Body Surface Area

Our goal was to compare the current CKD-EPI eGFRcr, eGFRcys, and eGFRcr-cys equations with equations developed with the use of two new approaches for GFR estimation that do not involve race.^{5 (link),9 (link)} As we described previously, the current approach to the development of CKD-EPI equations has been to model eGFR with the use of least-squares linear regression to relate log-transformed measured GFR to log-transformed filtration markers, age, sex, and race with separate slopes for higher as compared with lower levels of creatinine and cystatin C.^{5 (link),9 (link)} Race is an explanatory variable in the current eGFRcr and eGFRcr-cys equations but not in the current eGFRcys equation.
The first set of new equations uses the same coefficients for the intercept, age, sex and creatinine level as in the current eGFRcr and eGFRcr-cys equations but removes the Black race coefficient in computing eGFR, thereby assigning the eGFRcr and eGFRcr-cys values for non-Black persons to Black persons. For the second set of new equations, we fit new models using eGFRcr and eGFRcr-cys by means of the same regression function as the current equations but without inclusion of race as an explanatory variable. In total, we evaluated seven equations (three current and four new equations). Because all equations were developed by the CKD-EPI research group, we refer to them only by the filtration marker or markers (creatinine [eGFRcr], cystatin C [eGFRcys], or creatinine–cystatin C [eGFRcr-cys]) and the demographic factors (age, sex, and race [ASR] or age and sex [AS]) that were used in their development. We use the term non-Black (NB) to refer to ASR equations that were fit with a race term but in which the Black race coefficient was removed for computation of eGFR. Additional details are provided in the Methods section in the Supplementary Appendix.
In the development data sets, we assessed bias (systematic error) as the difference between measured GFR and eGFR and assessed model fit using root-mean-square error.^{5 (link),9 (link)} In the validation data set, we assessed accuracy overall and within race groups as bias, percentage of estimates less than 30% different from measured GFR (P₃₀, with 1−P₃₀ corresponding to large errors that may be clinically significant), and agreement of eGFR with measured GFR categories using guideline-recommended GFR stages (<30, 30 to 44, 45 to 59, 60 to 89, and >90 ml per minute per 1.73 m² of body-surface area).⁸ A P₃₀ value of 80 to 90% is considered to be acceptable for GFR evaluation in many circumstances, and a P₃₀ value of 90% or higher is preferred; these values correspond to approximately 60 to 70% agreement and more than 70% agreement of eGFR with measured GFR in GFR categories, respectively.^{5 (link),8 ,9 (link)} We also focused on differential bias (systematic differences) between race groups because it could lead to systematic differences in treatment for the same measured GFR level. Confidence intervals for bias were calculated by means of bootstrap methods. We assessed accuracy in subgroups according to eGFR (as defined above), age (<40, 40 to 65, and >65 years), sex, and body-mass index (BMI, the weight in kilograms divided by the square of the height in meters: ≤25, 25 to <30, and ≥30).
In sensitivity analyses, we weighted the proportion of Black participants in the development data set from 0 to 100% to evaluate the effect on accuracy. In the validation data set, we calibrated measured GFR to account for differences between measurement methods as compared with the development data sets,^{5 (link),9 (link),23 (link)} and we compared equations that were developed by other research groups to estimate GFR in adults.^{24 (link)–28 (link)}