Validation of the European Kidney Function Consortium Equation in Chinese Adult Population: An Equation Standing on the Shoulders of Predecessors Open Access

Glomerular filtration rate (GFR) has been widely accepted as an important indicator of renal function [1]. A precise assessment of GFR is crucial for diagnosing and staging chronic kidney disease (CKD) and might also significantly influence the drug dose adjustment and prognosis prediction [2]. However, considering the measurement of GFR by exogenous substances is invasive and complex [3], Kidney Disease: Improving Global Outcomes (KDIGO) guidelines recommended the use of equations based on endogenous biomarkers to calculate the estimated GFR (eGFR) [2]. Among them, serum creatinine (SCr) is inexpensive and convenient to use commonly.

Among a spectrum of SCr-based formulas, the CKD Epidemiology Collaboration (CKD-EPI) equation published in 2009 is suggested by KDIGO guidelines [2], but it is likely to overestimate GFR [4]. In addition, its application to the elderly has been questioned as few old people were included during the development [5]. In 2016, the full age spectrum (FAS) equation was published by Pottel et al. [6, 7], which focuses on the coherence among different age groups. But it was reported to overestimate GFR in people with low creatinine or poor renal function [8]. Thus, to achieve one more applicable formula, the European Kidney Function Consortium (EKFC) has published a novel equation EKFC in 2021. It was developed in a European population (all nonblack) by combining the respective advantages of the above two equations, addressing their deficiency to some extent [9]. Delanaye et al. [10] reported EKFC performed well in the whole age range by verifying it in a large cohort from Africa and Europe. However, only a few studies have been reported on the Asian population. In China, Xia et al. [11] assessed its applicability in hospitalized old patients rather than all age groups, and the sample size was relatively small. Therefore, this study was conducted to evaluate the potential performance of the new EKFC equation in Chinese adults and draw a comparison with its predecessors, including CKD-EPI₂₀₀₉ and FAS.

A total of 3,692 Chinese adults (≥18 years) were consecutively enrolled in the study at the First Affiliated Hospital of Nanjing Medical University from April 2007 to September 2021. The Ethics Committee of Nanjing Medical University approved the study protocol, and all participants gave informed consent (registration number 2021-SR-508).

Exclusion criteria included individuals less than 18 years, non-Chinese, lacking diagnostic content, acute kidney injury, severe hydronephrosis, acute infection, pleural or abdominal effusion, serious edema or malnutrition, skeletal muscle atrophy, amputation, ketoacidosis, and renal transplantation. Besides, those who had recently received glucocorticoid and hemodialysis therapy were also excluded. Demographic and past medical records were retrieved from the electronic data capture. All diagnoses were in accordance with the international classification of diseases (ICD-10). Detailed inclusion and exclusion process was shown in online supplementary Figure S1 (for all online suppl. material, see https://doi.org/10.1159/000531030).

Venous blood samples were drawn in the fasting state between 8:00 and 10:30 a.m. SCr concentration was measured by the enzymatic method (sarcosine oxidase-PAP; Shanghai Kehua Bio-Engineering Co., Ltd, China), traceable to the National Institute of Standards and Technology (isotope-dilution mass spectrometry calibrated) and creatinine standard reference material (SRM909b). The intra- and inter-assay coefficients of variations for creatinine measurement were consistently ≤6%, with stable quality control. All enzyme activities were measured at 37°C.

All participants underwent the 99mTc-DTPA renal dynamic imaging measurement. After measuring height and weight, subjects were demanded to drink 300–500 mL water and empty the bladder. A bolus of 185 MBq 99mTc-DTPA (purity 95–99%; Nanjing Senke Co., Ltd, China) was injected intravenously into the elbow vein, and the renal dynamic imaging measurement was carried out immediately. After image acquisition and processing, reference GFR (rGFR) was automatically calculated by a computer according to the Gates method.

To verify its accuracy, we further compared the 99mTc-DTPA renal dynamic imaging GFR with the dual plasma sample 99mTc-DTPA clearance GFR. If their difference was significant, a regression equation would be established and used for the correction. According to the method used by Ma et al. [12, 13], we determined the minimum required sample size to be 36 based on 95% confidence interval (CI) and 80% power. Then, we randomly selected 50 participants receiving renal dynamic imaging in the morning (GFR: 18.1–116.5 mL/min/1.73 m²) and arranged for them to perform dual plasma sample clearance later. Specific contents: drug radioactivity count was measured and converted to D before injection. After image acquisition, blood samples were collected, respectively, from the contralateral forearm at 120 min (t1) and 240 min (t2) postinjection. We collected 1 mL after separating the plasma and measured the radioactivity (P1 and P2) in a multifunction well counter (ZD-6000 multifunction instrument; Zhida Technology Co., Xian, People’s Republic of China). Then, we plugged values above into the following formula: GFR (mL/min/1.73 m²) = (D ln[P1/P2]/[t1-t2])exp([t1ln P2-t2ln P1])/(t1-t2)×1.73/BSA [14].

Paired sample t test showed that GFR measured by above two approaches was not significantly different (p > 0.05). The intraclass correlation coefficient (ICC) > 0.75. Thus, the renal dynamic imaging measurement results were adopted as the rGFR.

The CKD-EPI, FAS, and EKFC equations were selected to calculate eGFR separately. Both FAS and EKFC are based on the standardized SCr value calculated by SCr/Q, where Q represents the sex- and age-specific median creatinine value in healthy individuals. In Table 1, detailed expressions of above equations and values of Q were listed.

Normally distributed variables (p > 0.05, Kolmogorov-Smirnov test) were expressed as mean ± standard deviation. Otherwise, they were presented as median with 25th and 75th percentiles. The count data were presented as percentages. Correlations between eGFR and rGFR were measured by Spearman correlation analysis. The ability to detect rGFR<60 mL/min/1.73 m² was assessed using sensitivity, specificity, and the area under the receiver operating characteristic curve (ROC^AUC). Bland-Altman analysis was performed to assess the concordance between eGFR and rGFR.

According to the recommendations of KDOQI guidelines, we also evaluated the performance of these equations from bias (median difference between eGFR and rGFR), precision (interquartile range [IQR] of the median difference), and accuracy (percentage of estimates within 30% of the measured values [p30] and root-mean-square error [RMSE]). 95% CIs for bias, IQR, and RMSE were calculated using bootstrap methods with 3,000 replications. Wilcoxon signed-rank test was performed to compare bias. McNemar test was used to compare P30 values. Above analyses were repeated in subgroups stratified by age, sex, and rGFR stage. When grouping participants by age, we used 40 and 60 years as nodes, respectively. We also subdivided SCr values by quartile (Q1-Q4) for further investigation at different levels of SCr.

A two-tailed p value <0.05 was considered statistically significant. Data were analyzed with SPSS (version 26.0 SPSS, IBM Corp), R 3.6.3, and MedCalc (version 11.6.1.0; MedCalc Software). The graphic drawing was implemented in Prism 8 (GraphPad Software, LLC) and MATLAB (version 2016b, The MathWorks).

3,692 participants were eventually enrolled in this study, 58.50% of which were males. Median age and SCr were 54 (43, 64) years and 0.91 (0.72, 1.19) mg/dL, respectively. Mean rGFR was 74.2 ± 24.2 mL/min/1.73 m². Detailed anthropometric results and biochemical characteristics are shown in Table 2.

All equations exhibited a significantly positive relationship with rGFR through the calculation of Spearman correlation coefficient. Comparing their correlations, no significant difference was found, and EKFC had a relatively stronger correlation (R = 0.749, 95% CI: 0.733, 0.764). EKFC displayed a larger ROC^AUC (0.902, 95% CI: 0.892, 0.911), higher sensitivity (84.59), but lower specificity (80.02) than CKD-EPI (ROC^AUC = 0.899, 95% CI: 0.889, 0.912, sensitivity = 80.00 and specificity = 83.87) (Table 3). By testing consistency, EKFC had the smallest mean difference (3.9 mL/min/1.73 m²). Scatter plots and Bland-Altman plots of these equations versus rGFR are shown in Figure 1. Performance of these equations is summarized in Table 4. Among all formulas, EKFC exhibited the best accuracy, with the highest P30 (73.3%, 95% CI: 70.6, 76.1) and the lowest RMSE (18.96, 95% CI: 18.50, 19.41). FAS achieved the same P30 as CKD-EPI, both 67.5% (95% CI: 64.9, 70.2). All three equations overestimated GFR. Bias and IQR of EKFC were the lowest (bias = 3.61, 95% CI: 2.84, 4.42, and IQR = 24.09, 95% CI: 23.15, 25.02).

When comparing the diagnostic consistency of GFR staging between the eGFR and rGFR, similar results were obtained (Fig. 2). EKFC was relatively more accurate for GFR staging when 60≤rGFR<90 and rGFR<30 mL/min/1.73 m², but tended to classify it to a higher stage.

The EKFC equation achieved a significantly higher P30 (73.0%, 95% CI: 69.6, 76.5), compared to CKD-EPI (66.9%, 95% CI: 63.7, 70.2) and FAS (64.7%, 95% CI: 61.5, 68.0) in the subgroup of age<60 years. RMSE and IQR of EKFC were found to be the lowest (RMSE = 20.28, 95% CI: 19.69, 20.86, and IQR = 25.45, 95% CI: 24.43, 26.51). All equations overestimated GFR in this subgroup. FAS had the highest bias (11.61, 95% CI: 10.69, 12.86), and EKFC had the lowest (6.98, 95% CI: 6.10, 8.14) conversely (Table 4).

In the crowd aged 60 years and older, EKFC still depicted a higher P30 (74.0%, 95% CI: 69.3, 78.8) than FAS (72.7%, 95% CI: 68.1, 77.5) and CKD-EPI (68.7%, 95% CI: 64.2, 73.4), but not significant. EKFC and FAS underestimated GFR, and CKD-EPI overestimated it significantly (Table 4).

In addition, we screened out participants younger than 40 years, which was an important age knot in these formulas. EKFC performed significantly better, with the greatest accuracy (P30 = 73.5%, 95% CI: 67.4, 80.0, and RMSE = 21.70, 95% CI: 20.65, 22.75) and the lowest IQR (27.05, 95% CI: 24.81, 29.28) and bias (9.76, 95% CI: 7.55, 11.47) (online suppl. Table S1).

All equations showed better diagnostic performance in the male, with lower bias and higher accuracy. Among them, the EKFC equation had the best applicability in subgroups, with the lowest bias, IQR, and RMSE, and relatively higher P30 (Table 4).

In subgroups based on rGFR staging, P30 of all equations decreased with decreasing rGFR. In the individuals with rGFR≥90 mL/min/1.73 m², the EKFC equation had relatively lower IQR (21.46, 95% CI: 19.73, 23.24) and RMSE (16.83, 95% CI: 15.88, 17.84), and P30 of it (94.8%, 95% CI: 88.7, 101.2) was higher, followed by CKD-EPI (89.9%, 95% CI: 84.0, 96.1). EKFC systematically underestimated GFR, which contrasted with the findings of the other two equations in this subgroup. P30 of EKFC remained the highest (72.2%, 95% CI: 68.2, 76.4) in the subjects with 60≤rGFR<90 mL/min/1.73 m², followed by FAS (66.4%, 95% CI: 62.5, 70.4) and CKD-EPI (64.6%, 95% CI: 60.8, 68.5). All equations overestimated GFR in this subgroup, and CKD-EPI was more pronounced (bias: 13.47, 95% CI: 12.36, 14.35). When rGFR<60 mL/min/1.73 m², the P30 values were all less than 60%. There was no significant difference in performance of the three formulas (Table 4).

In Q1 (SCr: 0.23–0.72 mg/dL), the EKFC equation achieved relatively lower IQR (24.29, 95% CI: 22.43, 26.23) and RMSE (21.98, 95% CI: 21.07, 22.91), and the highest P30 (72.6%, 95% CI: 67.2, 78.3), followed by CKD-EPI (59.9%, 95% CI: 55.0, 65.1). FAS was relatively worse, and its P30 was even less than 50%. All equations overestimated GFR, and bias of FAS was the highest (27.23, 95% CI: 25.14, 28.58). In Q2 (SCr: 0.72–0.91 mg/dL), EKFC still performed better. FAS was found to make significant progress at P30 and bias, compared with the Q1 group. In Q3 and Q4, there was no statistical difference in P30 for each equation. Considering bias, all equations in Q3 (SCr: 0.91–1.19 mg/dL) overestimated GFR but underestimated it in Q4 (SCr: 1.19–18.99 mg/dL) (Table 4). The relationships between SCr and bias were shown through scatter plots (online suppl. Fig. S2).

In this study, we identified the EKFC equation in Chinese adult population. We compared three SCr-based equations in 3,692 participants and observed that EKFC performed relatively better in terms of diagnostic value, bias, precision, and accuracy. In each subgroup, EKFC never performed poorly compared to the other two, and it was especially better in participants with normal or mildly impaired renal function (rGFR≥60 mL/min/1.73 m²), and low creatinine.

EKFC was developed based on available formulas. It complemented the advantages of CKD-EPI and FAS. So, we started our analysis with the specific contents of three equations. For the convenience of development, equations were established on a basic statistical model [7], which can fundamentally establish that both FAS and EKFC differed from CKD-EPI. CKD-EPI was developed on the model that GFR is in a state of decline with aging since 18 years [5], whereas FAS and EKFC considered the age of 40 as the turning point, before which there was no age-dependent decline of renal function [6, 9]. Current findings based on healthy participants suggested that GFR could potentially remain stable or decline slowly until the age of 40–50 due to sufficient renal reserve [7, 15], and previous studies in healthy Chinese also sketched this trajectory of change [16, 17]. The model of FAS and EKFC may be closer to the reality of the Chinese. To balance the different age decline rates of the 18–40 years in comparison with the 40-year age range, CKD-EPI adopted one higher age term (0.993^age). But it can lead to a greater overestimation [7]. Comparison of these formulas in participants younger than 40 years may further identify the mathematical model of FAS, and EKFC was reasonable in Chinese. Thus, in the development of formulas, paying attention to the differences between age groups may be one key point to improve their applicability.

Overall, by applying a model considering age-related changes in GFR, EKFC effectively reduced the deviation. However, FAS with a similar model did not work as well as it did. When SCr was low, the GFR was infinitely overestimated in some individuals, which could be apparent from the function graphs we constructed (Fig. 3). This might be related to the set of the normalized node Q. The smaller the SCr is than Q, the greater deviation would be obtained. Interestingly, CKD-EPI does not depict this problem. It set node for SCr and applied different coefficients, mitigating the overestimation of eGFR to some extent. And EKFC drew on this, limiting overestimation at low creatinine. From our results, the method of targeted combining advantages of existing formulas can indeed address the shortfall to some extent, which could be used for reference when developing new formulations.

Notably, our results indicated that EKFC did not significantly perform better in participants with high SCr and poor renal function, in comparison with the other two equations. One probable explanation was that older adults accounted for the majority of these subgroups. For the middle-aged and the elderly with high creatinine, content of EKFC did not noticeably change from FAS. Second, the stringency of P30 can vary between high and low GFR. For instance, when rGFR = 100 mL/min/1.73 m², the eGFR within ±30% of rGFR is 70–130 mL/min/1.73 m². However, if rGFR = 10 mL/min/1.73 m², this range would be significantly narrowed to 7–13 mL/min/1.73 m². It is difficult for the estimated values to fall within such a small range, irrespective of which type of formula was used. Therefore, the diagnosis consistency in GFR staging and bias deserve more attention in individuals with severely decreased eGFR, because the actual kidney function might be worse if the formulas used were to overestimate GFR. At this time, serum cystatin C might be a good choice for further assessment, which could make up for some deficiencies of creatinine [18, 19], although it costs more. Serum cystatin C-based estimations are much superior in predicting morbidity and mortality [20]. And combined creatinine-cystatin C equations are considered to perform better than single-indicator formulas [21]. Currently, new formulas have been developed, including EKFC_Cys and EKFC_Cr-Cys, which alleviate differences between blacks and whites, men and women [22]. They outperform previous formulas in validation. Yet their performance in Asians is unclear, more research is needed.

In another validation based on the Chinese elderly population over 65 years, EKFC performed even slightly worse than FAS [11]. It may be due to the small proportion of participants whose rGFR≥60 mL/min/1.73 m², which was the advantage of EKFC. Our results were consistent with that reported by Pottel et al. [9], showing the superiority of the new equation. However, in their study, the P30 values of EKFC were 82.1%, 85.9%, and 83.6% in young, middle-aged, and older participants, respectively, which were significantly higher than our results. The first possible reason was that different exogenous markers were used in the measurement. Some studies have shown variations in the rGFR obtained by the distinct pathways [23, 24]. Cohorts using nonradioactive markers such as inulin are needed in China to solve this question. Another reason may be attributed to racial difference, which has been regarded as an important consideration affecting the performance and applications of SCr-based eGFR formulas [25]. So, it is important that sex- and age-specific median creatinine values (Q values) should be set appropriately with race. A study from South Korea suggested that Korean males had higher Q value [26]. Yet rare studies have been conducted to achieve appropriate Q values for the Chinese. We still adopted the Q values of healthy white people when comparing. To confirm the influence of racial factor, we developed Chinese Q values based on the SCr of 27,830 healthy people (age: 18–88 years). Details were presented in the supplement (online suppl. Section 4). Results showed that equations with the new Q values had better accuracy and lower bias (online suppl. Table S5), proving the influence of race on equations. There are now growing calls to eliminate the racial coefficient. A recent meta-analysis also showed that the equation without race coefficient was favorable for black adults [27]. However, at least for Chinese individuals in our study, racial factor still needs to be considered when using SCr-based equations.

Advantages of this study included a relatively large sample size of participants, ranging from 18 to 99 years. And we clarified highlights worthy of emulation by analyzing the structure of equations. Furthermore, we preliminarily deduced the Chinese Q values and verified the influence of ethnic factors on equations. Nevertheless, several limitations in our study deserved mention. First, all participants belonged to a single center, which might result in selection bias. Second, all subjects had only one measurement of creatinine, which might also influence the results.

The EKFC equation was developed by combining the advantages of CKD-EPI and FAS, achieving better diagnostic ability and performance in estimating GFR. However, ethnic differences limited the accuracy of the equation. Developing equations that fit the Chinese based on correct theory would be more accurate. Before then, EKFC could yet be regarded as a good choice, especially in health care.

This study was approved by the Ethics Committee of the First Affiliated Hospital of Nanjing Medical University and conducted in accordance with the Declaration of Helsinki, registration number 2021-SR-508. All participants gave their informed written consent.

The authors declared no conflict of interest.

This study was supported by the grants from the National Key R&D Program of China (2018YFC2002100, 2018YFC2002102), National Natural Science Foundation of China (82171585), Jiangsu Province Older Adults Health Introduction New Technique Project (LX2021003), Jiangsu Province Hospital Clinical Ability Improvement Project (JSPH-MC-2022-22), Jiangsu Province Graduate Training Innovation Project (SJCX21_0620), Jiangsu Province Cadres Health Care Project (BJ17018), and Jiangsu Province Hospital 511 Project (JSPH-511A-2018-5).

Design: Yao Ma, Lu Wei, and Zhenzhu Yong. Data collection and statistical analyses: Yao Ma and Zhenzhu Yong. Writing and revision: Yao Ma, Lu Wei, Yue Yu, Bei Zhu, and Weihong Zhao.

Yao Ma, Lu Wei and Zhenzhu Yong contributed equally to this work and are co-first authors.Bei Zhu and Weihong Zhao share senior authorship.

The data underlying this article will be shared on reasonable request to the corresponding author.