Cystatin C-based estimated GFR performs best in identifying individuals with poorer survival in an unselected Chinese population: results from the China Health and Retirement Longitudinal Study (CHARLS).

Background: The decline in estimated glomerular filtration rate (eGFR) has been reported as a risk factor for mortality. However, it remains unclear which eGFR equation is most useful in predicting death in the general Chinese population.
Methods: The association was examined between eGFR and all-cause mortality using data from the China Health and Retirement Longitudinal Study. Participants with complete data in 2011 and survival follow-up in 2013, 2015 and 2018 were included and analyzed in three separate cohorts, which included 8160, 8154 and 8020 participants, respectively. Logistic regression analyses, receiver operating characteristic curve, continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were computed to compare the discriminative power of eGFR derived by abbreviated Modification of Diet in Renal Disease (MDRD), Chinese coefficient-modified MDRD (MDRD-CN), Japanese coefficient-modified MDRD (MDRD-JPN), CKD-EPIcr, Japanese coefficient-modified CKD-EPIcr (CKD-EPIcr-JPN), CKD-EPIcys, CKD-EPIcr-cys, CKD-EPIcr fit without race and CKD-EPIcr-cys fit without race.
Results: A decreased eGFR (<60 ml/min/1.73 m2) was significantly associated with increased mortality at 2 years no matter which eGFR equation was used (odds ratio ranged between 2.02 and 4.94, all P < 0.001). The association remained significant after adjusting multiple covariates when MDRD-CN, CKD-EPIcys or CKD-EPIcr-cys fit without race was used. CKD-EPIcys showed the highest discriminative power for mortality (area under the curve 0.744 ± 0.40) and outperformed other equations (all P < 0.001) except for CKD-EPIcr-cys. The overall risk classification was also improved when the CKD-EPIcys equation was adopted as indicated by continuous NRI and IDI. Similar results were observed at 4 and 7 years. Conclusions: A decline in eGFR by all equations could predict poorer survival, among which the CKD-EPIcys equation showed the best discriminative power.

INTRODUCTION

Glomerular filtration rate (GFR) is the most well-established marker for assessing kidney function. However, it is complicated to measure GFR accurately by inulin or by radioisotope renography. Under such circumstances, the estimated GFR (eGFR) is a proper substitute and has been widely used in detecting and managing chronic kidney disease (CKD) [1]. Several popular equations were proposed based on different demographic and blood biomarkers to calculate eGFR, including serum creatinine–based Modification of Diet in Renal Disease (MDRD) equation, Chronic Kidney Disease Epidemiology Collaboration creatinine (CKD-EPIcr) equation, CKD-EPI cystatin C (CKD-EPIcys) equation, CKD-EPI creatinine-cystatin C (CKD-EPIcr-cys) equation and CKD-EPI equations fit without race [2-6].

An increasing number of studies have shown that decreased eGFR was an independent risk factor for mortality in the general population and selected populations with diabetes, hypertension and cardiovascular diseases (CVDs) [7-12]. It was also supported by the findings in some studies that eGFR estimated by the cystatin C–based equation showed a more robust association with mortality compared with that estimated based on creatinine [13-19]. However, the studies were either not carried out in the Chinese population or targeted patients under specific disease conditions, such as acute heart failure or after percutaneous coronary intervention (PCI). It remains unclear which equation is most useful for predicting death in the general Chinese population. To clarify this issue, this study aims at investigating the association of eGFR derived by different equations with all-cause mortality from the China Health and Retirement Longitudinal Study (CHARLS).

MATERIALS AND METHODS

Population

All data were collected from the CHARLS, which is a nationally representative longitudinal survey of middle-aged and elderly individuals in China carried out by Peking University. The study was initiated in 2011 and followed up in 2013, 2015 and 2018. Participants were chosen in 150 randomly selected counties all over the country with a probability-proportional-to-size method. More details were described in former studies [20]. The study was approved by the Institutional Review Board of Peking University Health Science Center. Written informed consent was obtained from each subject before participation.

Grouping

A total of 17 708 participants took part in the baseline survey (in 2011), among which 8849 participants had complete data. Participants with complete data were reinvestigated in 2013 (Cohort 1, n = 8160), in 2015 (Cohort 2, n = 8154) and in 2018 (Cohort 3, n = 8020). Survival information was obtained through a questionnaire survey by family members. It should be noted that cumulative death information was gathered for each cohort. For example, patients who died during the 2013 follow-up would be included in the analyses for Cohort 2 and Cohort 3. Participants were analyzed in three separate cohorts in order to make the most of the available data, as some of the participants missed one or two follow-ups (Figure 1).

FIGURE 1:

Enrollment flowchart of the study.

Blood sample collection and analysis

The venous blood samples were collected by trained staff of the Chinese Center for Disease Control and Prevention (CDC). The samples were centrifugated before they were stored at −20°C at the local laboratory. After that, the samples were transported to the Chinese CDC in Beijing within 2 weeks. They were stored at −80°C until tests were performed. Serum creatinine was tested by the rate-blanked and compensated Jaffe creatinine method. Cystatin C was tested by particle-enhanced turbimetric assay. Blood urea nitrogen (BUN) was measured by the enzymatic ultraviolet (UV) method with urease. Uric acid (UA) was assayed by the UA plus method. Hemoglobin A1c (HbA1c) was assayed by boronate affinity high-performance liquid chromatography. Fasting blood glucose (FBG), total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) were assayed by the enzymatic colorimetric test. High-sensitivity C-reactive protein (hs-CRP) was tested by immunoturbidimetric assay.

Different eGFR equations

A total of nine equations were used to estimate GFR, including the four-variable abbreviated MDRD equation, Chinese coefficient–modified MDRD (MDRD-CN) equation, Japanese coefficient–modified MDRD (MDRD-JPN) equation, CKD-EPIcr equation, Japanese coefficient–modified CKD-EPIcr (CKD-EPIcr-JPN) equation, CKD-EPIcys equation, CKD-EPIcr-cys equation, CKD-EPIcr equation fit without race and CKD-EPIcr-cys equation fit without race [3, 5, 6, 21–23]. The exact formulas are listed in the Supplementary data, Appendix Table S1.

Assessment of covariates

Demographic variables (gender and age), anthropometric measurements (blood pressure), health behavior (smoking and drinking) and medical history (hypertension, diabetes, dyslipidemia, heart disease, stroke and tumor) were collected by trained staff. Smoking was classified as never, used to and current smoker, while drinking was classified as never, less than once per month and more than once per month. Hypertension was defined as an average blood pressure of three measurements ≥140/90 mmHg and/or a self-reported hypertension history. Diabetes was defined as meeting one or more of the following criteria: FBG ≥126 mg/dl, HbA1c ≥6.5% or a self-reported history. Dyslipidemia was diagnosed if one or more of the following abnormalities were reported: TC ≥200 mg/dl, TG ≥150 mg/dl, LDL-C ≥130 mg/dl, HDL-C <40 mg/dl  or a self-reported history. Heart disease, stroke and tumor were defined as self-reported history.

Statistical analysis

Baseline characteristics were presented as mean [standard deviation (SD)] or median [interquartile range (IQR)] for continuous variables with normal and skewed distribution, respectively. For categorical variables, frequencies and percentages were provided. Associations between baseline eGFR in 2011 derived by different equations and 2-year mortality (2011–2013), 4-year mortality (2011–2015) and 7-year mortality (2011–2018) were evaluated using logistic regression analysis, since the exact time of death was not available. eGFR was analyzed in two approaches, either as a continuous variable or as a binary variable with a cutoff of 60 mL/min/1.73 m2, which is the recommended threshold for defining CKD solely based on eGFR. Multivariate logistic regression analyses were further performed, incorporating the confounding effect of age, gender, smoking, drinking, hypertension, diabetes, dyslipidemia, heart disease, stroke, tumor, BUN, UA and hs-CRP.

The discriminative performance of eGFR derived by different equations was assessed with the area under the curve (AUC) of each receiver operating characteristics (ROC) curve. Differences between the equation with the highest AUC (CKD-EPIcys) and others were compared with the DeLong method. In addition, continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were computed to evaluate the predictive value of equation-derived eGFR for all-cause mortality at 2, 4 and 7 years. Continuous NRI is a parameter that measures the reclassification improvement of an updated model compared with the initial model. It is independent of any prespecified risk categories but relies on the proportions of individuals with events correctly assigned a higher probability and individuals without events correctly assigned a lower probability. IDI indicates the difference in average predicted risks between the individuals with and without the event between the two models.

All statistical analyses were performed using packages implemented in R version 4.0.2 (http://cran.us.r-project.org/). A two-sided P-value <0.05 was considered statistically significant.

RESULTS

Baseline characteristics and the prevalence of CKD

A total number of 8160 participants were included in Cohort 1 (2011–2013) after excluding subjects with incomplete data, among whom 163 died by the year 2013. Similarly, 8154 and 8020 participants were included in Cohort 2 (2011–2015) and Cohort 3 (2011–2018), respectively, among whom 465 died by the year 2015 and 953 died by 2018 (Figure 1). Baseline characteristics of the participants were generally similar in three cohorts and are presented in Table 1.

Table 1.

Baseline characteristics

Characteristics	Cohort 1 (2011–2013)	Cohort 2 (2011–2015)	Cohort 3 (2011–2018)
Patients, n	8160	8154	8020
Gender, n (%)
Male	3807 (46.7)	3818 (46.8)	3746 (46.7)
Female	4353 (53.3)	4336 (53.2)	4274 (53.3)
Age ( years), mean (SD)	60.45 (10.24)	60.40 (10.23)	60.42 (10.25)
SBP (mmHg), mean (SD)	131.84 (25.42)	131.75 (25.44)	131.79 (25.78)
DBP (mmHg), mean (SD)	75.59 (12.22)	75.49 (12.17)	75.52 (12.15)
Creatinine (mg/dl), median (IQR)	0.76 (0.66, 0.89)	0.76 (0.66, 0.88)	0.76 (0.66, 0.88)
Cystatin C (mg/l), median (IQR)	0.98 (0.86, 1.14)	0.98 (0.86, 1.13)	0.98 (0.86, 1.13)
BUN (mg/dl), mean (SD)	15.86 (4.76)	15.89 (4.75)	15.88 (4.74)
UA (mg/dl), mean (SD)	4.47 (1.28)	4.47 (1.27)	4.46 (1.27)
HbA1c (%), median (IQR)	5.1 (4.9–5.4)	5.1 (4.9–5.4)	5.1 (4.9–5.4)
FBG (mg/dl), median (IQR)	102.4 (94.2–113.8)	102.4 (94.3–113.6)	102.4 (94.3–113.7)
TC (mg/dl), mean (SD)	193.23 (38.85)	193.06 (38.74)	193.11 (39.01)
TG (mg/dl), median (IQR)	106.20 (75.23–155.76)	102.42 (94.32–113.58)	105.32 (75.23–154.88)
LDL-C (mg/dl), mean (SD)	116.31 (35.04)	116.20 (35.02)	116.21 (35.12)
HDL-C (mg/dl), mean (SD)	51.12 (15.37)	51.17 (15.29)	51.17 (15.39)
hs-CRP (mg/l), median (IQR)	1.06 (0.56–2.24)	1.05 (0.55–2.23)	1.06 (0.55–2.27)
eGFR (mL/min/1.73 m2)
CKD-EPIcr	91.03 (15.55)	91.17 (15.52)	91.16 (15.55)
CKD-EPIcr-JPN	73.88 (12.62)	73.99 (12.59)	73.99 (12.62)
CKD-EPIcys	77.54 (21.16)	77.63 (21.17)	77.66 (21.22)
CKD-EPIcr-cys	84.15 (17.97)	84.26 (17.98)	84.28 (18.03)
CKD-EPIcr fit without race	90.29 (14.66)	90.41 (14.61)	90.41 (14.65)
CKD-EPIcr-cys fit without race	87.81 (18.52)	87.92 (18.52)	87.95 (18.57)
MDRD, median (IQR)	89.68 (77.17–103.65)	89.78 (77.35–103.77)	89.94 (77.25–103.80)
MDRD-JPN, median (IQR)	72.46 (62.35–83.75)	72.54 (62.50–83.85)	72.67 (62.41–83.87)
MDRD-CN, median (IQR)	104.50 (89.00–122.27)	104.55 (89.19–122.52)	104.83 (89.19–122.57)
Hypertension, n (%)	2949 (42.7)	2929 (42.4)	2894 (42.6)
Diabetes, n (%)	1443 (18.0)	1433 (17.9)	1401 (17.8)
Dyslipidemia, n (%)	5269 (65.0)	5243 (64.8)	5155 (64.8)
Heart disease, n (%)	1066 (13.2)	1033 (12.8)	1023 (12.9)
Stroke, n (%)	209 (2.6)	199 (2.5)	199 (2.5)
Tumor, n (%)	78 (1.0)	81 (1.0)	78 (1.0)
Smoking, n (%)
Never	4945 (61.9)	4939 (62.0)	4849 (61.8)
Used to	716 (9.0)	692 (8.7)	681 (8.7)
Smoking	2322 (29.1)	2338 (29.3)	2312 (29.5)
Drinking, n (%)
Never	5534 (68.0)	5530 (68.0)	5426 (67.9)
Less than once per month	583 (7.2)	586 (7.2)	573 (7.2)
More than once per month	2018 (24.8)	2012 (24.8)	1994 (24.9)

Mean or median eGFR varied greatly between those derived from different equations. For example, the mean/median eGFR was highest when estimated with the MDRD-CN equation and lowest when derived by the MDRD-JPN equation (104.50 and 72.46 mL/min/1.73 m2, respectively). When defined as eGFR <60 mL/min/1.73 m2,  the prevalence of CKD in participants ranged between 2.4% (MDRD-CN) and 21.1% (CKD-EPIcys), as seen in Table 2. The distribution density of eGFR estimated by different equations in Cohort 1 is illustrated in Figure 2. Similar results were observed in Cohort 2 and Cohort 3 and thus were not repeatedly displayed.

Table 2.

Prevalence of CKD (eGFR <60 mL/min/1.73 m2) estimated by different equations at baseline in Cohort 1

	Cohort 1 (2011–2013)
eGFR derived from different equations	>65 years old (n = 2511), n (%)	≤65 years old (n = 5649), n (%)	Overall (N = 8160), n (%)
CKD-EPIcr	270 (10.8)	69 (1.2)	339 (4.2)
CKD-EPIcr-JPN	745 (29.7)	378 (6.7)	1123 (13.8)
CKD-EPIcys	1227 (48.9)	494 (8.7)	1721 (21.1)
CKD-EPIcr-cys	612 (24.4)	150 (2.7)	762 (9.3)
CKD-EPIcr fit without race	264 (10.5)	68 (1.2)	332 (4.1)
CKD-EPIcr-cys fit without race	481 (19.2)	117 (2.1)	598 (7.3)
MDRD	291 (11.6)	133 (2.4)	424 (5.2)
MDRD-JPN	855 (34.1)	772 (13.7)	1627 (19.9)
MDRD-CN	138 (5.5)	54 (1.0)	192 (2.4)

FIGURE 2:

Distribution of eGFR in Cohort 1.

Association of eGFR derived from different equations with all-cause mortality

Compared with eGFR ≥60 mL/min/1.73 m2, those with a decreased eGFR (<60 mL/min/1.73 m2) had a significantly higher mortality rate in 2 years, no matter which eGFR equation was used (Table 3). The odds ratio (OR) was highest for eGFR estimated by CKD-EPIcys, which was 4.94 (95% CI 3.61–6.76). In descending order, the OR was 4.48 (95% CI 2.58–7.77) for eGFR estimated by MDRD-CN, 4.26 (95% CI 3.01–6.01) for CKD-EPIcr-cys, 4.18 (95% CI 2.89–6.06) for CKD-EPIcr-cys fit without race, 3.44 (95% CI 2.13–5.58) for CKD-EPIcr fit without race, 3.37 (95% CI 2.08–5.45) for CKD-EPIcr, 2.95 (95% CI 1.86–4.67) for MDRD, 2.94 (95% CI 2.11–4.14) for CKD-EPIcr-JPN and 2.02 (95% CI 1.45–2.82) for MDRD-JPN. Similar trends were observed when the association between eGFR and 4-year mortality in Cohort 2 and 7-year mortality in Cohort 3 were examined. All P-values were <0.001. The associations were attenuated after adjusting by multiple covariates, including gender, age, smoking, drinking, hypertension, diabetes, dyslipidemia, heart disease, stroke, tumor, BUN, UA and hs-CRP. After adjustment, the association remains statistically significant between 2-year mortality and eGFR estimated by MDRD-CN, CKD-EPIcys and CKD-EPIcr-cys fit without race, with an adjusted OR of 2.43 (95% CI 1.15–5.13), 1.85 (95% CI 1.19–2.89) and 1.74 (95% CI 1.04–2.91), respectively. When the association between 4-year mortality and eGFR was examined in Cohort 2, except for MDRD-CN, CKD-EPIcys and CKD-EPIcr-cys fit without race, the eGFRs from CKD-EPIcr, CKD-EPIcr fit without race and CKD-EPIcr-cys were also significantly associated with 4-year mortality. Similar results were observed for 7-year mortality in Cohort 3.

Table 3.

Univariate and multivariate logistic models for all-cause mortality at 2, 4 and 7 years by different eGFR equations (eGFR <60 mL/min/1.73 m2 versus ≥60 mL/min/1.73 m2)

	Cohort 1 (2011–2013)			Cohort 2 (2011–2015)			Cohort 3 (2011–2018)
Equation	OR	95% CI	P-value	OR	95% CI	P-value	OR	95% CI	P-value
CKD-EPIcr
Unadjusted	3.37	2.08–5.45	<0.001	4.72	3.54–6.28	<0.001	4.46	3.52–5.65	<0.001
Model 1	1.18	0.71–1.97	0.514	1.77	1.29–2.41	<0.001	1.60	1.23–2.08	<0.001
Model 2	1.31	0.73–2.33	0.364	1.77	1.24–2.53	0.002	1.54	1.14–2.07	0.004
Model 3	1.38	0.73–2.60	0.318	1.82	1.23–2.71	0.003	1.45	1.05–2.01	0.024
CKD-EPIcr-JPN
Unadjusted	2.94	2.10–4.12	<0.001	3.17	2.58–3.90	<0.001	3.11	2.65–3.64	<0.001
Model 1	1.18	0.82–1.71	0.377	1.34	1.07–1.69	0.012	1.24	1.03–1.49	0.020
Model 2	1.13	0.73–1.75	0.574	1.27	0.97–1.66	0.085	1.18	0.96–1.45	0.110
Model 3	1.15	0.71–1.85	0.575	1.26	0.94–1.69	0.127	1.10	0.88–1.38	0.392
CKD-EPIcys
Unadjusted	4.94	3.61–6.76	<0.001	4.79	3.96–5.80	<0.001	5.05	4.38–5.82	<0.001
Model 1	1.88	1.30–2.71	0.001	1.88	1.50–2.35	<0.001	1.86	1.58–2.20	<0.001
Model 2	1.95	1.27–2.98	0.002	1.89	1.46–2.44	<0.001	1.78	1.47–2.15	<0.001
Model 3	1.85	1.19–2.89	0.007	1.82	1.39–2.39	<0.001	1.68	1.38–2.05	<0.001
CKD-EPIcr-cys
Unadjusted	4.26	3.01–6.01	<0.001	5.16	4.16–6.39	<0.001	5.34	4.50–6.33	<0.001
Model 1	1.43	0.97–2.10	0.073	1.89	1.49–2.41	<0.001	1.84	1.51–2.23	<0.001
Model 2	1.53	0.98–2.39	0.062	1.90	1.44–2.51	<0.001	1.73	1.39–2.16	<0.001
Model 3	1.53	0.94–2.50	0.090	1.96	1.43–2.67	<0.001	1.68	1.32–2.14	<0.001
CKD-EPIcr fit without race
Unadjusted	3.44	2.13–5.58	<0.001	4.85	3.64–6.46	<0.001	4.59	3.62–5.83	<0.001
Model 1	1.21	0.73–2.02	0.455	1.82	1.34–2.49	<0.001	1.66	1.27–2.16	<0.001
Model 2	1.36	0.76–2.43	0.296	1.83	1.28–2.62	0.001	1.62	1.20–2.18	0.002
Model 3	1.44	0.76–2.73	0.259	1.90	1.28–2.83	0.002	1.54	1.11–2.13	0.010
CKD-EPIcr-cys fit without race
Unadjusted	4.18	2.89–6.06	<0.001	5.38	4.28–6.76	<0.001	5.49	4.57–6.61	<0.001
Model 1	1.40	0.93–2.11	0.106	1.99	1.54–2.57	<0.001	1.90	1.54–2.34	<0.001
Model 2	1.68	1.06–2.66	0.029	2.02	1.50–2.70	<0.001	1.77	1.40–2.24	<0.001
Model 3	1.74	1.04–2.91	0.034	2.10	1.52–2.91	<0.001	1.72	1.32–2.22	<0.001
MDRD
Unadjusted	2.95	1.86–4.67	<0.001	3.51	2.65–4.64	<0.001	3.37	2.70–4.22	<0.001
Model 1	1.30	0.80–2.10	0.287	1.61	1.19–2.18	0.002	1.52	1.19–1.95	0.001
Model 2	1.42	0.82–1.46	0.212	1.53	1.08–2.17	0.018	1.43	1.08–1.89	0.013
Model 3	1.51	0.82–2.77	0.183	1.53	1.04–2.26	0.031	1.34	0.99–1.83	0.061
MDRD-JPN
Unadjusted	2.02	1.45–2.82	<0.001	2.19	1.79–2.67	<0.001	2.19	1.88–2.54	<0.001
Model 1	1.04	0.74–1.48	0.809	1.19	0.96–1.48	0.114	1.17	0.99–1.38	0.068
Model 2	1.07	0.71–1.61	0.759	1.17	0.91–1.50	0.233	1.10	0.91–1.34	0.304
Model 3	1.10	0.70–1.73	0.679	1.16	0.88–2.54	0.289	1.03	0.84–1.28	0.750
MDRD-CN
Unadjusted	4.48	2.58–7.77	<0.001	4.93	3.44–7.06	<0.001	4.38	3.22–5.97	<0.001
Model 1	2.00	1.13–3.54	0.018	2.19	1.49–3.21	<0.001	1.93	1.38–2.71	<0.001
Model 2	2.15	1.11–4.16	0.023	2.10	1.35–3.27	0.001	1.82	1.24–2.66	0.002
Model 3	2.43	1.15–5.13	0.020	2.16	1.31–3.57	0.003	1.70	1.12–2.58	0.013

Model 1 was adjusted by gender and age. Model 2 was adjusted by Model 1 plus smoking, drinking, hypertension, diabetes, dyslipidemia, heart disease, stroke and tumor. Model 3 was adjusted by Model 2 plus blood urea nitrogen, uric acid and hs-CRP.

As a continuous variable, the risk for 2-year mortality increased 57% per 10 mL/min/1.73 m2 decrease in eGFR estimated by CKD-EPIcys (Table 4). After adjustment, the association remains statistically significant between 2-year mortality and eGFR estimated by CKD-EPIcys, CKD-EPIcr-cys and CKD-EPIcr-cys fit without race with an adjusted OR of 1.29 (95% CI 1.13–1.48), 1.31 (95% CI 1.13–1.52) and 1.30 (95% CI 1.13–1.49), respectively. Similar results were observed in Cohort 2 and Cohort 3.

Table 4.

Univariate and multivariate logistic models for all-cause mortality at 2, 4 and 7 years by different eGFR equations (continuous variable)

	Cohort 1 (2011–2013)			Cohort 2 (2011–2015)			Cohort 3 (2011–2018)
Equation	OR	95% CI	P-value	OR	95% CI	P-value	OR	95% CI	P-value
CKD-EPIcr
Unadjusted	1.43	1.32–1.55	<0.001	1.47	1.39–1.54	<0.001	1.49	1.43–1.55	<0.001
Model 1	1.07	0.96–1.20	0.208	1.11	1.04–1.19	0.002	1.09	1.04–1.15	0.001
Model 2	1.09	0.96–1.24	0.164	1.11	1.03–1.20	0.007	1.08	1.02–1.15	0.012
Model 3	1.14	0.98–1.33	0.096	1.15	1.04–1.26	0.004	1.07	1.00–1.15	0.063
CKD-EPIcr-JPN
Unadjusted	1.55	1.40–1.71	<0.001	1.60	1.50–1.71	<0.001	1.63	1.55–1.72	<0.001
Model 1	1.09	0.95–1.25	0.208	1.14	1.05–1.24	0.002	1.12	1.05–1.19	0.001
Model 2	1.12	0.96–1.31	0.164	1.14	1.04–1.26	0.007	1.10	1.02–1.18	0.012
Model 3	1.18	0.97–1.42	0.096	1.18	1.05–1.33	0.004	1.09	1.00–1.19	0.063
CKD-EPIcys
Unadjusted	1.57	1.45–1.69	<0.001	1.56	1.48–1.64	<0.001	1.56	1.50–1.62	<0.001
Model 1	1.25	1.13–1.38	<0.001	1.25	1.17–1.32	<0.001	1.21	1.16–1.26	<0.001
Model 2	1.28	1.14–1.43	<0.001	1.25	1.17–1.34	<0.001	1.20	1.14–1.26	<0.001
Model 3	1.29	1.13–1.48	<0.001	1.26	1.17–1.36	<0.001	1.19	1.13–1.26	<0.001
CKD-EPIcr-cys
Unadjusted	1.56	1.44–1.69	<0.001	1.59	1.51–1.68	<0.001	1.61	1.55–1.68	<0.001
Model 1	1.22	1.10–1.35	<0.001	1.25	1.17–1.33	<0.001	1.22	1.16–1.28	<0.001
Model 2	1.24	1.10–1.41	<0.001	1.25	1.16–1.34	<0.001	1.20	1.13–1.27	<0.001
Model 3	1.31	1.13–1.52	<0.001	1.31	1.19–1.43	<0.001	1.21	1.14–1.29	<0.001
CKD-EPIcr fit without race
Unadjusted	1.50	1.34–1.58	<0.001	1.50	1.42–1.58	<0.001	1.53	1.47–1.60	<0.001
Model 1	1.09	0.97–1.22	0.151	1.13	1.05–1.21	0.001	1.10	1.04–1.17	<0.001
Model 2	1.11	0.97–1.27	0.121	1.13	1.04–1.22	0.004	1.09	1.02–1.16	0.009
Model 3	1.16	0.99–1.37	0.067	1.17	1.06–1.29	0.003	1.08	1.00–1.16	0.048
CKD-EPIcr-cys fit without race
Unadjusted	1.53	1.42–1.65	<0.001	1.56	1.49–1.64	<0.001	1.58	1.52–1.64	<0.001
Model 1	1.22	1.10–1.34	<0.001	1.24	1.17–1.32	<0.001	1.21	1.16–1.27	<0.001
Model 2	1.24	1.11–1.39	<0.001	1.24	1.16–1.33	<0.001	1.19	1.13–1.26	<0.001
Model 3	1.30	1.13–1.49	<0.001	1.30	1.19–1.41	<0.001	1.21	1.14–1.28	<0.001
MDRD
Unadjusted	1.16	1.07–1.25	<0.001	1.19	1.13–1.24	<0.001	1.18	1.14–1.22	<0.001
Model 1	1.01	0.94–1.08	0.803	1.04	1.00–1.09	0.077	1.04	1.01–1.07	0.023
Model 2	1.02	0.94–1.11	0.630	1.05	0.99–1.10	0.086	1.03	1.00–1.07	0.076
Model 3	1.03	0.94–1.13	0.522	1.05	0.99–1.12	0.081	1.03	0.98–1.07	0.227
MDRD-JPN
Unadjusted	1.20	1.09–1.31	<0.001	1.24	1.17–1.31	<0.001	1.23	1.18–1.28	<0.001
Model 1	1.01	0.93–1.10	0.803	1.05	0.99–1.11	0.077	1.05	1.01–1.09	0.023
Model 2	1.03	0.93–1.14	0.630	1.06	0.99–1.13	0.086	1.04	1.00–1.09	0.076
Model 3	1.04	0.92–1.17	0.522	1.07	0.99–1.15	0.081	1.03	0.98–1.09	0.227
MDRD-CN
Unadjusted	1.12	1.06–1.20	<0.001	1.16	1.11–1.20	<0.001	1.16	1.12–1.19	<0.001
Model 1	1.01	0.95–1.06	0.812	1.03	1.00–1.07	0.071	1.03	1.00–1.06	0.021
Model 2	1.02	0.95–1.08	0.651	1.04	1.00–1.08	0.080	1.03	1.00–1.06	0.069
Model 3	1.02	0.95–1.10	0.548	1.04	1.00–1.09	0.076	1.02	0.99–1.06	0.207

Data in continuous models are ORs per 10 mL/min/1.73 m2 decrease. Model 1 was adjusted by gender and age. Model 2 was adjusted by Model 1 plus smoking, drinking, hypertension, diabetes, dyslipidemia, heart disease, stroke and tumor. Model 3 was adjusted by Model 2 plus blood urea nitrogen, uric acid and hs-CRP.

Reclassification and discrimination statistics for all-cause mortality

The discriminatory power of each equation was first assessed by calculating the AUC for predicting death at specific time points (2-, 4- and 7-year). The AUC ranged between 0.577 and 0.744 for predicting 2-year mortality, 0.594 and 0.731 for predicting 4-year mortality and 0.595 and 0.733 for predicting 7-year mortality (Table 5). The AUC was highest for eGFR derived by CKD-EPIcys and lowest for that calculated by MDRD and MDRD-JPN in all three cohorts. Differences were statistically significant between the equation with the highest AUC (CKD-EPIcys) and other equations except for the CKD-EPIcr-cys equation (Figure 3 and Table 5). The finding was consistent among all three cohorts.

Table 5.

Reclassification and discrimination statistics for all-cause mortality at 2, 4 and 7 years by different eGFR equations

	AUC		NRI		IDI
Equation	Estimate (95% CI)	P-value	Estimate (95% CI)	P-value	Estimate (95% CI)	P-value
Cohort 1 (2011–2013)
CKD-EPIcys	0.744 (0.705–0.784)	–	Reference	–	Reference	–
CKD-EPIcr	0.676 (0.637–0.715)	<0.001	−0.520 (−0.668 to −0.372)	<0.001	−0.011 (−0.014 to −0.007)	<0.001
CKD-EPIcr-JPN	0.676 (0.636–0.715)	<0.001	−0.519 (−0.667 to −0.371)	<0.001	−0.011 (−0.014 to −0.007)	<0.001
CKD-EPIcr-cys	0.733 (0.695–0.771)	0.056	−0.290 (−0.444 to −0.136)	<0.001	−0.002 (−0.004–0.000)	0.128
CKD-EPIcr fit without race	0.687 (0.650–0.725)	0.004	−0.471 (−0.620 to −0.321)	<0.001	−0.010 (−0.013 to −0.006)	<0.001
CKD-EPIcr-cys fit without race	0.732 (0.693–0.770)	0.011	−0.328 (−0.481 to −0.174)	<0.001	−0.002 (−0.004–0.000)	0.107
MDRD/MDRD-JPN	0.577 (0.528–0.626)	<0.001	−0.752 (−0.891 to −0.612)	<0.001	−0.020 (−0.025 to −0.015)	<0.001
MDRD-CN	0.578 (0.530–0.627)	<0.001	−0.730 (−0.871 to −0.588)	<0.001	−0.020 (−0.025 to −0.015)	<0.001
Cohort 2 (2011–2015)
CKD-EPIcys	0.731 (0.706–0.756)	–	Reference	–	Reference	–
CKD-EPIcr	0.677 (0.652–0.702)	<0.001	−0.448 (−0.539 to −0.358)	<0.001	−0.025 (−0.039 to −0.019)	<0.001
CKD-EPIcr-JPN	0.676 (0.652–0.701)	<0.001	−0.455 (−0.545 to −0.364)	<0.001	−0.025 (−0.030 to −0.020)	<0.001
CKD-EPIcr-cys	0.724 (0.699–0.748)	0.074	−0.197 (−0.290 to −0.104)	<0.001	−0.003 (−0.005–0.000)	0.063
CKD-EPIcr fit without race	0.686 (0.661–0.710)	<0.001	−0.423 (−0.514 to −0.332)	<0.001	−0.023 (−0.028 to −0.0173)	<0.001
CKD-EPIcr-cys fit without race	0.723 (0.670–0.748)	0.015	−0.252 (−0.345 to −0.160)	<0.001	−0.003 (−0.005–0.000)	0.051
MDRD/MDRD-JPN	0.594 (0.565–0.623)	<0.001	−0.708 (−0.794 to −0.623)	<0.001	−0.048 (−0.055 to −0.042)	<0.001
MDRD-CN	0.598 (0.569–0.626)	<0.001	−0.713 (−0.798 to −0.628)	<0.001	−0.048 (−0.054 to −0.041)	<0.001
Cohort 3 (2011–2018)
CKD-EPIcys	0.733 (0.716–0.751)	–	Reference	–	Reference	–
CKD-EPIcr	0.687 (0.670–0.705)	<0.001	−0.438 (−0.503 to −0.372)	<0.001	−0.039 (−0.045 to −0.033)	<0.001
CKD-EPIcr-JPN	0.687 (0.669–0.704)	<0.001	−0.438 (−0.503 to −0.372)	<0.001	−0.039 (−0.045 to −0.033)	<0.001
CKD-EPIcr-cys	0.729 (0.712–0.746)	0.126	−0.170 (−0.238 to −0.103)	<0.001	−0.005 (−0.008 to −0.002)	<0.001
CKD-EPIcr fit without race	0.696 (0.679–0.713)	<0.001	−0.409 (−0.475 to −0.343)	<0.001	−0.036 (−0.043 to −0.030)	<0.001
CKD-EPIcr-cys fit without race	0.727 (0.710–0.744)	0.010	−0.230 (−0.297 to −0.162)	<0.001	−0.006 (−0.008 to −0.003)	<0.001
MDRD/MDRD-JPN	0.595 (0.575–0.616)	<0.001	−0.702 (−0.764 to −0.640)	<0.001	−0.078 (−0.085 to −0.071)	<0.001
MDRD-CN	0.602 (0.581–0.622)	<0.001	−0.704 (−0.766 to −0.642)	<0.001	−0.076 (−0.083 to −0.069)	<0.001

FIGURE 3:

ROC curves for all-cause mortality at (A) 2 years, (B) 4 years and (C) 7 years predicted by different eGFR equations.

To examine whether different equation-based eGFRs contribute to distinct risk reclassification power, NRI and IDI were further calculated. Compared with the CKD-EPIcys equation, the values of NRI and IDI were −0.752 (P < 0.001) and −0.020 (P < 0.001) for the MDRD and MDRD-JPN equation, −0.730 (P < 0.001) and −0.020 (P < 0.001) for MDRD-CN, −0.520 (P < 0.001) and −0.011 (P < 0.001) for CKD-EPIcr, −0.519 (P < 0.001) and −0.011 (P < 0.001) for CKD-EPIcr-JPN, −0.290 (P <0.001) and −0.002 (P = 0.128) for CKD-EPIcr-cys, −0.471 (P < 0.001) and −0.010 (P < 0.001) for CKD-EPIcr fit without race, −0.328 (P < 0.001) and −0.002 (P = 0.107) for CKD-EPIcr-cys fit without race. Similar results were observed for 4-year mortality in Cohort 2 and 7-year mortality in Cohort 3 (Table 5).

DISCUSSION

In the present study we showed that decreased eGFR estimated by all available equations was a risk factor for all-cause mortality at 2, 4 and 7 years in this large cohort of unselected middle-aged/elderly Chinese. The prevalence of CKD varied greatly when different eGFR equations were used. Among all available equations, CKD-EPIcys revealed the highest overall discriminative power and outweighed other formulations in correctly assigning higher predictive risk to those death cases. When cystatin C is not available, eGFR derived from MDRD-CN showed a good prediction for mortality with a cutoff of 60 mL/min/1.73 m2. Similar results were observed in three cohorts with 2-, 4- and 7-year mortality as an endpoint, respectively, indicating that the finding in this study is robust.

Serum creatinine is without doubt one of the earliest and the most widely accepted biomarker for evaluation of kidney filtration function. However, serum creatinine is known to be affected by race, gender, age, muscle mass, diet factors and specific medications [24, 25]. Cystatin C has attracted more and more attention in recent years. As a protein produced by all nucleated cells, cystatin C maintains a stable level in the human body [25, 26]. It has been established that cystatin C is filtered through the glomerulus and could be almost completely reabsorbed by tubular cells [27-30]. After that, it will be degraded in the lysosome, so it can neither return to the blood nor be secreted into the urine again [31, 32]. The serum concentration of cystatin C is susceptible to the change in GFR since glomerular filtration is the unique key process for it to be cleared from circulation. Compared with creatinine, cystatin C is less affected by muscular mass, race, gender and age [33]. Cystatin C may be more useful in detecting kidney disease in children, in the elderly and in individuals with underlying conditions affecting muscle mass [27]. Based on these findings, the CKD-EPIcys equation and CKD-EPIcr-cys equation were developed. Studies demonstrated that compared to eGFR calculated based on creatinine or cystatin C alone, the eGFR derived by the creatinine-cystatin C equation is closer to the directly measured GFR [5, 34]. However, a few conditions including hyperthyroidism, liver disease and high doses of corticosteroids were reported to influence the level of cystatin C [31]. Apart from being a better endogenous biomarker for assessing kidney filtration function [25, 31], cystatin C has been suggested to associate with all-cause mortality in the general population [35, 36]. In addition to the general population, increased cystatin C level is also reported to be associated with a higher mortality rate in patients with hypertension, coronary heart disease and stroke [37-39]. That may explain why CKD-EPIcys outperforms creatinine-based equations in predicting death occurrences in our study.

In this study, we creatively investigated the discriminative power and reclassification index of different eGFR equations for predicting death at specific time points in a large Chinese cohort. It should be noted that although the CKD-EPIcys equation showed a better discriminative power than other equations, the calculations of AUC, NRI and IDI were all based on models incorporating eGFR as a continuous variable. The OR for eGFR derived by CKD-EPIcys was much lower when it was incorporated as a categorical parameter with a cutoff of 60 mL/min/1.73 m2. On the other hand, Zhu et al. [40] showed that when technetium 99m diethylenetriaminepentaacetate renal dynamic imaging was adopted as a gold standard, the CKD-EPIcys equation might perform no better than other available equations in accuracy among the Chinese population. These findings indicate that to maximize the predictive value of CKD-EPIcys, a more appropriate threshold should be determined in the Chinese population or the equation could be modified by multiplying with a proper coefficient. Despite the good performance of cystatin C, the cost of it is significantly higher than that of creatinine. Results demonstrated that decreased eGFR derived from MDRD-CN equation is also an independent risk factor for all-cause mortality. When cystatin C is not available, MDRD-CN may be a highly cost-effective choice. As regards the two newly developed CKD-EPI equations fit without race in 2021, these challenge the previous CKD-EPI equation in that race in eGFR equations was a social but not a biological construct and its inclusion could even lead to the systemic racism in medicine [6, 41]. The results in our study showed that discriminative curves fitted with eGFR estimated by CKD-EPI equations in 2012 and that estimated by CKD-EPI equations fit without race in 2021 were quite similar. However, the CKD-EPIcr fit without race does perform better than the original CKD-EPIcr equation in our study (Supplementary data, Appendix Table S2).

In addition to the above mentioned, two important but commonly ignored points should be noted. First, a specific eGFR equation could be suitable for a population with different ranges of GFRs. Since the MDRD equation was originally developed with data from the CKD population, its major limitation is systematic underestimation of GFR at higher GFR levels [3, 42]. On the contrary, CKD-EPIcr equation is believed to perform better when GFR is >60 mL/min/1.73 m2 [4], while most experts agree that the most appropriate range should be 90–120 mL/min/1.73 m2 [43]. Second, different methods of serum creatinine measurement could introduce various levels of bias to the eGFR estimation when different equations are used. For example, Jaffe's method was used to measure creatinine when the MDRD equation was developed, while in CKD-EPI study, the creatinine was measured with Jaffe's method but recalibrated to the enzymatic method [3, 4]. Please refer to Supplementary data, Appendix Table S3 for the complete list of serum creatinine measurement methods during the development of different eGFR equations. Küme et al. [44] showed that the Jaffe's method provided higher serum creatinine values than the enzymatic method, especially at the low creatinine concentrations. In the CHARLS, serum creatinine was measured with the rate-blanked and compensated Jaffe's method by accounting for the noncreatinine chromogen interference. The compensated Jaffe's method proved to be negatively biased compared with the original Jaffe assay, causing predictable, clinically significant overestimation of GFR when the MDRD equation was used [45]. However, the value measured by the compensated Jaffe's assay and that by the enzymatic method were not significantly different [45]. It suggests that compared to MDRD equations, CKD-EPI equations may be more suitable for eGFR estimation in the CHARLS. That may partially explain why CKD-EPI series equations perform better than MDRD series equations in predicting prognosis in our study.

There are several strengths in our study. First, participants were recruited from 150 randomly selected counties all over the country with a probability-proportional-to-size method. The population was highly representative. The large sample size, well-designed study protocol and study execution by well-trained staff add to the data reliability. Second, three cohorts with different durations of follow-up were separately analyzed in this study to confirm the robustness of the conclusions. Third, the discriminative power of different equations was assessed by multiple methods, including AUC, NRI and IDI, which provided us with a whole picture of the advantages and weaknesses of each equation. Limitations are also apparent in our study. First, the reason for death was not available from the CHARLS database, which prevents us from further analyzing the association between eGFR and cause-specific mortalities, especially cardiovascular mortality. Second, the CHARLS focuses on the middle-aged and elder population, and the association found in this study should be examined in a younger population.

In conclusion, this study showed that a lower eGFR level derived from all currently available equations is a risk factor for all-cause mortality at 2, 4 and 7 years. Among these, models using the CKD-EPIcys equation displayed the best discriminative power, especially when incorporated as a continuous variable. Our findings suggest that eGRF derived from CKD-EPIcys may be most helpful in identifying the population with poorer survival. However, it still warrants further research in determining the Chinese-specific threshold and Chinese-modified CKD-EPIcys equation.

Click here for additional data file.