Towards equations for estimating glomerular filtration rate without demographic characteristics

Dear Editor, Estimated glomerular filtration rate (eGFR) is an essential index for identifying and monitoring chronic kidney disease (CKD). For estimating GFR, the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) research group recently proposed two new serum creatinine-based equations without race, eGFRcr(AS) and eGFRcr-cys(AS), to replace the widely used equations eGFRcr(ASR) and eGFRcr-cys(ASR), that considers serum creatinine, cystatin C (CysC), age, sex and race.1-4 The new equations were developed and validated using data mainly from White and Black participants.4 It remains unknown whether they are applicable to plasma samples as well as to non-White and non-Black subjects. The present study addressed these two questions with a focus on two Chinese CKD cohorts and showed that equations based on other blood metabolites could perform better. The two cohorts comprised 52 patients from Guangzhou (Third Affiliated Hospital) and 135 patients from Macau (Kiang Wu Hospital) for whom serum samples and plasma samples were available, respectively (Tables S1 and S2 for their characteristics). Performance of the equations was examined with reference to the GFR measured by iohexol plasma clearance (mGFR).2, 4 The bias and precision of the current and new CKD-EPI equations were similar for the two cohorts (Table 1). The accuracy of all the equations was better for the Macau cohort, as revealed by lower values of the root mean square error and higher values of the percent agreement within 30% of the mGFR (P30), the percent agreement within 20% of the mGFR (P20) and the percent agreement between the mGFR and eGFR categories.5 As plasma samples were used in the Macau cohort, our results suggested that both old and new CKD-EPI equations should be applicable to plasma samples. In the subsequent analyses, the data from the two Chinese cohorts were pooled for a larger sample size. A P30 value of 80%–90% is considered acceptable, and a P30 value of 90% or higher is preferred.4 However, P30 values of two new equations without race were less than 80% for the two Chinese cohorts (Table 1, Result S1 for statistical test results). LOWESS curve fitting of the agreement between the mGFR and eGFR indicated that all the serum creatinine-based equations overestimated the GFR in the range of 30–90 ml/min/1.73 m2 (Figure 1). Overestimation was the highest for the eGFRcr(AS) equation (Figure 1B), whereas it was not obviously for the eGFRcys(AS) equation (Figure 1E). In a recent study, applying the eGFRcr(AS) equation to the White European population also resulted in shifting a major proportion of CKD patients to a higher eGFR category.6 In addition to the unsatisfactory P30 values, the use of the new equations could lead to underestimation of the disease severity in Chinese patients with mild or moderate loss of kidney function. Therefore, we recommend not to use the new equations without race for Chinese CKD patients. To minimize the effect of systemic and systematic biases,6, 7 to ensure a metabolite marker can be applied to both serum and plasma specimens,8 and to avoid false-positive biomarkers,9 we adopted a stringent two-centre study design (Figure 2), involving plasma samples from 10 healthy volunteers in addition to the two Chinese cohorts (Table S3, Discussion S1 about study designs). The samples were subjected to untargeted metabolomics profiling using a Metabolon's Discovery HD4 platform.10 The final list of putative biomarkers contained 212 metabolites (207 negatively and 5 positively correlated with the mGFR, Figure 2 and Result S2 for details). Among the top 20 putative biomarkers (Table S5), hydroxyasparagine and N,N-dimethyl-proline–proline had not been previously shown to be inversely associated with kidney function (Table S6, Discussion S2 about accessibility in clinical practice). Furthermore, one of the top 20 putative biomarkers was creatinine, indicating the success of our study design. For proof-of-concept, we attempted to construct eGFR equations from the data of the top 20 putative biomarkers in the discovery dataset. S-Adenosylhomocysteine, gluconate, N6-succinyladenosine and hydroxyasparagine (p values <.05) were retained in the final regression equation named CKD-msMET4a (Table 2). As gluconate is a common component of food additives and supplements, the performance of CKD-msMET4a could be affected by gluconate intake. We hence constructed another equation named CKD-msMET4b with the exclusion of gluconate. The major difference from the first equation was the replacement of gluconate with gulonate (Result S3 for additional information). Gulonate and gluconate are stereoisomers of each other, indicating that the two equations are almost identical in nature. Further inclusion of age, sex, height, weight and/or serum/plasma CysC did not significantly improve the two equations (p values >.05). Using the data from the independent validation cases, the performance of the two equations was compared with that of the CKD-EPI eGFR equations (Figure 2). Concerning bias, precision, accuracy and GFR category agreement, either CKD-msMET4a or CKD-msMET4b equations appeared to be one of the best two equations although the statistical power was insufficient (Table 2, Result S4 for detailed results). The P30 values of both equations were 82%. They were not significantly different from those of the current eGFRcr-cys(ASR) equation (p-values >.890) but were significantly different from the P30 value of the new eGFRcr(AS) equation (p-values <.05). LOWESS curves revealed that the overestimation of GFR in the range of 30–90 ml/min/1.73 m2 was not observed for the CKD-msMET4a and CKD-msMET4b equations (Figure S1, Result S4 for detailed results). Although the male and female patients might have differences in age and total mass muscles, GFRs estimated by the two equations were not significantly different between the male and female patients (p-values <.005, Table S7, Result S4 for detailed results). This suggests that four metabolites could provide sufficient information for estimating the GFR without demographic characteristics. This may also suggest that these two equations are applicable to patients of other races, such as the White and Black populations. In conclusion, the new CKD-EPI creatinine-based equations without race could lead to underestimation of the disease severity in Chinese patients with mild or moderate loss of kidney function. However, our results suggest the possibility of developing new metabolite equations for estimating the GFR without demographic characteristics. This work was supported by The Macao Science and Technology Development Fund (FDCT) (Grant Number: 0032/2018/A1). We gratefully acknowledge the contribution of all members of field staff who were involved in planning and conducting this study. We thank all participants in our studies for their donation of blood and time. We also thank the Calibra-Metabolon Joint Laboratory (Hangzhou, China) for their support. Finally, we thank Yuanyan Tang for support with the study data. This work was also supported by the Faculty of Health Sciences of the University of Macau and Kiang Wu Hospital Charitable Association, Macau. The authors declare no conflict of interest. The datasets used and analysed during the current study are available from the corresponding authors on reasonable request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.