#NephJC Chat
Tuesday Oct 12 9 pm Eastern
Wednesday Oct 13 9 pm IST
Wednesday Oct 13 9 pm BST
N Engl J Med 2021 Sep 23. doi: 10.1056/NEJMoa2103753. Online ahead of print.
Race, Genetic Ancestry, and Estimating Kidney Function in CKD
Chi-Yuan Hsu, Wei Yang, Rishi V Parikh, Amanda H Anderson, Teresa K Chen, Debbie L Cohen, Jiang He, Madhumita J Mohanty, James P Lash, Katherine T Mills, Anthony N Muiru, Afshin Parsa, Milda R Saunders, Tariq Shafi, Raymond R Townsend, Sushrut S Waikar, Jianqiao Wang, Myles Wolf, Thida C Tan, Harold I Feldman, Alan S Go, CRIC Study Investigators
PMID: 34554660
N Engl J Med 2021 Sep 23. doi: 10.1056/NEJMoa2102953. Online ahead of print.
New Creatinine- and Cystatin C-Based Equations to Estimate GFR without Race
Lesley A Inker, Nwamaka D Eneanya, Josef Coresh, Hocine Tighiouart, Dan Wang, Yingying Sang, Deidra C Crews, Alessandro Doria, Michelle M Estrella, Marc Froissart, Morgan E Grams, Tom Greene, Anders Grubb, Vilmundur Gudnason, Orlando M Gutiérrez, Roberto Kalil, Amy B Karger, Michael Mauer, Gerjan Navis, Robert G Nelson, Emilio D Poggio, Roger Rodby, Peter Rossing, Andrew D Rule, Elizabeth Selvin, Jesse C Seegmiller, Michael G Shlipak, Vicente E Torres, Wei Yang, Shoshana H Ballew, Sara J Couture, Neil R Powe, Andrew S Levey, Chronic Kidney Disease Epidemiology Collaboration
PMID: 34554658
Additional Reading:
Last NephJC discussion on race and GFR
Summary: http://www.nephjc.com/news/raceandegfr
Podcast: http://www.nephjc.com/freelyfiltered/2020/8/20-freely-filtered-023-race-and-egfr
NephJC discussion on cystatin C http://www.nephjc.com/news/cystatinc
Original CrCl/GFR equation papers
Cockroft and Gault, Nephron, 1976
MDRD Equation, Annals of IM, 1999
CKD-EPI, Annals of IM, 2009
CKD-EPI 2011 validation, KI
CKD-EPI 2012 with cystatin C, NEJM 2012
Blog post at PBFluids on the controversy
Introduction
In 2020, the National Kidney Foundation and the American Society of Nephrology formed a joint task force to address the use of race in eGFR prediction equations. This came about as there was increasing awareness that the coefficient added for being an African American was not as solid as had been assumed. Accuracy apart, it was additionally responsible for delayed care and decreased primary transplants among African Americans with CKD. This was discussed in a previous NephJC chat and podcast. The coefficient results in a systematic bias upwards, which has different consequences as shown in the table below:
Additionally, the big question is how one determines race. Should it be self-identified? In that case, should patients know about the downstream consequences of this seemingly innocuous choice? How accurate is self-identified race anyways? Skin colour does not determine race - and sometimes that is used for these clinical calculations. Lastly, how does one deal with mixed-race individuals?
Another issue that is less appreciated is the relevance of these equations outside the US. Most of the data in these cohorts include African Americans from the AASK study (around 70%). Apart from the particular socio-demographics this represents, this cohort was mostly of people with CKD, and not preserved GFR, and it is not clear how creatinine was calibrated at that time, all of which are important points discussed in greater depth in a recent review article (Delanaye et al, Curr Op Neph HT, 2021). Hence the addition of the coefficient actually results in a more inaccurate GFR when studied outside the US (Bukabaye et al Kidney Int 2019; Rocha et al Int J Nephrol 2020; Moodley et al, Clin Biochem 2018; Delanaye et al, CJASN 2011).
With that in mind, let us review two studies that came out simultaneously. In the study from CRIC, we look at the impact of race and ethnicity on models for GFR based on either creatinine and/or cystatin C. Secondly, the original CKD-EPI group used some newer data to derive race-free formulas for eGFR.
Time to dig deeper.
The CRIC Study
Methods
The Chronic Renal Insufficiency Cohort (CRIC) study is a multicenter, prospective, observational study. 1423 of the CRIC participants were randomly selected to undergo measurement of urinary iothalamate clearance as a direct measurement of GFR (iGFR). The study population included 1248 of these participants who had the following data collected:
Race as reported by the participant
Genetic ancestry markers
Serum creatinine
Serum cystatin C
24 hour urinary creatinine levels
The overall goal of the study was to determine whether GFR could accurately be estimated without including race; three alternatives were considered as outlined in the conceptual map below.
Statistical analysis
Rather than use the existing equations, the authors derived GFR estimating equations using
Age, sex, serum creatinine
Adding race (self-identified) and then based on genetic ancestry
Same exercise with cystatin C rather than creatinine
The entire dataset was randomly divided into development and validation data sets for this purpose.
In addition, they also performed linear regression to examine the association of race or African ancestry with measures of body composition, protein intake, 24-hour urinary excretion, as well as tubular secretion of creatinine, to better understand what drives this difference and a potential need for a race coefficient.
Results
Fig S1 shows how the cohort was assembled
Table 1 shows the characteristics of the included population. One can see how well or not the genetic ancestry tallies with the self-identified race, along with the other socio-demographic, kidney function, and body composition variables.
The race and ancestry can perhaps be better appreciated in this figure from the supplement.
How well did the new equation fare? Just using age, sex and serum creatinine (without race or ancestry) provides a biased estimate compared to all other alternatives (see top left panel below).
Using cystatin C, there is no difference whether race is considered or not (bottom panels).
The other considerations are that of accuracy. As can be seen below, not including race with serum creatinine is the least accurate. One should note, however, that if you are looking for accuracy to within 10% (P10) none of the equations are that great.
What could explain this discrepancy with serum creatinine? Table 3 shows the body composition and other variables considered. This did result in a small attenuation of the difference from 12.8 to 8.7, but were unable to eliminate the difference in the relationship between creatinine and GFR in Black and non-Black people. There is some information that serum creatinine carries, that is not explained away after adding urinary creatinine excretion, fat-free mass, and other anthromorphic measures. See the accompanying NephJC commentary for a longer discussion of this issue.
The authors conclude that if one is using serum creatinine, then not using race gives you less accurate information - even if one is considering several body composition factors - which are impractical to measure in routine practice.
However, using cystatin C makes the race coefficient redundant and unnecessary.
New GFR Equation
Methods
First we had Cockroft and Gault, then MDRD, then CKD-EPI (2009) and CKD-EPI (2012, with cystatin C) and now we have the new improved version for 2021. Let us see how it stacks up and whether it truly is an improvement and solves more problems than it creates.
For any prediction equation, one needs a development and a validation - two different datasets. For this, they mostly used the old datasets along with a few more studies in slightly different combinations as shown below.
Otherwise the formulae themselves were derived using the same methods, least-squares linear regression, but without including the race coefficient. However, bias and accuracy were reported by race - where being Black or non-Black was self-reported, mostly (table S3 has the details on how it was actually done for each included study).
The terminology is also somewhat hard to follow, so here goes:
eGFRcr: formula with creatinine (not cystatin C)
eGFRcys: equation with cystatin C (not creatinine)
eGFRcr-cys: equation with both creatinine and cystatin C
Then the demographic factors get added on
ASR: equation with age, sex and race
AS: age and sex (but not race)
And more
NB refers to ASR equations that were fit with a race term but in which the Black race coefficient was removed for computation and all people are treated as non-Black
Evaluation of these equations was done and is presented in many different ways.
Bias or systematic error refers to the mean difference between the estimated eGFR and the measured GFR (closer to 0 is better; a negative number means eGFR is an overestimate and a positive number means eGFR is an underestimate)
P30 refers to the percentage of estimates which were within 30% of the measured GFR (closer to 100% is better)
Classification refers to agreement between eGFR and measured GFR for the different CKD staging categories
Lastly, there is a concern that changing the eGFRs for a huge swathe of population would change the immediate prevalence, and subsequent incidence of CKD. So the authors modeled this using the 1999 - 2002 NHANES and 2019 US Census data.
Results
What do the formulas look like? Table 2 gives the coefficients for those who like to see them and how it all panned out. The actual equations are laid out in a supplementary table (table S10) for those who want to code it all in.
The graphs below are useful to understand the bias (systematic error). In panel A, one can see that the existing equation (which includes the race coefficient) overestimates GFR in Black individuals by 3.7 ml/min/1.73m^2 and has a minimal bias in non-Black individuals. Removing the race coefficient - as some hospital systems have been doing in recent years - changes this to an underestimation by 7 ml/min/1.73m^2. A new equation, same for everyone, makes the bias ~ 3.6 - 3.7, but in different directions for Black and non-Black individuals. The P30 (percentage of eGFR within 30% of measured GFR is between 85 and 90% for all three versions in Black and non-Black individuals. Lastly the classification (by CKD staging) is lowest by merely removing the race coefficient from the old formula, and similar between the old formula with the race coefficient and the new formula (which does not have any race coefficient).
Panel B shows the same calculations using creatinine + cystatin C equations. Interestingly enough, the newest equation: with creatinine, cystatin C and no race coefficient, performs very well, with a very small bias, P30 just over 90% and almost 70% agreement on CKD stage classification.
Lastly, how about just using cystatin C? Again the bias is very low (even lower than creatinine + cystatin C for non-Black individuals) and the P30 is in the high 80s with 63-66% agreement on classification.
What are the implications of these new equations for CKD prevalence in the US? It means that some previously healthy individuals will suddenly receive a new CKD diagnosis - which, if it is an accurate diagnosis, may lead to better care and management. How many of these people are there? The actual numbers depend on which equation is used, and can be seen below. The new eGFR-cr (AS) results in about 700,000 Black individuals being diagnosed as having CKD, but also over 3 million non-Black individuals being cured of CKD. At the other extreme, the new eGFR-cr-cys (AS) which is possibly the most accurate when it comes to minimizing bias, improving precision and classification, will not result in a big change in numbers for Black individuals while reducing the number of non-Black individuals with CKD.
Discussion
The old equations have a systematic bias and lead to under recognition of CKD in a community in which CKD prevalence is high, progression is rapid, and transplant rates have been low, thus exacerbating existing disparities.
This change in equations is long overdue. There are multiple problems with using a race coefficient, as we have discussed above, and previously. There are issues about how race is determined, and how accurate these have been. The old equations have a systematic bias and lead to under recognition of CKD in a community in which CKD prevalence is high, progression is rapid, and transplant rates have been low, thus exacerbating existing disparities. As the accompanying commentary (see here, from Raj Mehta) discusses, there are many socioeconomic factors that play a role, and we are ignoring them in the GFR accuracy discussion. Will having new equations really solve this?
There is little doubt that the new equations are a step in the right direction - and are also better than the current compromise (using the 2009 CKD-EPI but without any race coefficient). However, one has to also accept that the CRIC study shows that creatinine might not be the perfect marker after all. Cystatin C has been around now for over a decade. It is not affected by tubular secretion, it is not affected by race, and it is a better marker of subsequent bad outcomes (see NephJC summary of study from Lees et al, Nat Med 2019). Right now it is more expensive, but the equations incorporating cystatin C seem to be the best of the lot. Should we change now from the 2009 to the 2021 creatinine equations (which will be a big change logistically) or be even more bold and take a bigger step to cystatin C?
The 65-70% agreement in correctly classifying CKD might give you some pause. But even the CKD stages are based on arbitrary thresholds (do you really think a GFR of 58 and 62 are all that different?) We should keep our eyes on the ball here. Albuminuria, for example, is hugely important and it’s presence or extent is often way more important than a few mls of GFR accuracy (see NephJC discussion here).
There are other opinions and discussions to have. These equations have been validated for US populations, and even the 2009 CKD-EPI has not fared well without additional tweaks for other countries and ethnicities. If not anything else, we will see a cottage industry of validation studies in other places and programs. Possibly more papers with alternate markers (see table below from Levey et al, CJASN 2020).
Lastly, one should consider how accurate these numbers are? Beyond the mean bias, the P30 is about 85-90%, so about 10-15% of individuals will get an eGFR that is more than 30% away from their true GFR. That eGFR of 33? It could be anywhere from 23 to 43 for 85% of your patients. And even higher or lower for the remaining 15%. This shows how foolish it is to make decisions based on dichotomanic hard line thresholds, such as stopping metformin at 30, or listing for transplant at 20. The seemingly good 70% agreement on classification means the accuracy will be so bad that you will get the wrong CKD stage in a whopping 30% of patients! If the exact, precise number really matters, equations cannot replace measurement. Most importantly, new equations don’t mean our mission in fixing the disparities in kidney disease and health outcomes is done. This is just a start.
Summary by
Laura Slattery,
Nephrology Specialist Registrar,
Cork, Ireland
Joel Topf,
Nephrologist, Detroit
and Swapnil Hiremath,
Nephrologist, Ottawa