of patients. In the elderly subgroup, rates of all 4 adverse outcomes were increased (Table 2). There was significant variability in the predicted eGFR calculated with the different equations (Figure 1), with median eGFRs of 89.0, 73.0, and 76.7 ml/min/1.73 m2 for the Cockcroft-Gault, MDRD, and CKD-EPI equations, respectively.
Receiver operating characteristic analysis
Results of the receiver operating characteristic analysis for the 4 endpoints can be found in Table 3. For AKI, CKD-EPI produced an area under the curve of 0.741, which was significantly greater than that produced with the MDRD and Cockcroft-Gault equations. Receiver operating characteristic analysis for mortality demonstrated an area under the curve of 0.734 for CKD-EPI, which was not significantly different from that of Cockcroft-Gault; however, it was significantly greater than that of MDRD. The area under the curve of CKD-EPI for NRD was not significantly different from MDRD but was significantly greater than that of Cockcroft-Gault. For the endpoint of blood product transfusions, Cockcroft-Gault had the largest area under the curve (0.713), which was significantly greater than that of MDRD and CKD-EPI.
CKD stage reclassification by eGFR categories and prediction of risk
The eGFR calculated by the MDRD and CKD-EPI equations classified patients as being in the same CKD stage 84.69% of the time (Figure 2). Of the 19,720 patients who were reclassified by CKD-EPI compared with MDRD, 11.64% were reclassified to a lower CKD stage (higher eGFR category) and 3.66% to a higher CKD stage (lower eGFR category). The results for the comparison of CKD-EPI and Cockcroft-Gault found a much lower percentage agreement (56.01%), and in this case, the vast majority of reclassification by CKD-EPI placed patients in a higher-severity CKD stage (34.14% in a lower eGFR category and 9.85% in a higher eGFR category). When Cockcroft-Gault was adjusted for body surface area, agreement in classification improved (71.02%), but in this case, the reclassifications to lower- or higher-severity CKD stages were similar (15.88% in a higher eGFR category and 13.09% in a lower eGFR category).
In the elderly subgroup, the highest agreement in CKD classification by eGFR also occurred in the CKD-EPI and MDRD comparison (Online Figure 1), but the majority of reclassification by CKD-EPI moved patients to a higher-severity CKD stage. Classifications by BIS1 and CKD-EPI were in agreement 65.96% of the time, with the majority of reclassification by CKD-EPI placing patients in a lower CKD stage (32.95%). Again, agreement between CKD-EPI and Cockcroft-Gault was lowest.
In the general population cohort, reclassification to a higher or lower CKD stage by CKD-EPI was found to more appropriately predict risk of AKI and NRD. This is demonstrated by the statistically significant negative NRIs for reclassification by MDRD and Cockcroft-Gault compared with CKD-EPI (Figure 3). For reclassification by MDRD, the NRI was negative in a statistically significant fashion for AKI (−6.21; 95% confidence interval [CI]: −5.11 to −7.31) and NRD (−6.67; 95% CI: −3.85 to −9.50). Cockcroft-Gault performed similarly. Reclassification by MDRD also resulted in a negative NRI for mortality and need for transfusion; however, Cockcroft-Gault had a significantly positive NRI for mortality (9.71; 95% CI: 6.36 to 13.05) and transfusion (12.80; 95% CI: 10.52 to 15.07). When Cockcroft-Gault was adjusted for body surface area, the NRI results remained significantly positive for mortality (5.03; 95% CI: 2.17 to 7.89) and need for transfusion (5.40; 95% CI: 3.44 to 7.36). For the comparison of classification with 2 weight definitions, Cockcroft-Gault (ideal body weight) had a significantly negative NRI for mortality but was significantly positive for AKI compared with Cockcroft-Gault using actual body weight. In the elderly subgroup, reclassification of CKD stage by BIS1 compared with CKD-EPI resulted in negative NRIs for all outcomes (Online Figure 2).
Various estimations for GFR impact on drug dosing
Of the 128,805 patients in our general population cohort, 41,483 (32.2%), 24,659 (19.1%), and 4,261 (3.3%) were estimated by at least 1 equation to have an eGFR <60, <50, and <30 ml/min/1.73 m2, respectively; however, the degree to which all 3 equations were in agreement on this classification decreased as the eGFR cutoff value decreased. The 3 equations were in agreement on the classification of 50.3% of patients to an eGFR <60 ml/min/1.73 m2, whereas agreement occurred for 45.0% and 34.3% of patients for the cutoffs of <50 and <30 ml/min/1.73 m2, respectively. The distributions of patients with a low eGFR, as classified by 1 or more methods, are described in the Venn diagrams (Central Illustration). When the Cockcroft-Gault eGFR was adjusted for body surface area, concordance improved; however, the 3 equations still only agreed on drug-dose adjustment in 62.9%, 56.2%, and 45.0% of patients for eGFR cutoffs of <60, <50, and <30 ml/min/1.73 m2, respectively (Online Figure 3).
The principal finding of our study is that there is wide variation in the estimation of GFR among the main equations used today, which leads to CKD stage reclassification in a large proportion of patients and can have important implications regarding risk for major adverse outcomes after PCI. Additionally, calculation of eGFR by the various equations resulted in large discrepancies in drug-dosing recommendations for commonly used antiplatelet and antithrombotic agents.
Our work expands on and integrates the extensive literature regarding the clinical implications of CKD staging on prediction of overall cardiovascular risk and risk of adverse outcomes after PCI with the studies comparing the accuracy and risk prediction of eGFR equations. Multiple studies have shown that cardiovascular risk increases with decreasing eGFR (23,24). In patients undergoing PCI, CKD is an independent risk factor for all-cause mortality, bleeding, and contrast-induced nephropathy, with the risk for adverse events increasing with decreasing eGFR (1,2). Therefore, accurate staging of CKD is of paramount importance to the assessment of a patient’s risk for adverse events with PCI. Our study demonstrates that the selection of the eGFR equation can result in significant CKD restaging, with disagreement as high as 45% between CKD-EPI and Cockcroft-Gault and as high as 15% between the 2 currently recommended equations (CKD-EPI and MDRD). When CKD stage is used as a prognostic indicator, the clinician must consider this discrepancy when selecting an eGFR equation and appreciate how risk assessment for an individual patient may hinge on this decision.
Interestingly, although the BIS1 equation was developed in the elderly and was shown to be more accurate in its original cohort, our results do not support its use in the elderly subpopulation given the negative NRI values for all adverse events. This discrepancy may be secondary to the development of BIS1 in a relatively healthier patient cohort (14.9% with previous myocardial infarction vs. 36% in our study) and the equation having not yet been validated externally.
Currently, the KDIGO guidelines recommend the use of the CKD-EPI equation because it has been found to be more accurate, more precise, and less biased than the MDRD equation (6,8). The Cockcroft-Gault equation is not recommended because it was developed before standardization of creatinine assays (8). However, regardless of these guidelines, a recent survey demonstrated that there is still wide variation among institutions providing an automatic eGFR from serum creatinine values, with 83% of reporting laboratories using the MDRD, 4% using Cockcroft-Gault, 2% using another equation, and 12% of responders being “unsure” of the equation used at their institution (10). Furthermore, although physicians may rely on these automatically reported eGFR values, pharmacists frequently recalculate renal function using an alternative equation, often the Cockcroft-Gault. The fact that providers use different equations to estimate renal function or are unaware of the equation they are using increases the potential for error when establishing the dose for medications that could potentially cause serious harm in patients with decreased renal function.
In regard to renally dosed medications, the eGFR equation selected to determine drug dosing in the clinic ideally would be identical to the choice of equation used in the drug’s original pharmacokinetic study; however, the U.S. Food and Drug Administration currently has no clear guidelines for equation selection in the pharmacokinetic studies of novel medications (personal communication, FDA Division of Drug Information, August 25, 2014) (25). Instead, the choice of equation is determined per drug company discretion, and many pharmacokinetic studies use the Cockcroft-Gault equation. Since 2010, the U.S. Food and Drug Administration has made a greater attempt to include information on which equations are being used, but this information is not always readily available. To ensure appropriate dosage reductions, our study results emphasize the importance of clear documentation of the choice of an eGFR equation in all pharmacokinetic studies and determination of individual drug-dosing recommendations with that same equation. Given the striking disagreements in eGFR between the various equations and the lack of consistent documentation of pharmacokinetic study design, our results may even argue that it would benefit our patients to enact a uniform equation recommendation for all new drug development. Because multiple studies have demonstrated the improved accuracy of the CKD-EPI equation for predicting true GFR, and because it demonstrates better reclassification for the renally based outcomes of NRD and AKI, we would argue for use of the CKD-EPI equation as the standard for pharmacokinetic studies.
First, our study’s purpose was not to identify which equation most accurately predicts true GFR, but rather which stratifies our patients in a way that best correlates with the risk of adverse outcomes after PCI. Therefore, we had no gold standard of true GFR with which to compare our estimated GFRs for each patient. Multiple other studies have investigated the accuracy, precision, and bias of each equation, and we took their conclusions into account in our final recommendations for the use of CKD-EPI (6,8). In addition, there are multiple definitions for AKI, and we would expect the chosen definition to affect our results. We selected the definition of a serum creatinine increase >0.5 mg/dl because we and others have shown its superiority for predicting adverse events in patients undergoing PCI (16,17).
Our study demonstrates that not only is there great heterogeneity in eGFR values calculated by the most widely used equations today, but this variability affects risk classification for adverse events with PCI. Our data support the use of the CKD-EPI equation in the evaluation of patients before PCI, given its improved accuracy and risk classification for AKI and NRD. Finally, our data suggest that transparency and standardization for the estimation of GFR for pharmacokinetic studies could potentially lead to improved dosage recommendations and better outcomes for our patients.
COMPETENCY IN MEDICAL KNOWLEDGE: Estimates of glomerular filtration rate vary depending on the equation used, which results in clinically important differences in dose calculations for antiplatelet and antithrombotic drugs administered to patients with chronic kidney disease undergoing percutaneous coronary interventions.
TRANSLATIONAL OUTLOOK: Prospective studies are needed to determine the optimal method for estimation of renal clearance of drugs used in patients with chronic kidney disease undergoing cardiovascular procedures.
The authors are indebted to all the study coordinators, investigators, and patients who participated in the Blue Cross Blue Shield of Michigan Cardiovascular Consortium registry.
For supplemental figures, please see the online version of this article.
Dr. Mehran has received research grants from Eli Lilly, AstraZeneca, The Medicines Company, Bristol-Myers Squibb, and Sanofi; has served as a consultant for AstraZeneca, Bayer, CSL Behring, Janssen Pharmaceuticals Inc., Merck & Co., Osprey Medical Inc., and Watermark Research Partners; and has served on scientific advisory boards for Abbott Laboratories, Boston Scientific, Covidien, Janssen Pharmaceuticals, The Medicines Company, and Sanofi. Dr. Gurm has served as a consultant for Osprey Medical Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. Peter McCullough, MD, MPH, served as the Guest Editor for this article.
- Abbreviations and Acronyms
- acute kidney injury
- Berlin Initiative Study
- chronic kidney disease
- Chronic Kidney Disease Epidemiology Collaboration
- estimated glomerular filtration rate
- glomerular filtration rate
- Kidney Disease Improving Global Outcomes
- Modification of Diet in Renal Disease
- new requirement for dialysis
- net reclassification index
- percutaneous coronary intervention
- Received March 6, 2015.
- Revision received April 12, 2015.
- Accepted April 20, 2015.
- American College of Cardiology Foundation
- Hanna E.B.,
- Chen A.Y.,
- Roe M.T.,
- Wiviott S.D.,
- Fox C.S.,
- Saucedo J.R.
- Tsai T.T.,
- Patel U.D.,
- Chang T.I.,
- et al.
- Thomas M.P.,
- Moscucci M.,
- Smith D.E.,
- et al.,
- Blue Cross Blue Shield of Michigan Cardiovascular Consortium
- ↵Miller G. Current status of reporting estimated glomerular filtration rate (eGFR). College of American Pathologists website. Available at: http://www.cap.org/apps/docs/committees/chemistry/current_status_of_reporting_eGFR.pdf. Accessed April 21, 2015.
- Gurm H.S.,
- Smith D.E.,
- Collins J.S.,
- et al.,
- Blue Cross Blue Shield of Michigan Cardiovascular Consortium (BMC2)
- Szummer K.,
- Lundman P.,
- Jacobson S.H.,
- et al.,
- for SWEDEHEART
- Michels W.M.,
- Grootendorst D.C.,
- Verduijn M.,
- Elliott E.G.,
- Dekker F.W.,
- Krediet R.T.
- Du Bois D.,
- Du Bois E.F.
- Devine B.J.
- ↵DailyMed. Drug label archives. NIH U.S. National Library of Medicine website. Available at: http://dailymed.Nlm.Nih.Gov/dailymed/index.Cfm. Accessed April 21, 2015.
- ↵R Foundation for Statistical Computing website. The R Project for Statistical Computing. Available at: http://www.R-project.org/. Accessed April 21, 2015.
- Chronic Kidney Disease Prognosis Consortium,
- Matsushita K.,
- van der Velde M.,
- Astor B.C.,
- et al.
- ↵Renal Impairment Guidance Working Group. Guidance for industry: pharmacokinetics in patients with impaired renal function: study design, data analysis, and impact on dosing and labeling. Center for Drug Evaluation and Research (CDER) U.S. Department of Health and Human Services Food and Drug Administration. March 2010. Available at: http://www.fda.gov/downloads/drugs/guidancecomplianceregulatoryinformation/guidances/ucm204959.pdf. Accessed April 21, 2015.