Time to Eliminate Health Care Disparities in the Estimation of Kidney Function

In the wake of the racial reckoning since the spring of 2020 in the United States, efforts have emerged to identify hidden structural determinants of systemic racism in medicine. Such efforts have led to examination of traditional medical algorithms that incorporate race modifiers and may have led to disparities in health outcomes according to racial and ethnic group. Prominent among these efforts has been reconsideration of race-based adjustments in equations that are used to calculate the estimated glomerular filtration rate (eGFR) in the assessment of kidney function. Approximately 90% of U.S. medical laboratories report the eGFR along with the serum creatinine concentration, and the use of a Black race coefficient in calculation of the eGFR has engendered numerous debates. Consequently, many institutions have already stopped using the race adjustment.

Some have argued that the way in which estimating equations for GFR were developed — they were initially derived from data on White persons1 — doomed them to be inequitable from their inception. When GFR was directly measured from endogenous clearance of creatinine or from exogenous substances such as iothalamate, there was a difference in GFR between Black and non-Black persons. That observation led to incorporation of a race coefficient in estimating equations that would inflate GFR estimates for Black patients in order to generate unbiased GFR estimates in Black and non-Black populations.2,3 However, concern has emerged that the inclusion of race as a coefficient in eGFR equations inequitably inflates GFR estimates in Blacks and lacks a biologic basis.4

One year ago, the National Kidney Foundation (NKF) and the American Society of Nephrology (ASN) formed a joint task force to develop a best-practice recommendation regarding the use of race in eGFR prediction equations. The final version of that recommendation is now being published.5 Now in the Journal, two investigator groups report on studies that address the use of race in the estimation of kidney function.6,7

Inker et al.6 formulated and cross-validated new GFR equations that eliminate the race coefficient, and they compared them with currently used equations — the 2009 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, which is based on the serum creatinine concentration (eGFRcr),3 and the 2012 CKD-EPI equation, which is based on both the serum creatinine concentration and the serum cystatin C concentration (eGFRcr-cys).8 Hsu et al.7 had a different, but similar, approach. They used a sample from the Chronic Renal Insufficiency Cohort study database to fit regression models of measured GFR with age, sex, and either serum creatinine or cystatin C as predictors of GFR. They then compared the accuracy and predictive bias (the difference between eGFR and measured GFR) of models that added either Black race or the percentage of African ancestry as a predictor.

Inker et al. found that the currently used 2009 CKD-EPI eGFRcr equation that includes age, sex, and race3 overestimated measured GFR in Black participants by a median of 3.7 ml per minute per 1.73 m2 of body-surface area (95% confidence interval, 1.8 to 5.4) and had negligible bias in non-Black participants. Agreement between measured GFR and eGFR within chronic kidney disease (CKD) categories in that model was 63.2% in Black participants and 68.5% in non-Black participants. The 2012 CKD-EPI eGFRcr-cys equations8 had less bias and higher percent agreement than the corresponding eGFRcr equations.

It is crucial to understand how the newly proposed equations “eliminate” the race coefficient. The use of the current CKD-EPI equations without application of the inflation factor for Black race (15.9% for the eGFRcr equation and 8.0% for the eGFRcr-cys equation) does not affect the accuracy of eGFR estimates for non-Black persons, but it underestimates the GFR and reduces accuracy for Black persons. Another approach is to refit the eGFR equation without using race as a predictor, which leads to new values for the other coefficients. Analyses presented in both articles suggest that refitted creatinine-based prediction equations underestimate GFR for Black persons by 3 to 4 ml per minute per 1.73 m2 and overestimate GFR for non-Black persons by 1 to 4 ml per minute per 1.73 m2. Relative to currently used CKD-EPI equations, the change in overall accuracy with refitted equations is minor, but race disparities in classification of the CKD stage may not be fully resolved (Table S14 in the Supplementary Appendix of the article by Inker et al., available with the full text of the article at NEJM.org).

Both articles showed that equations based on cystatin C yielded lower predictive bias and greater agreement with measured GFR across race groups than GFR estimation based on creatinine alone. Moreover, they showed that bias and agreement were not affected by the inclusion of race as a predictor.

It is essential to quantify the ability of new equations to improve GFR estimation and reduce racial disparities in the diagnosis of kidney disease. Both articles relied on measures of accuracy such as predictive bias (also called statistical bias or differential bias), the percentage of estimates less than 10% or less than 30% different from measured GFR (P10 and P30, respectively), and the correct classification (i.e., the percent agreement between measured GFR and eGFR categories within CKD stages). Although useful, these omnibus measures are difficult to interpret in a clinical context (e.g., the average bias of 3 ml per minute per 1.73 m2 is more meaningful at a low GFR than at a high GFR8). Context-specific measures such as the percentage of patients with CKD who would be misclassified are needed in order to understand the effect at the patient level.

The development of accurate predictions of GFR that do not rely on adjustment for Black race and avoidance of potential race-based disparity in the accuracy of CKD diagnoses have proved to be problematic in clinical practice. Inker et al. found that, relative to the currently used 2009 CKD-EPI eGFRcr equation, the same equation refitted without race had a similar percent agreement between eGFR and measured GFR within CKD stages but retained modest statistical bias. However, Hsu et al. found that in models of eGFR that were based on serum creatinine, exclusion of race-based predictors (i.e., Black race as reported by the participants or percentage of African ancestry) yielded increased predictive bias and diminished accuracy; furthermore, the effect of excluding race as a predictor could not be mitigated by replacing race with non-GFR determinants of the serum creatinine concentration. In contrast, race had no effect on the predictive accuracy of eGFR in equations that were based on cystatin C. Both articles point to the promise of cystatin C for uniformly more accurate GFR prediction without the need to include race-based adjustments.

The much-anticipated NKF-ASN task force report5 concludes that CKD-EPI equations that are refitted without race should be used in practice. The articles by Inker et al. and Hsu et al. provide evidence that equations based on cystatin C have greater predictive accuracy than those derived from serum creatinine. If the capacity to measure cystatin C routinely were widespread, cystatin C equations would become practical; thus, we suggest that the use of cystatin C measurements should be encouraged and funded.

Meaningful ways to alleviate health care inequities are overdue. That Black persons with CKD often lose kidney function more rapidly and have lower kidney transplantation rates than patients from other racial and ethnic groups indicates an urgent problem. The use of the most accurate estimates of GFR may permit earlier identification and care of persons at risk.9,10 Irrespective of the equations adopted, estimates of GFR are, by their very nature, imperfect. Some promising options will take time to implement, since measurement of cystatin C is currently neither routine nor uniform. Both existing and newly derived equations have strengths and weaknesses, and change inevitably induces unanticipated consequences. Most important, however, is that estimates do no harm but rather help us care for all patients equally.

留言 (0)

沒有登入
gif