Authors' Reply: Enhancing eGFRcr-cys Accuracy in Extreme Body Mass Index Groups: Addressing Body Composition and Methodological Heterogeneity

We thank Li et al.1 for their thoughtful letter and to JASN for the opportunity to respond. We agree that body mass index (BMI) is an imperfect surrogate for body composition. However, BMI is often the most accessible anthropometric measure in clinical care, and we sought to clarify eGFR performance across the BMI spectrum given the limited evidence on this relationship. As noted in our article,2 our study population reflects a referred group with comorbid conditions, where the relationship between BMI and underlying physiology may differ from that in other populations. The CKD Epidemiology Collaboration research group has validated both the combined 2012 eGFR calculated with creatinine and cystatin C (eGFRcr-cys) equation and the average of the 2009 eGFRcr calculated with creatinine only (eGFRcr) and 2012 eGFRcys calculated with cystatin C only (eGFRcys) equations.3 To address the second comment from Li et al.,1 we compared GFRcr-cys with the average of eGFRcr and eGFRcys in our study population. The overall median bias (95% confidence interval) was marginally but not meaningfully larger with average of eGFRcr and eGFRcys (1.7 [1.4 to 1.9] ml/min per 1.73 m2) compared with eGFRcr-cys (1.1 [0.0 to 1.3]). Differences remained minor across strata of low BMI (<20 kg/m2; 5.9 [5.1 to 6.8] versus 4.9 [4.0 to 5.8]) and high BMI (30–34.9 kg/m2: 0.3 [−0.4 to 0.9] versus −0.5 [−1.1 to 0.2]; 35–39.9 kg/m2: 0.4 [−0.9 to 1.9] versus −0.6 [−1.8 to 0.8]; ≥40 kg/m2: 0.1 [−1.9 to 1.9] versus −1.5 [−3.3 to 0.3]). Finally, precisely because comorbid conditions like diabetes are often more common at higher BMI in real-world contexts, we find it clinically valuable to characterize eGFR performance without adjustment for comorbid conditions. Because stratification is often favored as a more interpretable alternative to adjustment in this context,3 we compared median biases (95% confidence interval) in individuals with and without diabetes. We observed greater overestimation by eGFRcr (8.9 [8.4 to 9.5] ml/min per 1.73 m2 in individuals with diabetes versus 5.7 [5.3 to 6.1] in those without diabetes), greater underestimation by eGFRcys (−4.7 [−5.2 to −4.1] versus −2.7 [−3.1 to −2.3]), and similar bias for eGFRcr-cys (1.3 [0.9 to 1.7] versus 1.0 [0.7 to 1.3]). We also found that the relationship between BMI category and bias differed significantly between individuals with and without diabetes for eGFRcr (interaction P = 0.01) and eGFRcys (P = 0.03), but not for eGFRcr-cys (P = 0.55). However, we still observed that individuals without diabetes trended toward greater overestimation for eGFRcr and underestimation for eGFRcys in higher BMI categories. Among individuals with diabetes, eGFRcr bias was greater at BMI <20 and 20–29.9 kg/m2 but smaller (closer to 0) at BMI ≥40 kg/m2 compared with those without diabetes; eGFRcys bias was greater at BMI 20–24.9 kg/m2 in those with diabetes compared with those without.