ABSTRACT
Background
In 2021, an updated Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation for estimated glomerular filtration rate (eGFR) without a coefficient for race (CKD-EPI21) was developed. The performance of this new equation has yet to be examined among specific patient groups.
Methods
We compared the performances of the new CKD-EPI21 equation and the 2009 equation assuming non-Black race (CKD-EPI09-NB) in patients with GFR measured by chromium-51-EDTA plasma clearance at Aarhus University Hospital in Denmark during 2010–18. We examined bias, accuracy, precision and correct classification of chronic kidney disease (CKD) stage using chromium-51-EDTA clearance as the reference standard. We assessed the performance in the total cohort, cancer patients and potential living kidney donors. We also assessed the performance stratified by CKD stage in the total cohort.
Results
In this predominantly white population, the CKD-EPI21 equation performed slightly better than the CKD-EPI09-NB equation in both the total cohort (N = 4668), and in cancer patients (N = 3313) and potential living kidney donors (N = 239). In the total cohort, the CKD-EPI21 equation demonstrated a slightly lower median absolute bias (−0.2 versus −4.4 mL/min/1.73 m2), and a similar accuracy, precision and correct classification of CKD stage compared with the CKD-EPI09-NB equation. When stratified by CKD stage, the CKD-EPI09-NB equation performed slightly better than the CKD-EPI21 equation among patients with a measured GFR (mGFR) <60 mL/min/1.73 m2.
Conclusions
In a selected cohort of Danish patients with mGFR, the CKD-EPI21 equation performed slightly better than the CKD-EPI09-NB equation except for patients with a mGFR <60 mL/min/1.73 m2, where CKD-EPI09-NB performed slightly better although the differences were considered clinically insignificant.
Keywords: CKD-EPI equation, GFR, race
KEY LEARNING POINTS.
What was known:
An updated Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation for estimated glomerular filtration rate (eGFR) without a coefficient for race (CKD-EPI21) was developed and published in 2021.
This study adds:
In this study of predominantly white patients with GFR measured by chromium-51-EDTA plasma clearance, the overall performance of the CKD-EPI21 equation was slightly better than the original 2009 CKD-EPI equation assuming non-Black race (CKD-EPI09-NB), both overall and in cohorts of cancer patients and potential living kidney donors.
However, the CKD-EPI09-NB equation performed slightly better than the CKD-EPI21 equation for patients with a measured GFR <60 mL/min/1.73 m2.
Potential impact:
There were slight differences in the performance of CKD-EPI21 and CKD-EPI09-NB depending on the kidney function; however, these differences may not be clinically significant.
INTRODUCTION
Accurate estimation of glomerular filtration rates (GFRs) is pivotal, as the estimated GFR (eGFR) is recommended to identify and monitor patients with chronic kidney disease (CKD) and to guide adjustments to medications and contrast medium [1]. The first version of the widely used Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-based eGFR equation developed in 2009 (CKD-EPI09) based on patients included in American studies, included the variables serum creatinine, age, sex and race (Black versus non-Black) [2]. A new creatinine-based eGFR equation was proposed in 2021 that excludes the race term (CKD-EPI21) [3]. The authors concluded that the new equation is sufficiently accurate to be implemented in clinical practice. This led the National Kidney Foundation and American Society of Nephrology to recommend the use of this new equation in the USA [4]. However, Inker et al. [3] speculated that the new equation could lead to systematic differences in GFR estimation between race groups. This was confirmed by a recent study showing that exclusion of the race term would lead to fewer Black patients receiving the full dose of anticancer treatments, as more Black patients have an eGFR below kidney function cutoffs when using the CKD-EPI21 equation compared with the CKD-EPI09 equation [5]. Furthermore, in predominantly non-Black populations, such as Scandinavia [6, 7], the introduction of the CKD-EPI21 equation is observed to decrease the prevalence of CKD using eGFR by approximately 25%. Thus, the implementation of this new formula should be supported by studies establishing its accuracy compared with measured GFR (mGFR).
Some studies have examined the performance of the CKD-EPI21 equation in specific patient groups [8–14]. To further validate the new CKD-EPI21 equation, we examined its performance in a Danish and predominantly white population, including cohorts of cancer patients and potential living kidney donors, and compared the new equation with the original CKD-EPI09 equation considering all patients as non-Black (CKD-EPI09-NB). We assessed bias, precision, accuracy and the ability to categorize patients correctly according to CKD stage comparing the eGFR with mGFR using chromium-51-EDTA (51Cr-EDTA) plasma clearance.
MATERIALS AND METHODS
Setting and data sources
We conducted this cross-sectional study using data from Danish medical databases. Denmark's National Health Service provides tax-supported healthcare to the Danish population, ensuring free access to general practitioners and hospitals [15]. All Danish residents are assigned a unique personal identifier that permits individual-level linkage among Danish registries, including the Danish National Patient Registry (patient registry) [16], laboratory databases providing data on creatinine measurements from general practices and hospitals [17, 18], and the Danish Civil Registration System [19]. The creatinine was measured based on isotope dilution mass spectrometry (IDMS)-traceable creatinine enzymatic assays. The mGFR was calculated from the plasma 51Cr-EDTA clearance determined at Aarhus University Hospital, primarily for clinical purposes (e.g. before initiation of anticancer drug treatment or kidney donation). The 51Cr-EDTA was administered intravenously followed by oral hydration. Plasma 51Cr-EDTA was evaluated at 180, 200, 220 and 240 min after injection if the eGFR was >40 mL/min/1.73 m2, at 180, 210, 240, 270 and 300 min after injection if the eGFR was 20–40 mL/min/1.73 m2, and at 180, 210, 240, 270 and 300 min and 24 h after injection if the eGFR was <20 mL/min/1.73 m2 [20, 21]. The mGFR was modeled using the Brochner Mortensen method. The body surface area was calculated using the DuBois equation. We did not have access to direct information about indications for the mGFR, but we used the patient registry to identify the most likely indication (see below).
Study cohort
We identified patients with an mGFR from 1 January 2010 to 31 December 2018 and included the first recorded measurement for each patient. We excluded patients without a creatinine measurement in the laboratory database within the 3 months prior to the date of the mGFR, patients receiving dialysis within the year prior to the date of GFR measurement as recorded in the patient registry and patients <18 years of age (Fig. 1). We identified three study cohorts for analysis: a total cohort consisting of all eligible patients, a cancer cohort consisting of patients with an inpatient or outpatient cancer diagnosis in the patient registry within the 3 months prior to the date of the mGFR, and a kidney donor cohort consisting of non-cancer patients with an outpatient diagnosis code for potential living kidney donation in the patient registry within the 3 months prior to the date of the mGFR or with a surgery code for living kidney donation in the patient registry within 1 year after the mGFR (see the Codebook in the Supplementary data for codes). The most recent creatinine measurement prior to the date of mGFR was used to estimate GFR (see Box 1 for definitions of the applied equations).
Figure 1:
Flowchart of cohort construction.
Box 1. Applied definitions of eGFR equations.
-
(i) CKD-EPI21 = 142 × min(Scr/κ, L)α × max(Scr/κ, L)−1.200 × 0.9938Age × 1.012 [if female], where:
Scr = standardized serum creatinine in mg/dL
κ = 0.7 (female) or 0.9 (male)
α = −0.241 (female) or −0.302 (male)
(ii) CKD-EPI09-NB = 141 × min(SCr/k, L)α × max(SCr/k, L)−1.209 × 0.993Age × 1.018 [if female], where:
Scr = standardized serum creatinine in mg/dL
κ = 0.7 (female) or 0.9 (male)
α = −0.329 (female) or −0.411 (male)
Covariates
We included the following covariates: age and sex obtained from the Civil Registration System; morbidities obtained from the patient registry, including diabetes, heart failure, liver disease and lung disease; the presumed indication for mGFR, including candidacy for a kidney, liver, lung or heart transplant, congenital malformation of the urinary system, benign tumor in the urinary system, palsy, and psoriasis or atopic dermatitis possibly related to administration of calcineurin inhibitors; and CKD identified from the laboratory databases, defined as two outpatient eGFR values <60 mL/min/1.73 m2, according to either CKD-EPI21 or CKD-EPI09-NB, ≥90 days apart (see the Codebook in the Supplementary data for codes, definitions, and look-back and look-forward periods).
Statistical analysis
Descriptive analyses
We described the prevalence of the following conditions that were considered potential indications for mGFR: cancer, candidacy as a potential living kidney donor, congenital malformation of the urinary system, benign tumor in the urinary system, CKD, candidacy for a kidney, liver, heart or lung transplant, palsy, psoriasis and atopic dermatitis. We reported the number of patients and their distribution of age, sex, creatinine level, mGFR and year of mGFR, as well as the prevalence of diabetes, cancer, CKD, heart failure, liver diseas, and lung disease within 10 years before the mGFR (see the Codebook in the Supplementary data for codes and definitions). To investigate the time from the eGFR determination to the mGFR, we calculated the median time from eGFR to mGFR in the three cohorts.
Performance
We assessed the performance of the CKD-EPI09-NB and CKD-EPI21 equation in each of the three cohorts. We examined bias by calculating the absolute bias (i.e. the difference between eGFR and mGFR) and the relative bias (i.e. the difference between eGFR and mGFR divided by the mGFR). Precision was defined as the interquartile range (IQR) of the absolute bias. We assessed accuracy by calculating the percentage of patients with eGFR within the range of ±30% of the mGFR (P30). We plotted eGFR and the difference between the eGFR and mGFR against the mGFR. We then assessed bias, precision and accuracy in the three cohorts stratified by no CKD, or CKD stages 1/2, CKD stage 3 and CKD stages 4/5 based on mGFR (≥60, 30–59 and <30 mL/min/1.73 m2). Finally, we calculated the percentage of patients correctly classified into the three CKD stages.
Additional analyses
We repeated the analysis of bias, precision and accuracy for the new European Kidney Function Consortium (EKFC) eGFR equation [22], while stratifying by sex and age (</≥65 years), and while restricting to patients with an eGFR available within 5 days prior to the mGFR.
RESULTS
We identified 5639 patients with an mGFR. After excluding 398 lacking a creatinine measurement, 133 in dialysis within 1 year before the mGFR and 420 aged <18 years, we included 4688 patients in the total cohort (Fig. 1). Based on the presumed indication for mGFR, the majority (3313 patients) were included in the cancer cohort, whereas 239 patients were included in the kidney donor cohort (Supplementary data, Table S1). Individuals in the kidney donor cohort were younger, and had fewer comorbidities and higher mGFR than the total and cancer cohorts (Table 1).
Table 1:
Characteristics of the study cohorts.
| Total cohort (N = 4688) | Cancer cohort (N = 3313) | Kidney donor cohort (N = 239) | Non-cancer, non-kidney donor cohort (N = 1136) | |
|---|---|---|---|---|
| Age, median (IQR), years | 64 (54–72) | 66 (57–73) | 54 (45–63) | 58 (45–69) |
| Age <65 years | 2446 (52) | 1504 (45) | 192 (80) | 750 (66) |
| Female sex | 2219 (47) | 1688 (51) | 140 (59) | 391 (34) |
| mGFR, median (IQR), mL/min/1.73 m2 | 84 (63–105) | 86 (67–106) | 101 (83–116) | 73 (49–99) |
| Creatinine, median (IQR), μmol/L | 75 (62–96) | 72 (60–89) | 71 (62–81) | 92 (73–132) |
| Creatinine, median (IQR), μmol/L, women | 65 (56–79) | 64 (55–76) | 65 (59–73) | 75 (61–110) |
| Creatinine, median (IQR), μmol/L, men | 85 (72–108) | 81 (69–98) | 79 (74–88) | 100 (80–145) |
| Days from creatinine test to GFR, median (IQR) | 5 (1–13) | 6 (2–13) | 3 (1–6) | 3 (0–15) |
| Patients with an eGFR available within 5 days before the mGFR | 2474 (53) | 1649 (50) | 176 (74) | 649 (57) |
| Year of mGFR measurement | ||||
| 2010–12 | 470 (10) | 160 (5) | 74 (31) | 236 (21) |
| 2013–15 | 797 (17) | 408 (12) | 86 (36) | 303 (27) |
| 2016–18 | 3421 (73) | 2745 (83) | 79 (33) | 597 (53) |
| Diabetes | 422 (9) | 267 (8) | 0 (0) | 155 (14) |
| Cancera | 3550 (76) | 3313 (100) | <5c | Maskedc |
| Chronic kidney disease (CKD-EPI21)b | 932 (20) | 462 (14) | 10 (4) | 460 (40) |
| Chronic kidney disease (CKD-EPI09-NB)b | 1094 (23) | 592 (18) | 13 (5) | 489 (43) |
| mGFR level | ||||
| <30 mL/min/1.73 m2 | 196 (4) | 78 (2) | <5 | Maskedc (10) |
| 30–59 mL/min/1.73 m2 | 811 (17) | 508 (15) | Maskedc (4) | Maskedc (26) |
| ≥60 mL/min/1.73 m2 | 3681 (79) | 2727 (82) | Maskedc (96) | Maskedc (64) |
| Heart failure | 349 (7) | 131 (4) | 0 (0) | 218 (19) |
| Liver disease | 64 (1) | 40 (1) | <5c | Maskedc |
| Lung disease | 427 (9) | 294 (9) | <5c | Maskedc |
Values are given as N (%) unless otherwise noted.
The look-back period for the definition of cancer comorbidity was 3650 days, whereas the look-back period for the definition of the cancer cohort was 365 days.
Two outpatient eGFRs <60 mL/min/1.73 m2 ≥90 days apart within the look-back period.
Cells are masked, so it is not possible to identify or back-calculate numbers less than 5.
The median time from eGFR to mGFR was 5 days [first and third quartiles (Q1–Q3), 1–13 days] in the total cohort, 6 days (Q1–Q3, 2–13) in the cancer cohort and 3 days (Q1–Q3, 1–6) in the kidney donor cohort.
Performance
In all three cohorts, the biases were slightly closer to 0 using the CKD-EPI21 equation compared with the CKD-EPI09-NB equation, with biases being greatest in the kidney donor cohort (Table 2, Fig. 2 and Fig. 3). Precision (IQR of absolute bias) was similar for CKD-EPI21 and CKD-EPI09-NB in all three cohorts (Table 2), ranging from 24.1 to 25.7 mL/min/1.73 m2. The accuracy (P30) was markedly higher using the CKD-EPI21 equation than the CKD-EPI09-NB equation in the kidney donor cohort (88.3% versus 79.9%) and similar in the total cohort (72.4% versus 70.3%) and cancer cohort (73.4% versus 72.0%).
Table 2:
Bias, precision and accuracy of eGFR for the CKD-EPI09-NB and CKD-EPI21 equations across cohorts.
| CKD-EPI09-NB | CKD-EPI21 | ||
|---|---|---|---|
| Total cohort (N = 4688) | Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −4.4 (−17.1 to 8.0) | −0.2 (−13.2 to 12.1) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −4.7 (−5.3 to −4.1) | −0.6 (−1.2 to 0.0) | |
| Precision, mL/min/1.73 m2 | 25.1 | 25.3 | |
| Relative bias, %, median (Q1–Q3) | −5.6 (−18.8 to 11.7) | −0.3 (−14.3 to 17.9) | |
| Relative bias, %, mean (95% CI) | 1.9 (0.5–3.3) | 7.6 (6.1–9.1) | |
| Accuracy, % (95% CI) | 70.3 (69.0–71.6) | 72.4 (71.1–73.6) | |
| Cancer cohort (N = 3313) | Absolute bias, mL/min/1.73m2, median (Q1–Q3) | −4.1 (−16.8 to 8.8) | 0.5 (−12.6 to 13.1) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −4.3 (−5.0 to −3.6) | 0.0 (−0.7 to 0.7) | |
| Precision, mL/min/1.73 m2 | 25.6 | 25.7 | |
| Relative bias, %, median (Q1–Q3) | −4.8 (−17.6 to 12.2) | 0.6 (−13.1 to 18.5) | |
| Relative bias, %, mean (95% CI) | 2.6 (0.9–4.3) | 8.4 (6.6–10.3) | |
| Accuracy, % (95% CI) | 72.0 (70.4–73.5) | 73.4 (71.9–74.9) | |
| Kidney donor cohort (N = 239) | Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −9.9 (−22.0 to 2.1) | −5.8 (−17.6 to 6.6) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −9.2 (−11.6 to −6.7) | −4.9 (−7.3 to −2.5) | |
| Precision, mL/min/1.73 m2 | 24.1 | 24.2 | |
| Relative bias, %, median (Q1–Q3) | −10.6 (−20.4 to 2.2) | −5.6 (−16.7 to 7.8) | |
| Relative bias, %, mean (95% CI) | −4.6 (−10.0 to 0.8) | −0.0 (−5.6 to 5.6) | |
| Accuracy, % (95% CI) | 79.9 (74.5–84.6) | 88.3 (83.8–91.9) | |
| Non-cancer, non-kidney donor cohort (N = 1136) | Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −4.3 (−17.1 to 6.8) | −1.2 (−13.8 to 10.5) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −4.8 (−6.1 to −3.5) | −1.4 (−2.7 to −0.1) | |
| Precision, mL/min/1.73 m2 | 23.9 | 24.3 | |
| Relative bias, %, median (Q1–Q3) | −7.0 (−22.1 to 11.9) | −1.7 (−17.9 to 18.0) | |
| Relative bias, %, mean (95% CI) | 1.3 (−1.4 to 3.9) | 6.7 (4.0–9.5) | |
| Accuracy, % (95% CI) | 63 (61–66) | 66 (63–69) |
CI, confidence interval.
Absolute bias is calculated as the difference between eGFR and mGFR; relative bias is calculated as the difference between eGFR and mGFR divided by the mGFR; precision is calculated as the IQR of the bias; and the accuracy is calculated as the percentage of patients with eGFR within the range of ±30% of the mGFR (P30).
Figure 2:
Bland–Altman plots of the difference between eGFR and mGFR against the mean of estimated and measured GFR for the CKD-EPI09-NB and CKD-EPI21 equations across cohorts. CI, confidence interval.
Figure 3:
Comparison of mGFR and eGFR values across cohorts. CI, confidence interval.
Across all three cohorts, both eGFR equations overestimated the GFR at low values and underestimated GFR at high values (Fig. 2). When we stratified the total cohort by CKD stage, we observed that bias was slightly greater using the CKD-EPI21 equation compared with the CKD-EPI09-NB equation when mGFR was <60 mL/min/1.73 m2 (Table 3). When mGFR was ≥60 mL/min/1.73 m2, the reverse was seen—the bias of the CKD-EPI21 equation was slightly smaller than the bias of the CKD-EPI09-NB equation (Table 3).
Table 3:
Bias, precision and accuracy of eGFR for the CKD-EPI09-NB and CKD-EPI21 equations across CKD stages in the total cohort.
| CKD-EPI09-NB | CKD-EPI21 | |||
|---|---|---|---|---|
| Total cohort | GFR <30 ( N = 196) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | 7.7 (−0.1 to 16.5) | 9.7 (1.2–19.7) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | 11.7 (9.1–14.3) | 13.9 (11.1–16.6) | ||
| Precision, mL/min/1.73 m2 | 16.6 | 18.5 | ||
| Relative bias, %, median (Q1–Q3) | 39.4 (−0.7 to 95.0) | 49.6 (7.2–110.6) | ||
| Relative bias, %, mean (95% CI) | 90.2 (65.0–115.4) | 103.6 (76.6–130.6) | ||
| Accuracy, % (95% CI) | 30.6 (24.5–37.3) | 30.1 (24.0–36.8) | ||
| 30 ≤ GFR < 60 ( N = 811) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | 4.7 (−4.7 to 16.5) | 8.4 (−1.7 to 21.0) | |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | 7.7 (6.5–9.0) | 11.4 (10.1–12.6) | ||
| Precision, mL/min/1.73 m2 | 21.2 | 22.7 | ||
| Relative bias, %, median (Q1–Q3) | 10.0 (−10.5 to 36.5) | 18.3 (−3.7 to 45.9) | ||
| Relative bias, %, mean (95% CI) | 16.4 (13.7–19.0) | 24.2 (21.5–26.9) | ||
| Accuracy, % (95% CI) | 60.4 (57.0–63.7) | 54.7 (51.3–58.2) | ||
| GFR ≥60 ( N = 3681) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −7.8 (−20.9 to 4.7) | −3.3 (−16.5 to 9.2) | |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −8.3 (−8.9 to −7.6) | −4.0 (−4.6 to −3.3) | ||
| Precision, mL/min/1.73 m2 | 25.6 | 25.8 | ||
| Relative bias, %, median (Q1–Q3) | −8.5 (−20.2 to 5.5) | −3.5 (−15.8 to 10.8) | ||
| Relative bias, %, mean (95% CI) | −6.0 (−6.7 to −5.3) | −1.2 (−1.9 to −0.5) | ||
| Accuracy, % (95% CI) | 75 (73–76) | 78 (77–80) | ||
| Cancer cohort | GFR <30 ( N = 78) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | 15.8 (9.1–30.3) | 18.8 (11.4–33.2) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | 20.0 (16.2–23.7) | 22.9 (19.0–26.8) | ||
| Precision, mL/min/1.73 m2 | 21.2 | 21.9 | ||
| Relative bias, %, median (Q1–Q3) | 80.5 (38.1–175.6) | 93.3 (49.3–197.7) | ||
| Relative bias, %, mean (95% CI) | 156.5 (104.4–208.5) | 175.7 (119.5–231.8) | ||
| Accuracy, % (95% CI) | 12 (6–20) | 12 (6–20) | ||
| 30 ≤ GFR < 60 ( N = 508) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | 8.0 (−1.3 to 20.7) | 11.8 (2.2–25.5) | |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | 10.5 (9.0–11.9) | 14.6 (13.1–16.1) | ||
| Precision, mL/min/1.73 m2 | 22.0 | 23.4 | ||
| Relative bias, %, median (Q1–Q3) | 16.6 (−2.8 to 45.1) | 25.9 (4.4–56.0) | ||
| Relative bias, %, mean (95% CI) | 22.4 (19.3–25.5) | 31.0 (27.8–34.2) | ||
| Accuracy, % (95% CI) | 58 (54–63) | 50 (45–54) | ||
| GFR ≥60 ( N = 2727) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −7.1 (−19.8 to 4.9) | −2.7 (−15.3 to 9.5) | |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −7.8 (−8.5 to −7.1) | −3.4 (−4.1 to −2.6) | ||
| Precision, mL/min/1.73 m2 | 24.7 | 24.9 | ||
| Relative bias, %, median (Q1–Q3) | −7.7 (−19.4 to 5.8) | −2.7 (−14.8 to 11.4) | ||
| Relative bias, %, mean (95% CI) | −5.5 (−6.3 to −4.8) | −0.6 (−1.3 to 0.2) | ||
| Accuracy, % (95% CI) | 76 (75–78) | 80 (78–81) | ||
| Kidney donor cohorta | GFR ≥60 ( N = maskedb) |
Absolute bias, mL/min/1.73 m2, median (Q1–Q3) | −11.8 (−22.2 to 1.2) | −6.7 (−18.0 to 5.6) |
| Absolute bias, mL/min/1.73 m2, mean (95% CI) | −10.1 (−12.5 to −7.8) | −5.8 (−8.2 to −3.5) | ||
| Precision, mL/min/1.73 m2 | 23.4 | 23.6 | ||
| Relative bias, %, median (Q1–Q3) | −10.9 (−20.4 to 1.4) | −7.1 (−16.9 to 7.1) | ||
| Relative bias, %, mean (95% CI) | −7.7 (−10.0 to −5.4) | −3.3 (−5.7 to −1.0) | ||
| Accuracy, % (95% CI) | 80 (75–85) | 89 (84–92) |
CI, confidence interval.
Absolute bias is calculated as the difference between eGFR and mGFR; relative bias is calculated as the difference between eGFR and mGFR divided by the mGFR; precision is calculated as the IQR of the bias; and the accuracy is calculated as the percentage of patients with eGFR within the range of ±30% of the mGFR (P30).
aWe did not report the results for potential living kidney donors with mGFR <60 mL/min/1.73 m2 because of the low number of these patients.
bCells are masked, so it is not possible to identify or back-calculate numbers less than 5.
Overall, the classification according to no CKD, or CKD stages 1/2, CKD stage 3 and CKD stages 4/5 was very similar using the CKD-EPI21 and CKD-EPI09-NB equations in all three cohorts (Table 4). However, for CKD stage 3 and CKD stages 4/5, the CKD-EPI09-NB equation classified the patients more accurately than the CKD-EPI21 equation, whereas the CKD-EPI21 equation classified CKD stage more accurately than the CKD-EPI09-NB equation for mGFR ≥60 mL/min/1.73 m2 (Table 4).
Table 4:
Proportions of participants correctly classified according to GFR cut-offs.
| GFR <30 mL/min/1.73 m2 | 30 ≤ GFR < 60 mL/min/1.73 m2 | GFR ≥60 mL/min/1.73 m2 | Overall | ||
|---|---|---|---|---|---|
| Total cohort ( N = 4688) |
CKD-EPI09-NB | 110/196, 56% (49%–63%) | 473/811, 58% (55%–62%) | 3408/3681, 93% (92%–93%) | 3991/4688, 85% (84%–86%) |
| CKD-EPI21 | 100/196, 51% (44%–58%) | 449/811, 55% (52%–59%) | 3495/3681, 95% (94%–96%) | 4044/4688, 86% (85%–87%) | |
| Cancer cohort (N = 3313) |
CKD-EPI09-NB | 23/78, 29% (20%–40%) | 280/508, 55% (51%–59%) | 2541/2727, 93% (92%–94%) | 2844/3313, 86% (85%–87%) |
| CKD-EPI21 | 18/78, 23% (15%–33%) | 256/508, 50% (46%–55%) | 2607/2727, 96% (95%–96%) | 2881/3313, 87% (86%–88%) | |
| Kidney donor cohort (N = 239 |
CKD-EPI09-NB | Maskeda | Maskeda | Maskeda, 99% (97%–100%) | 232/239, 97% (94%–99%) |
| CKD-EPI21 | Maskeda | Maskeda | Maskeda, 99% (97%–100%) | 232/239, 97% (94%–99%) |
95% confidence intervals are given in parentheses.
Cells are masked, so it is not possible to identify or back-calculate numbers less than 5.
Additional analysis
For the EKFC equation in the total cohort, the absolute median bias was −8.4 mL/min/1.73 m2, precision was 24.6 mL/min/1.73 m2 and accuracy was 67% (Supplementary data, Table S3), suggesting that the EKFC performs slightly worse than both the CKD-EPI09-NB and CKD-EPI21 equations in these selected patient cohorts. When stratifying by sex we observed that both CKD-EPI equations performed worse for men than for women (Supplementary data, Table S4). The CKD-EPI09-NB equation performed slightly better than the CKD-EPI21 equation in women, while the reverse was seen for men. When stratifying by age, we observed that the CKD-EPI09-NB and CKD-EPI21 equation were equally biased for patients aged ≥65 years (Supplementary data, Table S5). The CKD-EPI21 equation performed slightly better than CKD-EPI09-NB equation for patients aged <65 years. Lastly, we did not observe major differences from the main analysis when restricting analysis to patients with an eGFR available within 5 days prior to the mGFR (Supplementary data, Table S6).
DISCUSSION
In a Danish cohort of patients with mGFR, we found an overall slightly lower bias, and similar accuracy, precision and classification of CKD stages with the CKD-EPI21 equation compared with the CKD-EPI09-NB equation. The performances were virtually consistent in both the total cohort, and the cohorts of cancer patients and potential living kidney donors. When stratifying by GFR, the CKD-EPI21 equation performed slightly better than the CKD-EPI09-NB equation when the mGFR was ≥60 mL/min/1.73 m2, whereas the CKD-EPI09-NB equation performed slightly better than the CKD-EPI21 when mGFR was <60 mL/min/1.73 m2. We found that the accuracy was considerable higher in the kidney donor cohort compared with the total and cancer cohort. This could potentially be explained by our findings that kidney donors generally had a higher mGFR compared with the total and cancer cohort, and that the CKD-EPI equations are more accurate at high mGFR. Furthermore, we observed that the new EKFC equation performed slightly worse than both the CKD-EPI09-NB and the CKD-EPI21 equation.
Our findings of a slightly better performance of the new CKD-EPI21 equation compared with the CKD-EPI09-NB equation are in contrast to the original validation study [3]. As there is a difference in creatinine between Black and non-Black individuals independent of kidney function [2], we expected that ignoring race, and thereby omitting information, would lead to more biased estimates. There are a couple of possible explanations behind this observation. First, the study population may not have been representative of the non-Black individuals used to model the CKD-EPI equations, which may have biased the equations. Second, the method used to measure GFR differed by study. The development of the CKD-EPI equation was mainly based on iothalamate [3]. This may have provided mGFR values that were different from those obtained by 51Cr-EDTA plasma clearance, since iothalamate clearance and 51Cr-EDTA plasma clearance have been shown to yield slightly different results [23]. Likewise, the difference in performance of the EKFC equation between our study and the original validation study may have been due to the fact that the EKFC equation was developed in a cohort where the GFR was measured mainly based on inulin and iohexol [22]. Thus there is a need to achieve better standardizations of the mGFR methods.
Notably, both eGFR equations tended to overestimate GFR at lower values and to underestimate GFR at higher values in all three cohorts. mGFR is often prescribed when a highly accurate GFR is required or when the eGFR is thought to be exceptionally inaccurate (e.g. if muscle mass is abnormal). This may have contributed to the discrepancy between eGFR values and mGFR in the total cohort and the cancer cohort. However, GFR is measured in virtually all patients undergoing kidney donor eligibility screening regardless of muscle mass. Thus, selection of patients with extreme body composition is unlikely to explain the non-uniform bias that we observed.
In the original validation study, which included data from various studies, Inker et al. found that the CKD-EPI21 equation provided equally biased estimates in Black individuals and slightly more biased estimates in non-Black individuals compared with estimates obtained using the CKD-EPI09 equation [3]. Inker et al. reported considerably higher accuracy than in our study. A couple of studies have subsequently validated the CKD-EPI21 equation in different patient groups [8–14, 24]. In line with our study, the studies generally reported no or only minor differences in the performance between CKD-EPI09-NB and CKD-EPI21 equations in patients with mGFR including kidney transplant recipients, living kidney donors, patients with CKD and cancer patients, among others [8–14, 24]. Although our study took advantage of comprehensive, routinely collected health data in Denmark, some limitations should be considered when interpreting our results. First, we lacked information on race, though the vast majority are thought to be white because approximately 90% of Danes have a Danish ancestor [25]. In addition, as the study was conducted in a predominantly non-Black population, our results may not be generalizable to other populations. Second, the cohorts consisted of patients whose GFR was measured due to a clinical indication, so the estimates may not be applicable to the general population or to cancer patients in general. Third, we lacked direct information on the indication for the mGFR. However, we expect that our algorithm for categorizing patients based on diagnosis and surgery codes identified the correct indication for the vast majority of the patients. Fourth, the median time of 5 days between the eGFR and mGFR could potentially affect the performance of the equations; however, we did not see major differences in the performance when restricting analysis to patients with an eGFR available within 5 days prior to the mGFR.
In conclusion, in a selected cohort of predominantly white Danish patients, including cancer patients and potential living kidney donors, the CKD-EPI21 equation performed slightly better than the CKD-EPI21-NB equation except for patients with a GFR <60 mL/min/1.73 m2; however, these differences may not be clinically significant.
Supplementary Material
ACKNOWLEDGEMENTS
The authors thank Lene Elsebeth Nielsen, Henriette Kristoffersen and Henriette Vendelbo Graversen for their support with data collection.
Contributor Information
Philip Vestergaard Munch, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark.
Uffe Heide-Jørgensen, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark.
Simon Kok Jensen, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark.
Henrik Birn, Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark; Department of Renal Medicine, Aarhus University Hospital, Aarhus, Denmark; Department of Biomedicine, Aarhus University, Aarhus, Denmark.
Søren Viborg Vestergaard, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark; Department of Clinical Pharmacology, Aarhus University Hospital, Aarhus, Denmark.
Jørgen Frøkiær, Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark; Department of Nuclear Medicine and PET-Centre, Aarhus University Hospital, Aarhus, Denmark.
Henrik Toft Sørensen, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark.
Christian Fynbo Christiansen, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Midtjylland, Denmark.
FUNDING
This study was supported by the Independent Research Fund Denmark (grant number 0134-00407B).
AUTHORS’ CONTRIBUTIONS
C.F.C., H. B., H.T.S., J. F., P.V.M., S.K. J., S.V.V., and U. H.-J. conceptualized the study. C.F.C., P.V.M., S.K.J., S.V.V., and U.H.-J. designed the study. P.V.M. reviewed the literature and wrote the initial draft of the article. U.H.-J. conducted the formal analysis. All authors interpreted the results and reviewed and edited the article.
DATA AVAILABILITY STATEMENT
The data underlying this study are available from the Danish Health Data Authority. Restrictions apply to the availability of these data. Researchers may apply for data access through the Danish Health Data Authority.
CONFLICT OF INTEREST STATEMENT
C.F.C., H.B., H.T.S., J.F., P.V.M., S.K.J., S.V.V. and U.H.-J. report employment with Aarhus University and Aarhus University Hospital. The Department of Clinical Epidemiology, Department of Biomedicine, Department of Renal Medicine, Department of Clinical Pharmacology and Department of Nuclear Medicine and PET-Centre at Aarhus University Hospital are involved in studies with funding from various companies as research grants to (and administered by) Aarhus University. None of these is related to this study. H.B. also reports consultancy agreements with AstraZeneca, Galapagos and Vifor Pharma, as well as research funding from GlaxoSmithKline (GSK) and Vifor Pharma, and reports honoraria from Alexion, AstraZeneca, MSD, Novartis Healthcare, NOVO Nordisk and Otsuka Pharmaceuticals. None of these is related to this study.
REFERENCES
- 1. Levin AS, Paul E, Bilous RW et al. ; Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group . KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease. Kidney Int Suppl 2013;3:1–150. [Google Scholar]
- 2. Levey AS, Stevens LA, Schmid CH et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 2009;150:604–12. 10.7326/0003-4819-150-9-200905050-00006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Inker LA, Eneanya ND, Coresh J et al. New creatinine- and cystatin C-based equations to estimate GFR without race. N Engl J Med 2021;385:1737–49. 10.1056/NEJMoa2102953 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Delgado C, Baweja M, Crews DC et al. A unifying approach for GFR estimation: recommendations of the NKF-ASN Task Force on reassessing the inclusion of race in diagnosing kidney disease. Am J Kidney Dis 2022;79:268–88.e1. 10.1053/j.ajkd.2021.08.003 [DOI] [PubMed] [Google Scholar]
- 5. Casal MA, Ivy SP, Beumer JH et al. Effect of removing race from glomerular filtration rate-estimating equations on anticancer drug dosing and eligibility: a retrospective analysis of National Cancer Institute phase 1 clinical trial participants. Lancet Oncol 2021;22:1333–40. 10.1016/S1470-2045(21)00377-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Vestergaard SV, Heide-Jørgensen U, Birn H et al. Effect of the refitted race-free eGFR formula on the CKD prevalence and mortality in the Danish population. Clin J Am Soc Nephrol 2022;17:426–8. 10.2215/CJN.14491121 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Fu EL, Coresh J, Grams ME et al. Removing race from the CKD-EPI equation and its impact on prognosis in a predominantly White European population. Nephrol Dial Transplant 2023; 38:119–28. 10.1093/ndt/gfac197 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Hundemer GL, White CA, Norman PA et al. Performance of the 2021 race-free CKD-EPI creatinine- and cystatin C-based estimated GFR equations among kidney transplant recipients. Am J Kidney Dis 2022;80:462–72.e1. 10.1053/j.ajkd.2022.03.014 [DOI] [PubMed] [Google Scholar]
- 9. Meeusen JW, Kasozi RN, Larson TS et al. Clinical impact of the refit CKD-EPI 2021 creatinine-based eGFR equation. Clin Chem 2022;68:534–9. 10.1093/clinchem/hvab282 [DOI] [PubMed] [Google Scholar]
- 10. Goodson DA, Chalupsky MR, Wiegley N et al. GFR estimation in potential living kidney donors: race and non-race based equations and measured GFR. Kidney Med 2022;100558. 10.1016/j.xkme.2022.100558 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Pei X, Zhao W, Du X. A validation study of the 2021 CKD-EPI equations: data from two cohort studies in Nanjing, China. J Nephrol 2022;35:2155–6. 10.1007/s40620-022-01384-z [DOI] [PubMed] [Google Scholar]
- 12. Delanaye P, Vidal-Petiot E, Björk J et al. Performance of creatinine-based equations to estimate glomerular filtration rate in White and Black populations in Europe, Brazil and Africa. Nephrol Dial Transplant 2023;38:106–18. 10.1093/ndt/gfac241 [DOI] [PubMed] [Google Scholar]
- 13. Delanaye P, Masson I, Maillard N et al. The new 2021 CKD-EPI equation without race in a European cohort of renal transplanted patients. Transplantation 2022;106:2443–7. 10.1097/TP.0000000000004234 [DOI] [PubMed] [Google Scholar]
- 14. Silva AM, Shen W, Heo M et al. Ethnicity-related skeletal muscle differences across the lifespan. Am J Hum Biol 2010;22:76–82. 10.1002/ajhb.20956 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Schmidt M, Schmidt SAJ, Adelborg K et al. The Danish health care system and epidemiological research: from health care contacts to database records. Clin Epidemiol 2019;11:563–91. 10.2147/CLEP.S179083 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Schmidt M, Schmidt SA, Sandegaard JL et al. The Danish National Patient Registry: a review of content, data quality, and research potential. Clin Epidemiol 2015;7:449–90. 10.2147/CLEP.S91125 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Grann AF, Erichsen R, Nielsen AG et al. Existing data sources for clinical epidemiology: the clinical laboratory information system (LABKA) research database at Aarhus University, Denmark. Clin Epidemiol 2011;3:133–8. 10.2147/CLEP.S17901 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Arendt JFH, Hansen AT, Ladefoged SA et al. Existing data sources in clinical epidemiology: laboratory information system databases in Denmark. Clin Epidemiol 2020;12:469–75. 10.2147/CLEP.S245060 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Schmidt M, Pedersen L, Sørensen HT. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol 2014;29:541–9. 10.1007/s10654-014-9930-3 [DOI] [PubMed] [Google Scholar]
- 20. Murray AW, Barnfield MC, Waller ML et al. Assessment of glomerular filtration rate measurement with plasma sampling: a technical review. J Nucl Med Technol 2013;41:67–75. 10.2967/jnmt.113.121004 [DOI] [PubMed] [Google Scholar]
- 21. Jødal L, Brøchner-Mortensen J. Reassessment of a classical single injection 51Cr-EDTA clearance method for determination of renal function in children and adults. Part I: analytically correct relationship between total and one-pool clearance. Scand J Clin Lab Invest 2009;69:305–13. 10.1080/00365510802566882 [DOI] [PubMed] [Google Scholar]
- 22. Pottel H, Björk J, Courbebaisse M et al. Development and validation of a modified full age spectrum creatinine-based equation to estimate glomerular filtration rate : a cross-sectional analysis of pooled data. Ann Intern Med 2021;174:183–91. 10.7326/M20-4366 [DOI] [PubMed] [Google Scholar]
- 23. Soveri I, Berg UB, Björk J et al. Measuring GFR: a systematic review. Am J Kidney Dis 2014;64:411–24. 10.1053/j.ajkd.2014.04.010 [DOI] [PubMed] [Google Scholar]
- 24. Raynaud M, Al-Awadhi S, Juric I et al. Race-free estimated glomerular filtration rate equation in kidney transplant recipients: development and validation study. BMJ 2023;381:e073654. 10.1136/bmj-2022-073654 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. StatisticsDenmark . FOLK1C: population at the first day of the quarter by region, sex, age (5 years age groups), ancestry and country of origin. https://www.statbank.dk/FOLK1C (7 February 2023, date last accessed). [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this study are available from the Danish Health Data Authority. Restrictions apply to the availability of these data. Researchers may apply for data access through the Danish Health Data Authority.