Predictive performance of the new race-free Chronic Kidney Disease Epidemiology Collaboration equations for kidney outcome in Korean patients with chronic kidney disease

Study population

This study was conducted using data from the Korean Cohort Study for Outcome in Patients with Chronic Kidney Disease (KNOW-CKD). The KNOW-CKD was a nationwide, multicenter, prospective cohort study. The detailed design and methods of the KNOW-CKD have been previously described (NCT01630486 at http://www.clinicaltrials.gov) [12]. Patients aged 20 to 75 years with CKD of various causes were recruited from 2011 to 2016. As a result, 2,238 patients from CKD G1 to G5ND were included in the KNOW-CKD. After excluding 89 patients without measurement of serum cystatin C levels, a total of 2,149 patients were included. CKD severity was defined according to the Kidney Disease: Improving Global Outcomes guidelines [13]. This study was performed in accordance with the Declaration of Helsinki, and the study protocol was approved by the Institutional Review Board of the participating centers; Seoul National University Hospital (1104-089-359), Seoul National University Bundang Hospital (B-1106/129-008), Severance Hospital (4-2011-0163), Kangbuk Samsung Medical Center (2011-01-076), The Catholic University of Korea, Seoul St. Mary’s Hospital (KC11OIMI0441), Gachon University Gil Hospital (GIRBA2553), Eulji Medical Center (201105-01), Chonnam National University Hospital (CNUH-2011-092), and Inje University Pusan Paik Hospital (11-091).

Data collection

Baseline socioeconomic data, anthropometric measurements, and medical histories were obtained by trained healthcare providers using a standardized protocol. These data were extracted from the electronic data management system of KNOW-CKD. A history of CVD was defined as a composite of history of coronary artery disease, ischemic stroke, hemorrhagic stroke, or peripheral vascular disease. All blood samples were obtained after overnight fasting and centrifuged within 1 hour. The following laboratory parameters were measured: complete blood count, serum calcium, phosphate, albumin, total cholesterol, high-sensitive C-reactive protein (hsCRP), and parathyroid hormone levels. Proteinuria was determined using spot urine protein-to-creatinine ratio (PCR).

Calculation of estimated glomerular filtration rate

Serum creatinine and cystatin C levels were measured at the central laboratory of KNOW-CKD (Lab Genomics). Serum creatinine levels were measured using the Jaffe method (ADVIA Chemistry XPT System; Siemens) and a coefficient of variation (CV) of creatinine was 5.1%. Serum cystatin C levels were measured using a particle-enhanced nephelometric immunoassay (BN II System; Siemens) and CV of cystatin C was 4.3%. Serum creatinine level was standardized using the isotope-dilution mass spectrometry-traceable method. eGFR was calculated based on serum creatinine and both markers (creatinine and cystatin C). We used the current CKD-EPI equations based on creatinine (eGFR_cr) and both markers (eGFR_cr-cys) without the Black race coefficient [10], as well as eGFR based on creatinine [eGFR_cr (NEW)] and both markers [eGFR_cr-cys (NEW)] with the newly developed race-free CKD-EPI equations [11] as follows:

eGFR_cr (NEW) = 142 × min (creatinine/κ, 1)^α × max (creatinine/κ, 1)^–1.2 × 0.9938^age × 1.012 (if female) [α = –0.241 (female), –0.302 (male); κ = 0.7 (female), 0.9 (male)]

eGFR_cr-cys (NEW) = 135 × min (creatinine/κ, 1)^α × max (creatinine/κ, 1)^–0.544 × min (cystatin C/0.8, 1)^–0.323 × max (cystatin C/0.8, 1)^–0.778 × 0.9961^age × 0.963 (if female) [α = –0.219 (female), –0.144 (male); κ = 0.7 (female), 0.9 (male)].

Outcome assessment

The primary outcome of this study was the onset of KFRT, defined as maintenance of long-term kidney replacement therapy (KRT; hemodialysis or peritoneal dialysis) for over three months or kidney transplantation during the follow-up period.

Statistical analysis

Continuous variables are expressed as mean ± standard deviation and were compared using a one-way analysis of variance. Categorical variables are expressed as numbers and percentages and were compared using the chi-square test. Data with skewed deviations are presented as median and interquartile ranges (IQRs) and compared using the Jonckheere-Terpstra test. The linear-by-linear association among CKD grades was evaluated and presented as a p for trend. Agreement between eGFRs based on the current and new equations was visualized using a Bland-Altman plot [14]. In addition, we assessed the predictive performance of eGFRs for the 5-year risk of KFRT. We compared the prediction performances of the eGFRs based on the current and new equations using the time-dependent area under the receiver operating characteristic curve (AUC). Moreover, reclassification and discrimination improvement by the new eGFRs compared to the current eGFRs were assessed using the continuous net reclassification index (NRI) and integrated discrimination improvement (IDI) statistics. To compare predictive performances by AUC, NRI, and IDI, we adjusted for the following variables: age, sex, history of CVD, diabetes mellitus, body mass index, systolic blood pressure, serum total cholesterol, calcium-phosphate product, urine PCR, use of renin-angiotensin system (RAS) blockade, and statin use. Furthermore, we compared the predictive performance of each eGFR for 2-year and 8-year KFRT risk, and performed subgroup analysis based on age (<60 and ≥60 years), sex, and cause of CKD. Finally, we calculated the four-variable Kidney Failure Risk Equation (KFRE) score [15], which included age, sex, eGFR, and urine albumin-to-creatinine ratio, and compared predictability of eGFR equations using this score. All analyses were conducted using the R language (version 3.6.2; R Foundation for Statistical Computing).

Baseline characteristics

The baseline characteristics of the patients according to CKD severity based on eGFR_cr (NEW) are presented in Table 1. The mean age was 53.8 years, and 61.2% were men. Among 2,149 patients, 33.9%, 95.9%, and 13.4% had type 2 diabetes mellitus, hypertension, and a prior history of CVD, respectively. The Charlson comorbidity index was higher in higher CKD grades (p for trend < 0.001). The most common cause of CKD was glomerulonephritis (35.7%), followed by diabetic nephropathy (23.3%), hypertensive nephropathy (18.5%), and polycystic kidney disease (16.4%). As expected, hemoglobin and serum calcium levels gradually decreased in advanced CKD grades (p for trend < 0.001, both). In addition, there were graded increases in serum phosphate, hsCRP, and parathyroid hormone levels in more severe CKD (p for trend < 0.001, all). Median urine PCR was 0.5 g/g (IQR, 0.1–1.5 g/g), and 85.5% of patients used RAS blockade at baseline.

Table 1.

Baseline characteristics of patients according to CKD stages based on new creatinine-based eGFR equation

Differences between current and new estimated glomerular filtration rates and reclassification of chronic kidney disease grades

The mean eGFR_cr and eGFR_cr (NEW) of the whole cohort were 53.2 ± 30.8 and 56.0 ± 31.6 mL/min/1.73 m², respectively. Thus, the difference between the two eGFRs was 2.90 mL/min/1.73 m² (95% confidence interval [CI], 2.82–2.94 mL/min/1.73 m²). The difference between the current and new eGFRs was more likely to be larger in early CKD (Fig. 1A). Based on CKD severity with the current eGFR_cr, the difference was largest in CKD G2 (4.2 ± 1.0 mL/min/1.73 m²) and decreased in CKD G3 or greater (p for trend < 0.001) (Table 2). Of the 2,149 patients, 303 (14.1%) were reclassified as having lower CKD grades. In patients with CKD G3a based on eGFR_cr, 81 (23.1%) were reclassified as G2 based on eGFR_cr (NEW). Therefore, the number of patients with eGFR of <60 mL/min/1.73 m² decreased from 1,393 (64.8%) to 1,312 (61.1%) in the entire cohort (Supplementary Fig. 1, available online).

Figure 1.

Bland-Altman plot for agreement between the current and new eGFR equations.

(A) eGFR based on creatinine. (B) eGFR based on creatinine and cystatin C. The average of the two eGFRs (x-axis) is plotted against the difference of the two eGFRs (y-axis). Difference was calculated as new eGFR – current eGFR. Solid horizontal lines represent means and dashed lines represent standard deviation of eGFR differences.

eGFR, estimated glomerular filtration rate.

Table 2.

Reclassification of CKD stages based on new creatinine-based eGFR equation

With regard to eGFR with both markers, the mean eGFR_cr-cys and eGFR_cr-cys (NEW) were 52.3 ± 32.1 and 54.6 ± 33.1 mL/min/1.73 m², respectively. The difference between the two eGFRs was 2.3 mL/min/1.73 m² (95% CI, 2.20–2.32 mL/min/1.73 m²). There was a similar pattern in the difference between the current and new eGFR_cr-cys across CKD severity as observed in analysis with eGFR_cr. The difference was greater in CKD G1 and G2 (Fig. 1B). Among the 326 patients classified as CKD G3a based on eGFR_cr-cys, 53 (16.3%) were reclassified as G2 based on eGFR_cr-cys (NEW), and one patient originally classified as CKD G2 was regrouped to G3 (Table 3). Therefore, the number of patients with eGFR of <60 mL/min/1.73 m² decreased from 1,413 (65.8%) to 1,361 (61.1%) (Supplementary Fig. 1, available online).

Table 3.

Reclassification of CKD stages based on new creatinine and cystatin C-based eGFR equation

In subgroup analysis, the difference in eGFR_cr between the current and new equations was significantly higher in men than in women across all CKD grades (Supplementary Table 1, Supplementary Fig. 2; available online). Accordingly, 200 men (15.2%) were reclassified into lower CKD grades, and 103 women (12.4%) were reclassified (Supplementary Table 2, available online). In contrast, the difference in eGFR_cr-cys between the two equations was slightly higher in women than in men; 138 men (10.5%) and 82 women (9.8%) were reclassified (Supplementary Table 3, available online). In subgroup analysis by age, eGFR differences between the current and new equations were significantly higher in patients aged ≥60 years than those aged <60 across all CKD grades (Supplementary Table 4, Supplementary Fig. 3; available online). Therefore, patients aged ≥60 years were reclassified into lower CKD grades than those aged <60 years, regardless of eGFR equation (Supplementary Table 5, 6; available online).

Predictive performance of estimated glomerular filtration rate equations for kidney failure with replacement therapy

During a median of 5.6 years of follow-up, 624 KFRT events (29.0%) occurred. The AUC of eGFR_cr (NEW) for the 5-year KFRT risk was 0.896 (95% CI, 0.877–0.915), which was slightly higher than that of eGFR_cr (0.895; 95% CI, 0.876–0.915) (Table 4; Fig. 2A, B). The AUCs were similar between eGFR_cr and eGFR_cr (NEW) after adjusting for covariables. The IDI and continuous NRI showed that eGFR_cr (NEW) had slightly better discrimination and reclassification abilities than eGFR_cr. Without adjusting for covariables, eGFR_cr-cys (NEW) showed slightly superior predictive performance compared to eGFR_cr-cys (Table 4; Fig. 2C, D). However, in the adjusted model, all AUC, IDI, and continuous NRI analyses indicated that eGFR_cr-cys and eGFR_cr-cys (NEW) had similar predictive performance. We further analyzed the 2-year and 8-year risks of KFRT. The results showed that the predictability of KFRT by AUC was similar between the current and new eGFR equations across all adjusted models (Supplementary Table 7, available online). In the adjusted models of IDI and continuous NRI analyses, eGFR_cr (NEW) outperformed eGFR_cr for 2-year KFRT risk but not for 8-year risk (Supplementary Table 8, available online). Meanwhile, eGFR_cr-cys and eGFR_cr-cys (NEW) showed similar performance across all time-point KFRT risk in the adjusted models. When we calculated four-variable KFRE score, all scores showed higher AUCs than eGFR alone in all analyses (Supplementary Fig. 4, 5; available online). However, we found no significant differences in KFRE scores between current and new eGFR equations for both 2-year and 5-year KFRT risk.

Table 4.

Comparison of predictive performance of eGFR equations for 5-year KFRT risk

Figure 2.

Time-dependent receiver operating characteristic curves of eGFR equations for kidney failure with replacement therapy.

(A) Current creatinine-based equation, (B) new creatinine-based equation, (C) current creatinine and cystatin C-based equation, (D) new creatinine and cystatin C-based equation. Each curve was derived based on Cox proportional hazard regression model. In the multivariable model, the following covariables were adjusted for: age, sex, history of cardiovascular disease, diabetes mellitus, body mass index, systolic blood pressure, serum total cholesterol, calcium-phosphate product, urine protein-to-creatinine ratio, and use of renin-angiotensin system blockade or statin. Values in the index are presented as area under curve (95% CI).

eGFR, estimated glomerular filtration rate; eGFR_cr, eGFR based on creatinine with current equation; eGFR_cr (NEW), eGFR based on creatinine with new equation; eGFR_cr-cys, eGFR based on creatinine and cystatin C with current equation; eGFR_cr-cys (NEW), eGFR based on creatinine and cystatin C with new equation.

In subgroup analysis, IDI and continuous NRI statistics showed that eGFR_cr (NEW) had slightly better performance than eGFR_cr in both sexes, but AUCs were similar between eGFRs (Supplementary Table 9, available online). eGFR_cr-cys and eGFR_cr-cys (NEW) performed similarly in both sexes. In addition, eGFR_cr (NEW) showed superior predictive performance in all analyses compared with eGFR_cr in patients aged <60 years, whereas these were similar in those aged ≥60 years (Supplementary Table 10, available online). The predictive performance of eGFR_cr-cys and eGFR_cr-cys (NEW) was similar in patients aged <60 years. However, IDI and continuous NRI showed inferior discrimination and reclassification ability for eGFR_cr-cys (NEW) compared to eGFR_cr-cys. We further compared the AUCs of eGFR equations by CKD cause. We found that all eGFRs showed excellent performance for predicting 5-year KFRT risk in all subgroups (Supplementary Fig. 6, available online). However, there was no significant difference in AUCs between the current and new equations.

Finally, we compared the predictive performance of the new eGFR equations. There was no difference in the AUC between eGFR_cr (NEW) and eGFR_cr-cys (NEW) across all time points of KFRT risk (Supplementary Table 11, available online). Moreover, IDI and continuous NRI also indicated that the predictive performance was similar between eGFR_cr (NEW) and eGFR_cr-cys (NEW) (Supplementary Table 12, available online).