Adoption of CKD-EPI (2021) for Glomerular Filtration Rate Estimation: Implications for UK Practice

The worldwide incidence of chronic kidney disease (CKD) continues to grow [1], in part owing to an ageing, multi-morbid population. The number of patients requiring some form of kidney replacement therapy (including kidney transplantation) is growing in parallel with this. In the UK, the estimated cost of 1 year of outpatient haemodialysis/haemodiafiltration is somewhere between GBP 28,000 and GBP 33,000 [2]. For the National Health Service in England, the total cost attributed to CKD in 2009–2010 was GBP 1.45 billion, 1.3% of the overall annual budget [3].

Since 2002, estimated glomerular filtration rate (eGFR) has been used to classify CKD into stages relating to severity of disease [4]. A patient’s eGFR sways treatment decisions in both the inpatient and outpatient domain. For the nephrologist, eGFR guides referral from primary to secondary care, specialist referrals within the secondary care service itself (e.g., to specialist “low clearance” or “advanced kidney disease” clinics), and referrals made in preparation for renal transplantation. In the more acute setting, eGFR will guide, e.g., medication dosing, fluid therapy, and contrast administration related to cross-sectional imaging. Research studies appraising drug pharmacokinetics also reference patient eGFR as a patient’s volume of distribution will influence the body’s handling of drugs. CKD carries with it increased cardiovascular mortality and morbidity [5, 6], potentially impacting significant decision points in a patient’s journey, such as admission to intensive care, dosing of chemotherapy, or perhaps the more prosaic matter of determining insurance premiums.

The Modification of Diet in Renal Disease (MDRD) equation to calculate eGFR, first described in 1999 [7], was fed into the 2002 recommendations from National Kidney Federation Kidney Disease Outcomes Quality Initiative (NKF-KDOQI). The practice changing use of the MDRD equation was perhaps an unintended consequence of Levey’s work and not the group’s original intention [8]. Other problems have since been described with the MDRD equation, including the use of coefficients for racial groups (a social, rather than a biological, construct) and the application of findings in a North American population to other populations worldwide. Interestingly, the MDRD group did not directly measure muscle mass in patients, and the hypothesis linking the African American racial group to increased creatinine generation has since been queried [9, 10].

In 2021, the NKF recommended the use of updated CKD-EPI equations [11] which no longer carried within them coefficients related to ethnicity or racial group [12]. The UK has followed the example from the USA and now advises that equation adjustments made for race or ethnicity may not be accurate [13]. With this in mind, it is likely that an increasing number of biochemistry laboratories in the UK will move to using the updated CKD-EPI equation to report eGFR [14]. Continental European opinion differs here, and it is likely that CKD-EPI (2009) will continue to be used without the racial coefficient [15, 16].

Our group wished to evaluate the potential impact of adoption of the newer equation to estimate glomerular filtration rate (GFR), including effects upon CKD disease incidence and assessing any change to the current burden on clinical services. For individual patients, this could affect the monitoring of their disease progression and medication dosing. Furthermore, because estimating formulae use parameters which are themselves associated with variability in serum creatinine (age and gender), the extent to which moving from one formula to another affects eGFR may vary within patient groups.

In this study, we undertook a comparison of eGFR calculated by the CKD-EPI (2021) equation, the CKD-EPI (2009) equation, the four variable MDRD equation, and the European Kidney Function Consortium (EKFC) equation [17] using the Bland-Altman method to determine whether the tests are acceptably interchangeable. We sought to quantify the difference in estimated kidney function that would occur between the equations in the same patient. We assessed the whole population effect and the effect in subgroups of patients of different ages, sex, and ethnicity.

The department of Renal Medicine based at Salford Royal Hospital serves the northern and western sectors of the Greater Manchester conurbation of the UK, with a catchment population of 1.55 million. The proportion of Greater Manchester residents self-identifying as South Asian (14%) or African/Caribbean (5%) is closely reflective of England as a whole [18]. Greater Manchester suffers from above-average deprivation, with four local authority districts ranked among the twenty most deprived in England. Three of these local authority districts (Salford, Oldham, and Rochdale) have their renal services provided for by Salford Royal Hospital.

Pseudo-anonymised data were drawn from the electronic patient record of the Northern Care Alliance NHS Foundation Trust. Data were captured for individual adult (age ≥18 years old) patient episodes, occurring over a period of 4 years (2014–2018) which involved an acute inpatient admission to Salford Royal Hospital. Written informed consent from participants was not required in accordance with local guidelines as all data utilised were pseudo-anonymised. Parameters collected were those necessary to determine eGFR (creatinine, age, ethnicity, sex). The analysis was restricted to patients with an eGFR of 60 mL/min/1.73 m² or less calculated by the MDRD equation (the standard result currently provided by our biochemistry laboratory). The local creatinine assay used is traceable via the isotope dilution mass spectrometry reference method through correlation of patient samples and reference material SRM967 from the National Institute of Standards and Technology (NIST) of the USA.

We compared calculated eGFRs using the 4 equations for each patient (CKD-EPI [2021], CKD-EPI [2009], MDRD, and EKFC), using the Bland-Altman method [19]. In the EKFC equation, Q values for white Europeans were referenced from Pottel’s original paper; work from Delanaye’s group [20] provided the Q values for black patients in our study. In the absence of South Asian-specific Q values, we once again used the Q values for white Europeans in the EKFC equation. We studied the equations in subgroups based on age, sex, and ethnicity. The bias was calculated as the mean difference between the respective equations. Precision was calculated as the standard deviation (SD) of the mean difference.

Finally, we compared the CKD stage that each patient would fall into for each formula to determine if differences in the calculated eGFR between the equations would lead to a difference in CKD epidemiology.

From the original dataset (124,473 patients), a total of 16,861 patients were identified as suffering with CKD stage 3 or worse using the MDRD equation. Of these patients, 118 self-identified as African or Caribbean (hereafter referred to as black), 254 self-identified as South Asian, and 16,489 self-identified as white British or white Irish (hereafter referred to as white). As might be expected, there was a high prevalence of hypertension and type 2 diabetes. Over half of all patient groups suffered with hypertension. At least a third of all patient groups suffered with type 2 diabetes in all demographic groups, rising to over two thirds in the South Asian patient group. A full breakdown of patient characteristics can be found in Table 1.

Owing to the much larger number of white patients included in the study, the analysis in this patient group was performed within the following age groups: 18–39 years, 40–59 years, 60–79 years, and over 80 years. The mean age of the patients in our cohort was 77.5 years (±13.4 years).

The mean eGFR calculated from this CKD patient dataset using the CKD-EPI (2021) equation was 45 (±14) mL/min/1.73 m². Using the reported eGFR (MDRD), mean was 42 (±12) mL/min/1.73 m². Here, Bland-Altman bias was +2 mL/min/1.73 m², and precision was 2.3 mL/min/1.73 m². By comparison, CKD-EPI (2009) had a mean eGFR of 42 (±13) mL/min/1.73 m² (bias +3 mL/min/1.73 m², precision 1 mL/min/1.73 m²). Mean eGFR by the EKFC equation was 39 (±12) mL/min/1.73 m² (bias +6 mL/min/1.73 m², precision 3 mL/min/1.73 m²).

When moving to CKD-EPI (2021), all equations showed a lowering of mean eGFR in the black cohort of patients included in the study (excepting EKFC in black females). This effect was most pronounced when moving from MDRD and lowest when moving from EKFC (see Table 2). MDRD had the widest limits of agreement, and CKD-EPI (2009), the narrowest.

All equations showed a higher population mean eGFR when compared against CKD-EPI (2021) in South Asian patients included in the study. MDRD had the largest absolute change in mean eGFR, and EKFC demonstrated the largest relative change in mean eGFR by percentage (see Table 2).

As with the black patient cohort, MDRD had the widest limits of agreement, and CKD-EPI (2009), the narrowest. Plotting the results together elegantly illustrates the changes seen in mean eGFR depending upon equation used across both black and South Asian patient groups (see Fig. 1a–c).

For white patients in our study, MDRD demonstrated a higher mean eGFR when compared to CKD-EPI (2021). Both CKD-EPI (2009) and EKFC showed a lower mean eGFR, with EKFC having the largest relative change in mean eGFR in both absolute and percentage terms (most notably in white female patients). As in previous groups, CKD-EPI (2009) had the narrowest limits of agreement (see Table 2).

When white patients in our study were separated by age, the higher mean eGFR seen in the MDRD equation fell as patients aged (Tables 3, 4). The opposite was seen in both CKD-EPI (2009) and EKFC, where the higher mean eGFR increased as patients aged; this was most pronounced in white females over 80 years old. All equations studied had a lower mean eGFR when compared to CKD-EPI (2021). With regards to precision, CKD-EPI (2009) once again performed well, having the narrowest limits of agreement of the three equations. The limits of agreement of the MDRD equation narrowed as mean eGFR fell. In parallel, the limits of agreement of the EKFC equation widened as mean eGFR increased.

Comparing white male and females in our study, a higher mean eGFR and wider limits of agreement were seen in white females (compared to males) when both MDRD and EKFC were studied against CKD-EPI (2021). White males were not so fortunate when CKD-EPI (2009) was studied against CKD-EPI (2021); on this occasion, they had a higher mean eGFR and wider limits of agreement compared to white females.

Figures 2 and 3 demonstrate the changes in mean eGFR graphically; as one might expect, there were far more patients in the 60–79 and over 80 years categories. However, the trends in mean eGFR changes can still be appreciated.

When assessed by CKD stage, all equations exhibited a positive change in mean eGFR, indicating a higher population mean eGFR when compared to the CKD-EPI 2021 equation. In all cases, the positive change in mean eGFR was highest in patients with CKD 3a and lowest in patients with CKD 5. EKFC showed the largest positive change in mean eGFR (7.6 mL/min/1.73 m²) in CKD 3a patients; MDRD, the lowest (2.4 mL/min/1.73 m²) (see Table 5). Equation limits of agreement were similarly widest in patients with CKD 3a and narrowest in patients with CKD 5. CKD-EPI (2009) once again demonstrated the narrowest limits of agreement across all CKD stages, with EKFC and MDRD displaying similar performance.

CKD staging by eGFR was compared using both equations. In the overall dataset of 16,861 patients, a total of 1,671 patients (10%) had their CKD stage changed when moving from MDRD to CKD-EPI (2021). The proportions increased when moving from CKD-EPI (2009) (2,350, 14%) and EKFC (4,649, 28%).

For South Asian patients in our study, 12–25% of patients moved CKD stage depending upon the equation used. In keeping with the positive change in mean eGFR, more patients moved to a numerically higher CKD stage. This was most pronounced when moving from MDRD to CKD-EPI (2021) (Tables 6-8).

It is important to note here that, in the absence of proteinuria, microalbuminuria, other markers of kidney damage, or structural abnormalities, an eGFR >60 mL/min/1.73 m² would not be defined as CKD. This is especially true within the primary care setting.

The movement between CKD stages was less evident in black patients in our study (Table 6). Between 2 and 18% of patients moved CKD stage. The general trend of a negative change in mean eGFR in all equations studied meant that more patients were seen to move to a numerically lower CKD stage. This was most obviously seen when MDRD was studied against CKD-EPI (2021).

The earlier findings related to changes in mean eGFR impacted the white patient groups in our study in a similar fashion to the minority ethnic groups, with anywhere from 7 to 64% of patients having their CKD stage reclassified (Tables 9-11). Moving from EKFC to CKD-EPI (2021) resulted in over half of older (80+) white patients changing CKD stage. Such change was far less evident when moving from either MDRD or CKD-EPI (2009) to CKD-EPI (2021). In the case of MDRD, the largest proportion of CKD stage changes were seen in younger male and female patients. Across all three equations, patients were nearly always reclassified into a numerically higher CKD stage.

Two themes emerge in the results of this study: the first was the sheer proportion of patients undergoing reclassification of their CKD stage, owing to a change of equation; the second is that changes in mean eGFR was seen in all groups studied, but this was more notable in certain patient groups (specifically young white patients, black patients, and South Asian patients). The implications of both these themes are also noteworthy: patients may have been given an incorrect CKD staging and that differences in mean eGFR may have, in a small number of instances, resulted in interventions (such as kidney replacement therapy, including kidney transplantation) being offered earlier or later than necessary in certain patient subgroups. Of equal importance is the implication that a significant number of patients may have been prescribed statins and ACE inhibitors (or other antihypertensive agents) to help manage their misclassified CKD.

With somewhere between 10 and 28% of all patients having their CKD stage reclassified, there are serious implications for the patient burden on secondary care in the NHS. The most recent figures from Public Health England suggest that 2.6 million patients over 16 years old in England have CKD [21]; should our findings be reflected nationally, then between 260,000 and 728,000 patients would have their disease stage changed. In our cohort, there is potential for a considerable reduction in the frequency of clinic attendance and disease monitoring as a consequence of improved CKD staging.

The negative mean change in eGFR in black patients indicates that all other equations give a higher estimated GFR when compared to the updated CKD-EPI (2021) equation. These patients may well have worse kidney function than was originally thought, now strongly implied using the CKD-EPI (2021) equation. The delay in initiation of kidney protective strategies could potentially explain the higher incidence of kidney replacement therapy seen in black patients with CKD [22] and may also go some way to explaining in part the health inequalities seen in the black ethnic minority community [23].

The positive mean change in eGFR seen in the South Asian patient cohort results in fewer patients being categorised as having advanced CKD (stage 3a or worse). These patients may have been attending specialist clinics unnecessarily and have been denied investigations (such as contrast-enhanced cross-sectional imaging), owing to an incorrectly perceived CKD stage. The narrow limits of agreement seen when moving from CKD-EPI (2009) to CKD-EPI (2021) may reflect similarities in the derivation equations and validation datasets used by the CKD-EPI investigators.

There may be no single reason for the large positive change in mean eGFR seen in our older white male and female patients when EKFC and CKD-EPI (2021) are compared. The changes seen may reflect the heterogeneity among white populations in both Europe and North America. In addition, the mean age of the internal and external validation datasets for the EKFC study group was 42 and 51 years, respectively [17]. Mean measured GFR (mGFR) in the EKFC internal and external validation datasets was, respectively, 78 and 79 mL/min/1.73 m². For CKD-EPI (2021), mean age in the validation dataset was 57 years and mean mGFR, 77 mL/min/1.73 m². This evidently contrasts with our dataset where mean age was 78 years and mean eGFR was 45 mL/min/1.73 m² CKD-EPI (2021). CKD-EPI (2009) was developed in response to the MDRD equation being found to be imprecise with systemic GFR underestimation in patients with an mGFR >60 mL/min/1.73 m²; our inclusion of patients with an MDRD eGFR >60 mL/min/1.73 m² could then be regarded as a potential error bias, although the findings for CKD-EPI (2009) and EKFC should still stand.

The mean age of the patients included in the MDRD study was 51.6 years (±12.7 years) with 12% of patients being classified as African American and no other ethnic minority groups recorded [7]. By contrast, CKD-EPI (2021) equation was derived with reference to a dataset with a mean age of 47 years (±15 years) with 32% of patients self-reporting their racial group as African American and 1% as Asian [24]. One problem we may have in the UK is the use of an estimating equation validated in an American population which does not closely reflect UK patient demographics. The lower mean eGFR and higher limits of agreement in the older white male and female patient groups may simply relate to the larger patient numbers in the group we have studied and result from mathematical artefact.

Another problem with our work lies in the demographics of the dataset analysed, which does not closely reflect the population served with respect to ethnic makeup (1.7% South Asian and 1.1% black compared to 13.6% and 4.7%, respectively, in Greater Manchester). Mean age and eGFR among the minority ethnic groups differs from the white patient groups (unpublished), and therefore, the changes seen may not be solely reflective of ethnicity. We do not have records of patient height or weight, which would help evaluate the influence of any sarcopenia upon our findings, particularly in the older patient cohort. We could expand our dataset (and potentially strengthen our findings) by evaluating patients with evidence of albuminuria/proteinuria and eGFR >60 mL/min/1.73 m². Furthermore, we do not have a “gold standard” mGFR available for the patients included in the dataset. This means that we were not able to compare the true accuracy of any of the equations studied relative to mGFR.

The recommendation to move to a race-free estimating equation for eGFR has resulted in a rich seam of research looking at the implications of any change upon racial disparity and CKD outcomes [25, 26], the effects in other nations [27, 28], and the impact upon kidney transplantation waiting lists [29]. To our knowledge, this is the first study looking at the impact upon the UK population, validating many of the findings in black patients and perhaps prompting more questions to answer in white patients and South Asians. For white patients with CKD, our results are in largely keeping with the work done in Denmark by Vestergaard’s group [30].

Periodic monitoring of mGFR, rather than eGFR, is the answer to these problems, but we are currently some way from achieving this goal (although there is now an increasing awareness and availability of point of care mGFR testing). Caution should be exercised and consensus sought when recommendations or guidance around use of estimating equations is issued by authorities. Nephrologists must remain cognizant of the limitations of all estimating equations and the implications of moving from one equation to another. An estimating equation with low Bland-Altman bias and high Bland-Altman precision across multiple ages and ethnicities is the goal. Hopefully, the day will soon come when such an equation is reported.

This retrospective review of patient data did not require ethical approval in accordance with local guidelines. Written informed consent from participants was not required in accordance with local guidelines as all data utilised were pseudo-anonymised.

The authors declare no conflicts of interest.

This study was not supported by any sponsor or funder.

Darren Green and Maharajan Raman originally proposed assessing the impact of any equation change. Reuben Roy obtained and analysed the data, drafted the first version of the manuscript, and revised the work with the equal help of Paul M. Dark, Philip A. Kalra, and Darren Green.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from Reuben Roy upon reasonable request.