OUP user menu

Estimating glomerular filtration rate: comparison of the CKD-EPI and MDRD equations in a large UK cohort with particular emphasis on the effect of age

J.L. Carter, P.E. Stevens, J.E. Irving, E.J. Lamb
DOI: http://dx.doi.org/10.1093/qjmed/hcr077 839-847 First published online: 7 June 2011

Abstract

Background: The chronic kidney disease (CKD)-Epidemiology Collaboration (CKD-EPI) equation was developed to address the underestimation of measured glomerular filtration rate (GFR) by the Modification of Diet in Renal Disease (MDRD) equation at levels >60 mL/min/1.73 m2.

Aim: To assess the impact of the CKD-EPI equation on the estimation of GFR in a large adult UK population (n = 561 400), particularly looking at the effect of age.

Design: Serum creatinine results (ID-MS-aligned enzymatic assay) were extracted from the pathology database during 1 year on adult (≥18 years) patients from primary care.

Methods: The first available creatinine result from 174 448 people was used to estimate GFR using both equations and agreement assessed.

Results: Median CKD-EPI GFR was significantly higher than median MDRD GFR (82 vs. 76 mL/min/1.73 m2, P < 0.0001). Overall mean bias between CKD-EPI and MDRD GFR was 5.0%, ranging from 13.0% in the 18–29 years age group down to −7.5% in those aged ≥90 years. Although statistically significant at all age groups the difference diminished with age and the agreement in GFR category assignment increased. Age-adjusted population prevalence of CKD Stages 3–5 was lower by CKD-EPI than by MDRD (4.4% vs. 4.9%).

Conclusion: CKD-EPI produces higher GFR and lower CKD estimates, particularly among 18–59 year age groups with MDRD estimated GFRs of 45–59 mL/min/1.73 m2 (Stage 3A). However, at ages >70 years there is very little difference between the equations, and among the very elderly CKD-EPI may actually increase CKD prevalence estimates.

Introduction

Equations to estimate glomerular filtration rate (GFR) based on serum creatinine measurement are routinely used to assess kidney function. The most commonly used equation is the Modification of Diet in Renal Disease (MDRD) Study equation.1 In the UK, automated reporting of GFR whenever a serum creatinine is requested was introduced from April 2006. The lack of studies to guide treatment and referral of people identified with chronic kidney disease (CKD) in this incidental fashion has led to some confusion and inconsistency, particularly among those with milder degrees of kidney dysfunction [estimated GFR (eGFR) 45–60 mL/min/1.73 m2].2 Furthermore, the MDRD equation was developed in people with significant CKD and has been shown to systematically underestimate measured GFR at levels >60 mL/min/1.73 m2 and thus overestimate CKD prevalence. A new equation, developed by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI), is based on more diverse pooled research and clinical populations and reduces bias against reference GFR measurements in comparison to the MDRD equation.3 This has now been tested in several populations including transplant patients,4 Japanese patients,5 middle-aged subjects from the USA without a history of cardiovascular disease6 and non-institutionalized adult Australians.7 Overall, these studies demonstrate that the CKD-EPI equation produces a more accurate estimate of GFR and lower prevalence of CKD.

In the present study, we assessed the likely impact of introducing the CKD-EPI equation for estimating GFR in a UK population, particularly looking at the effect of age. We compared the prevalence of CKD using both the established MDRD equation and the new CKD-EPI equation based on creatinine results from primary care over a 12-month period across East Kent, a semirural area of Southern England with a population of just over half a million.

Methods

Patients

All serum creatinine results obtained for standard clinical purposes during a 12-month period (May 2009 to April 2010) for adults aged ≥18 years were extracted from the pathology database. Patients from locations other than primary care/general practice, those <18 years of age and those with missing demographics necessary to estimate GFR were excluded (Figure 1). Concurrent albuminuria and glycated haemoglobin (HbA1c) data were also extracted. GFR was estimated using the MDRD1 and CKD-EPI3 equations (for white men and women) and a comparison of GFR categories using the two equations was performed. Albuminuria was defined as those with an albumin to creatinine ratio (ACR) ≥2.5 mg/mmol in men and those with an ACR ≥3.5 mg/mmol in women. Proteinuria was defined as those with an ACR ≥30 mg/mmol.8 The prevalence of CKD stages using both equations and albuminuria/proteinuria data was adjusted for age using UK population data from England, 2009,9 and the known age distribution of the NHS Eastern and Coastal Kent Primary Care Trust (PCT) population in mid-2009.10 CKD Stages 1–5 were defined according to the internationally agreed system.11 Diabetes was defined as any patient with an HbA1c result: in the UK HbA1c is not currently used as a diagnostic test.

Figure 1.

Flow diagram showing inclusion criteria for creatinine tests collected over 12 months across East Kent.

Laboratory analyses

Serum creatinine was measured using an enzymatic method on Abbott Architect analysers (Abbott Diagnostics, Maidenhead, Berkshire, UK). The enzymatic method for creatinine is standardized against NIST SRM 967 and thus is traceable to isotope dilution mass spectrometry (ID-MS). The assay was related to an ID-MS assay according to the equation: Abbott enzymatic = 0.982 (ID-MS) + 3.3 (n = 203, E.J. Lamb). The assay is related to the Roche creatinine plus enzymatic assay (Hoffman-La Roche, Basel, Switzerland) used to re-express the MDRD equation according to the equation: Abbott enzymatic = 1.0338 (Roche enzymatic) + 0.98 (unpublished data, E.J. Lamb). The reference ranges for serum creatinine were 49–90 µmol/l in women and 64–104 µmol/l in men.12 External quality assessment data from the UK National External Quality Assessment Scheme (UKNEQAS) for general chemistry during the study period revealed that laboratory creatinine data were consistently within ±5% of the enzymatic group target mean result. Further, UKNEQAS data specifically from the GFR scheme showed creatinine data to consistently be within ±4% of the reference ID-MS target. Urinary albumin and creatinine concentrations were measured on an Abbott Architect analyser using turbidimetric immunoassay and enzymatic creatinine assays, respectively, and the ACR was calculated. HbA1c was measured by high-performance liquid chromatography (HPLC) on an ADAMS™ A1c HA-8160 analyser (A.Menarini, Wokingham, Berkshire, UK) with standardization traceable to that used in the Diabetes Control and Complications Trial (DCCT).

Statistics

Analyses were performed using Analyse-it™ (Analyse-it™ Software, Ltd, Leeds, UK). A P-value of <0.05 was considered statistically significant. Most of the data were not normally distributed as demonstrated using the Shapiro–Wilk W test. The non-parametric Wilcoxon matched pairs signed ranks test was used for comparison between CKD-EPI GFR and MDRD GFR. Comparison between the GFR stages derived from the two GFR estimates was undertaken using the weighted kappa test for agreement: kappa statistic (κ) 0.21–0.40 is considered fair agreement, 0.41–0.60 moderate agreement, 0.61–0.80 substantial agreement and 0.81–1.00 near-perfect agreement. Log-transformed data were used for Bland–Altman difference plots (difference between CKD-EPI GFR and MDRD GFR vs. mean of CKD-EPI GFR and MDRD GFR).

Results

After excluding patients aged <18 years, those being requested in secondary care and those with incomplete records, 263 658 creatinine tests were performed from an adult population of 561 400. For each individual only the first available creatinine result was used, giving a final study cohort of 174 448 (men 78 952; women 95 496) (Figure 1). Albuminuria was tested in 17 809 of these. Prevalence of diabetes was 5.3% (33 050 people) which is similar to the ascertainment level of 5.9% recorded by the Quality and Outcomes Framework (QOF) for NHS Eastern and Coastal Kent PCT patients in 2009/2010.13

CKD-EPI eGFR was significantly higher than MDRD eGFR in the whole group combined and also when the cohort was subdivided separately into men, women, those with and without diabetes and/or proteinuria (Table 1). At most ages CKD-EPI eGFR was significantly higher than MDRD eGFR, with the exception of those aged ∼80 years and older when it was significantly lower (Table 2).

View this table:
Table 1

Characteristics of all patients from primary care with the first available creatinine result in the study period (n = 174 448)

AllMenWomenDiabetesAlbuminuriaProteinuria
NoYesNoYesNoYes
n174 44878 95295 496141 39833 050166 0138435173 589859
Median age (years)62 (49 to 74)63 (50 to 73)62 (47 to 75)62 (47 to 73)66 (56 to 76)62 (48 to 73)75 (65 to 82)62 (49 to 74)75 (65 to 82)
Median creatinine (µmol/l)78 (67 to 92)88 (78 to 100)70 (62 to 81)78 (67 to 91)80 (65 to 95)78 (67 to 91)97 (78 to 125)78 (67 to 92)111 (83 to 153)
Median MDRD eGFR (mL/min/1.73 m2)76 (63 to 89)77 (65 to 90)74 (62 to 88)76 (64 to 88)75 (61 to 90)76 (64 to 89)57 (44 to 78)76 (64 to 89)49 (34 to 70)
Median CKD-EPI eGFR (mL/min/1.73 m2)82 (66 to 95)*82 (67 to 94)*81 (65 to 96)*82 (67 to 96)*80 (62 to 94)*82 (67 to 96)*58 (43 to 82)***82 (66 to 95)*49 (33 to 72)**
  • Data shown as median (IQR). CKD-EPI eGFR is significantly higher than MDRD eGFR (*P < 0.0001; **P < 0.005; ***P < 0.05).

View this table:
Table 2

Median (IQR) creatinine, MDRD and CKD-EPI eGFR by age group in individuals from primary care with the first available creatinine result during the study period (n = 174 448)

Age (years)nMedian creatinine (µmol/l)Median MDRD eGFR (mL/min/1.73 m2)Median CKD-EPI eGFR (mL/min/1.73 m2)
18–2911 41570 (62 to 80)98 (87 to 110)114 (102 to 124)*
30–3912 33972 (63 to 82)89 (80 to 100)105 (93 to 114)*
40–4922 26974 (65 to 85)84 (75 to 94)96 (85 to 105)*
50–5928 52576 (66 to 87)80 (70 to 90)88 (78 to 98)*
60–6940 42479 (68 to 91)74 (65 to 84)79 (69 to 90)*
70–7933 93884 (71 to 99)67 (56 to 78)69 (57 to 81)*
80–8921 77589 (74 to 110)59 (47 to 72)58 (45 to 71)*
90–107376394 (76 to 119)52 (41 to 65)49 (38 to 62)*
  • *CKD-EPI eGFR is significantly different from MDRD eGFR (P < 0.0001).

Agreement between GFR categories using MDRD and CKD-EPI equations was substantial (κ = 0.77) (Figure 2A). When the cohort was subdivided into those aged <70 years (n = 114 972) (Figure 2B) and those aged ≥70 years (n = 59 476) (Figure 2C), the agreement improved in older people (<70 years cohort κ = 0.60 vs. ≥70 years cohort κ = 0.88). A similar pattern was seen when men and women were considered separately (women <70 years κ = 0.55 vs. women ≥70 years κ = 0.88; men <70 years κ = 0.68 vs. men ≥70 years κ = 0.88). Overall, the difference observed between CKD-EPI and MDRD eGFR in women (Supplementary Figure 1 and Table 1) was similar but slightly greater than that in men (Supplementary Figure 2 and Table 2).

Figure 2.

Differences in GFR categories in all individuals from primary care over 12 months (n = 174 448) (A) and in those aged <70 years (n = 114 972) (B) and in those aged ≥70 years (n = 59 476) (C) using MDRD (hatched bars) and CKD-EPI (solid bars) equations. Data are expressed as n (%).

Tables 3 and 4 further illustrate the impact of age on both estimates of GFR. The CKD-EPI equation produces higher GFR estimates, particularly marked among 18–59 year age-groups with GFR >60 mL/min/1.73 m2. In contrast, at ages 70–79 years there was less difference between the CKD-EPI and MDRD eGFRs (κ = 0.89), and among the very elderly the CKD-EPI equation actually gave lower estimates of GFR (Table 4). Overall the mean bias between MDRD and CKD-EPI equations was 5.0%, ranging from 13% in the 18–29 year age group down to −7.5% in those aged >90 years. Similar age-dependent differences between the two equations were observed among people with diabetes (Supplementary Table 3).

View this table:
Table 3

Differences in prevalence of eGFR categories (mL/min/1.73 m2) within the tested population by age groups for men and women in primary care using the MDRD and CKD-EPI equations

Age groups (years)Percent prevalence of CKD-EPI eGFR − percent prevalence of MDRD eGFR (y − x)
>12090–12060–8945–5930–4415–29<15
18–2923.3−2.4−20.7−0.2000
30–390.930.4−30.4−0.9000
40–49−2.232.8−28.8−1.7−0.100
50–59−1.823.0−17.9−3.1−0.200
60–69−1.111.2−5.4−4.1−0.4−0.10
70–79−0.6−3.76.6−2.2−0.30.10.1
80–89−0.4−5.23.2−0.11.41.00.1
90–107−0.3−4.2−2.503.92.90.2
  • Values are expressed as percentages.

View this table:
Table 4

Agreement between MDRD eGFR (x) and CKD-EPI eGFR (y) for GFR categories using the κ-test and Bland–Altman analysis

Patient groupsAge subgroups (years)nAgreement analysis
Weighted κ-statisticBland–Altman analysis (y − x)Limits of agreement (%) (95% CI of estimate)
Mean bias (%) (95% CI)LowerUpper
All patientsAll174 4480.775.0 (5.0 to 5.0)−9.8 (−9.8 to −9.8)22.2 (22.2 to 2.2)
18–2911 4150.3713.0 (13.0 to 13.2)−0.7 (−0.7 to 0.5)28.5 (28.2 to 28.8)
30–3912 3390.4313.0 (13.0 to 13.2)−0.2 (−0.5 to 1.0)28.2 (27.9 to 28.5)
40–4922 2690.4211.2 (11.2 to 11.2)−1.1 (−1.4 to −1.1)25.3 (25.0 to 25.3)
50–5928 5250.537.9 (7.6 to 7.9)−4.1 (−4.3 to −4.1)21.3 (21.1 to 21.3)
60–6940 4240.734.5 (4.5 to 4.7)−6.2 (−6.2 to −6.1)16.7 (16.4 to 16.7)
70–7933 9380.890.7 (0.5 to 0.7)−8.6 (−8.6 to −8.4)10.7 (10.7 to 10.7)
80–8921 7750.88−3.6 (−3.6 to −3.4)−12.1 (−12.1 to −11.9)5.7 (5.7 to 5.9)
90–10737630.75−7.5 (−7.5 to −7.3)−15.5 (−15.7 to −15.3)1.2 (0.9 to 1.6)
All menAll78 9520.793.3 (3.3 to 3.3)−10.7 (−10.7 to −10.5)19.1 (19.1 to 19.1)
All womenAll95 4960.756.4 (6.4 to 6.4)−8.8 (−8.8 to −8.8)23.9 (23.9 to 24.2)
Patients with diabetesAll33 0500.822.3 (2.1 to 2.3)−11.7 (−11.7 to −11.5)18.3 (18.3 to 18.6)
Patients with albuminuriaAll84350.87−0.9 (−1.1 to −0.7)−13.1 (−13.3 to −12.9)13.0 (12.7 to 13.2)
Patients with proteinuriaAll8590.86−1.4 (−1.8 to −0.9)−13.1 (−13.7 to −12.5)12.0 (11.2 to 12.7)
  • Bland–Altman plots were calculated following log transformation.

Based on eGFR derived from a single creatinine measurement, the CKD-EPI equation yields a lower estimated prevalence of CKD Stages 3–5 (4.4%) in our population than the MDRD equation (4.9%) (Figure 3).

Figure 3.

Prevalence of CKD stages in UK using MDRD (hatched bars) and CKD-EPI (solid bars) equations based on age-adjusted census data. Data are expressed as percentages.

Discussion

Currently in the UK the MDRD study equation remains that recommended for use by the Department of Health.14 However, a growing body of literature suggests that the CKD-EPI equation provides a less biased estimate of GFR than the MDRD study equation,3,5 particularly at GFR levels >60 mL/min/1.73 m2.15 Furthermore, it appears to offer advantages over the MDRD equation in terms of risk stratification.6,16 Therefore, it seems likely that at some point the CKD-EPI equation will be introduced into routine clinical practice in the UK. It is consequently important to understand how this will affect CKD prevalence rates and the ‘phenotype’ of CKD in the UK population.

We have modelled the likely impact of introduction of the CKD-EPI equation in the UK using a large, representative sample of adult patients presenting to their primary care physician in a 1-year period. As observed by others3,5,15 overall the CKD-EPI equation produced higher estimates of GFR and hence lower prevalence estimates of CKD than the MDRD equation. In our population, the overall prevalence of CKD Stages 3–5 was 4.4% compared with 4.9% when applying the MDRD equation. Similar differences were seen when the two equations were applied to the National Health and Nutrition Examination Survey (NHANES CKD Stages 3 and 4, 6.7 vs. 8.2%);3 the Australian Diabetes, Obesity and Lifestyle Study (AusDiab CKD Stages 3 and 4, 5.8 vs. 7.8%)7 and the Quality Improvement in CKD (QICKD Stages 3–5 4.80 vs. 5.41%)16 databases. The difference in CKD prevalence between the two equations is mainly attributable to a decrease in the prevalence of Stage 3A CKD when the CKD-EPI equation is applied. Consistent with other studies, the largest effect of the CKD-EPI equation on GFR estimation we observed was seen in middle-aged and female patients and at GFR levels >60 mL/min/1.73 m2. The presence or absence of diabetes did not appear to affect the relative biases of the equations.

The effect of the CKD-EPI equation on GFR estimation is of particular relevance in terms of outcomes associated with CKD. We know that presence of albuminuria affects outcome at all levels of GFR,17 but we also know that outcomes in those without albuminuria and GFR levels of 30–59 mL/min/1.73 m2 are better than those with albuminuria and higher levels of GFR.18,19 Roderick et al.20 have shown that in subjects aged ≥75 years in the UK, as kidney function decreases there is a graded and independent increase in all-cause and cardiovascular mortality risk, especially in those with eGFR <45 mL/min/1.73 m2. In this study, the presence of dipstick proteinuria did not add to cardiovascular mortality risk. The relevance of a finding of reduced kidney function alone, in the absence of any comorbid conditions, should be interpreted differently according to the likelihood of the need for consideration of renal replacement therapy during a given patient's lifetime. To do this requires confidence in the accuracy of the prediction equations we use in clinical practice.

The consistency of our data with that from other populations in which the CKD-EPI equation has been studied could suggest that it is probably applicable to a UK population. The reduction in prevalence of Stage 3A CKD would be a welcome development to many concerned about disease ‘labelling’ in patients at low risk.21 However, a note of caution should be sounded. The CKD-EPI equation was neither derived nor tested in large numbers of older people (e.g. only 219/5504 of the development and 109/2750 of the internal validation sets were >70 years old).3 The NHANES and AusDiab cohorts were significantly younger than our study group, which represented a real-life clinical sampling scenario. Our data, and that of others,22 suggest that the CKD-EPI equation produces lower estimates of GFR in older people. In the Nijmegen study,22 the age thresholds at which this became apparent was in women aged >75 and men aged >70 years, whereas in our data (with significantly greater numbers of subjects) the age thresholds were slightly higher, ∼85 years in women and 80 years in men (Supplementary Tables 1 and 2). Hence, the CKD-EPI equation will not act as a panacea for reducing apparent CKD prevalence in older people; in fact, it may identify more such individuals. This is certainly relevant to the UK where in two recently published studies following introduction of the MDRD equation the age of patients referred to nephrology units for assessment significantly increased from 66 to 70 years23 and from 63 to 69 years.24

For obvious reasons our study did not include a reference measure of GFR. Hence, we can draw no conclusions regarding the relative accuracy of the two equations in a UK population. Given the consistency of the effects we have observed to those studies that have included a reference measure,3,15 it may be reasonable to assume improved accuracy of CKD-EPI estimates in UK individuals, although it should be noted that Nyman et al.25 did not observe improved accuracy of the CKD-EPI equation in a Caucasian Swedish population.

Our study has some other weaknesses. We did not have ethnicity data and hence neither the MDRD nor CKD-EPI equations have been corrected for black race. However, the East Kent population is predominantly Caucasian: population data show only 1.3% of the population to be black and a further 1.9% recorded as Asian (Office of National Statistics, Experimental Statistics, Table EE1, 2007). We have only used samples from primary care, so we will not have captured, for example, those patients with established renal failure that are seen within our kidney care centre. Nevertheless, while this may have resulted in a conservative estimate of the prevalence of CKD Stage 5, our emphasis was on the relative impact of the two equations and it is well known that both equations produce very similar estimates of GFR at severely reduced levels of kidney function.3 Further, we have been unable to obtain an accurate prevalence estimate of CKD Stages 1 and 2. Albuminuria testing is not universal in the UK, predominantly being undertaken annually in patients with diabetes: only 3% of our population underwent albuminuria testing. Hence, our data must significantly underestimate the prevalence of CKD Stages 1 and 2 (e.g. 0.63% compared to 7% in the Health Survey for England26 and 7.3% in NHANES3).

An earlier UK primary care study27 which used universal sampling and based CKD stage on a single creatinine measurement and the MDRD equation obtained a prevalence estimate of CKD Stages 3–5 of 8.5%, compared to 4.9% in the present study. In the present study, only 31% of the total adult population of East Kent had a blood test for serum creatinine. Inevitably therefore our prevalence estimate must be conservative because we have assumed that the population who have not had a blood test for serum creatinine have a GFR level >60 mL/min/1.73 m2. A similar recently published study in the UK, in which 36% of a primary care population were sampled produced a prevalence estimate of CKD Stages 3–5 of 6.4% using the MDRD equation when CKD staging was based on a single creatinine measurement.16 The recorded prevalence in primary care records of diagnosed CKD Stages 3–5 among people aged ≥18 years (i.e. those who have had a blood test for serum creatinine, have a GFR <60 mL/min/1.73 m2 and have been entered on the CKD disease register) in NHS Eastern and Coastal PCT 2010 was 4.1%.28

In common with many other epidemiological studies in this area, our estimates have been based on a single creatinine result. The definition of CKD requires the demonstration of reduced kidney function for >3 months and this is likely to reduce the prevalence estimates observed in single sample studies. The QICKD study modelled this effect and observed that applying this more accurate definition of CKD reduced their CKD Stages 3–5 prevalence estimates by ∼1%.16

In a large, representative sample of the UK population using an IDMS aligned enzymatic creatinine assay we have confirmed the observations from other countries that the CKD-EPI equation reduces the apparent prevalence of CKD, in particular CKD Stage 3A and that this effect is most marked among middle-aged females. Before recommending switching to the CKD-EPI equation for eGFR reporting in the UK the potential implications of lower GFR estimates in the elderly require further assessment.

Supplementary Data

Supplementary data are available at QJMED online.

Acknowledgements

We are grateful to Dr Neil Dalton for the ID-MS assay comparison data and also to Alan Young for the extraction of data from the pathology database.

Conflict of interest: None declared.

References

View Abstract