Advertisement

Comparison of bias and accuracy using cystatin C and creatinine in CKD-EPI equations for GFR estimation

  • Lu-Xi Zou
    Affiliations
    Xuzhou Medical University, Xuzhou, Jiangsu, China
    Search for articles by this author
  • Ling Sun
    Correspondence
    Corresponding author: Division of Nephrology, Xuzhou Central Hospital, Medical College of Southeast University, No.199, Jiefang South Road, Xuzhou, Jiangsu, China. Postcode: 221009. Ph: +86 0516 83956891 / Fax:+86 0516 83840486
    Affiliations
    Division of Nephrology, Xuzhou Central Hospital, Medical College of Southeast University, Xuzhou, Jiangsu, China

    Department of Clinical Medicine, Xuzhou Medical University, Xuzhou, Jiangsu, China
    Search for articles by this author
  • Susanne B. Nicholas
    Correspondence
    Corresponding author: Divisions of Nephrology and Endocrinology, Department of Medicine, David Geffen School of Medicine at University of California, David Geffen School of Medicine at UCLA, Los Angeles, CA, USA. Ph: 310-206-6741, Fax: 310-825-6309
    Affiliations
    Divisions of Nephrology and Endocrinology, Department of Medicine, David Geffen School of Medicine at University of California, Los Angeles, CA, USA
    Search for articles by this author
  • Yan Lu
    Affiliations
    Department of Clinical Medicine, Xuzhou Medical University, Xuzhou, Jiangsu, China
    Search for articles by this author
  • Satyesh Sinha K
    Affiliations
    Division of Nephrology, Department of Medicine, David Geffen School of Medicine at University of California, Los Angeles, CA, USA
    Search for articles by this author
  • Ruixue Hua
    Affiliations
    Department of Clinical Medicine, Xuzhou Medical University, Xuzhou, Jiangsu, China
    Search for articles by this author
Open AccessPublished:June 08, 2020DOI:https://doi.org/10.1016/j.ejim.2020.04.044

      Highlights

      • Cystatin C based CKD-EPI performed better than CKD-EPIcrea in estimating GFR.
      • Bias of CKD-EPIcys was lowest, the gaps increased in low mGFR, non-elderly and Asian.
      • Accuracy estimates (P30) of CKD-EPIcrea/cys eGFR was highest in all subgroups.
      • CKD-EPIcrea/cys eGFR had the strongest correlation with mGFR.

      Abstract

      Background: The directly measured glomerular filtrate rate (mGFR) is the gold standard for kidney function, but it is invasive and costly. The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations have been widely used to estimate GFR, however, the comparative accuracy of estimated GFR (eGFR) using creatinine and cystatin C in CKD-EPI equations remains unclear. We performed this meta-analysis to assess the bias and accuracy of eGFR using equations of CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys in adult populations relevant to primary health care.
      Methods: Pubmed, Web of Science, EMBASE, and the Cochrane Library were searched from inception until December 2019 for related studies.
      Results: A total of 35 studies with 23,667 participants, which reported the data on the bias, and/or P30, and/or R were included. The difference in the bias of eGFR using CKD-EPIcys was 4.84 mL/min/1.73 m2 (95% CI, 1.88~7.80) lower than using CKD-EPIcrea, and 1.50 mL/min/1.73 m2 (95% CI, 0.05~2.95) lower than using CKD-EPIcrea/cys. These gaps increased in subgroups of low mGFR (<60 mL/min/1.73 m2). CKD-EPIcrea/cys eGFR achieved the highest accuracy, 7.50% higher than CKD-EPIcrea (95% CI, 4.81~10.18), and 3.21% higher than CKD-EPIcys (95% CI, -0.43~6.85); and the best correlation with mGFR, with Fisher's z transformed R of 1.20 (95% CI, 0.89-1.50).
      Conclusions: CKD-EPIcrea/cys and CKD-EPIcys gave less bias and more accurate estimates of mGFR than CKD-EPIcrea. More variables and coefficients could be added in CKD-EPI equations to achieve less bias and more accuracy in future research.

      Graphical abstract

      Keywords

      1. Introduction

      Glomerular filtrate rate (GFR) is the most useful index of kidney function. Reduction of GFR may indicate chronic kidney disease (CKD) or acute kidney injury (AKI) [
      • Stevens LA
      • Padala S
      • Levey AS
      Advances in glomerular filtration rate-estimating equations.
      ]. CKD is a growing global health concern, and a risk factor for end-stage renal disease (ESRD), cardiovascular disease (CVD), and mortality [
      • Lewis JR
      • Lim W
      • Dhaliwal SS
      • Zhu K
      • Lim EM
      • Thompson PL
      • et al.
      Estimated glomerular filtration rate as an independent predictor of atherosclerotic vascular disease in older women.
      ,
      • MacKinnon HJ
      • Wilkinson TJ
      • Clarke AL
      • Gould DW
      • O'Sullivan TF
      • Xenophontos S
      • et al.
      The association of physical function and physical activity with all-cause mortality and adverse clinical outcomes in nondialysis chronic kidney disease: a systematic review.
      ]. GFR is central to CKD diagnosis and management and is important in appropriate medication dosing. Therefore, having a precise and convenient measure of GFR would assist in preventing or delaying the progression of CKD.
      The gold standard for evaluating renal function is directly measured GFR (mGFR), using clearances of infused exogenous substances, such as inulin, iohexol, and iothalamate, which is invasive and may be too impractical and costly for large-scale application [
      • Machado JD
      • Camargo EG
      • Boff R
      • Rodrigues LD
      • Camargo JL
      • Soares AA
      • et al.
      Combined creatinine-cystatin C CKD-EPI equation significantly underestimates measured glomerular filtration rate in people with type 2 diabetes mellitus.
      ]. Estimated GFR (eGFR), using equations based on endogenous biomarkers, has been extensively used as a surrogate for mGFR, with adjustments for age, sex, and race [
      • Matsushita K
      • Selvin E
      • Bash LD
      • Astor BC
      • Coresh J
      Risk Implications of the New CKD Epidemiology Collaboration (CKD-EPI) Equation Compared With the MDRD Study Equation for Estimated GFR: The Atherosclerosis Risk in Communities (ARIC) Study.
      ]. In recent years, estimation formulas have been updated, and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations proved to be more accurate than other equations [such as the modification of diet in renal disease (MDRD) equation], particularly in those with eGFR ≥60 mL/min/1.73 m2 [
      • Levey AS
      • Stevens LA
      • Schmid CH
      • Zhang YP
      • Castro AF
      • Feldman HI
      • et al.
      A New Equation to Estimate Glomerular Filtration Rate.
      ,
      • McFadden EC
      • Hirst JA
      • Verbakel JY
      • McLellan JH
      • Hobbs FDR
      • Stevens RJ
      • et al.
      Systematic Review and Metaanalysis Comparing the Bias and Accuracy of the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration Equations in Community-Based Populations.
      ,
      • Matsushita K
      • Mahmoodi BK
      • Woodward M
      • Emberson JR
      • Jafar TH
      • Jee SH
      • et al.
      Comparison of Risk Prediction Using the CKD-EPI Equation and the MDRD Study Equation for Estimated Glomerular Filtration Rate.
      ].
      The biomarkers that are used to estimate GFR include serum creatinine and serum cystatin C [
      • Machado JD
      • Camargo EG
      • Boff R
      • Rodrigues LD
      • Camargo JL
      • Soares AA
      • et al.
      Combined creatinine-cystatin C CKD-EPI equation significantly underestimates measured glomerular filtration rate in people with type 2 diabetes mellitus.
      ]. Serum creatinine is the most common endogenous biomarker used to estimate GFR, but there are limitations to its use. For example, serum creatinine is not sensitive to true GFR decline, as it will not increase until kidney function loss is >50% [
      • Shemesh O
      • Golbetz H
      • Kriss JP
      • Myers BD
      Limitations of creatinine as a filtration marker in glomerulopathic patients.
      ]; and it could be influenced by multiple factors, e.g., secretion from renal tubules, changes in muscle mass, protein-containing meals, and physical activity, with changes of concentration ranging from 10% to 100% [
      • Delanaye P
      • Mariat C.
      The applicability of eGFR equations to different populations.
      ]. Cystatin C is a non-glycosylated protein of 13,260 daltons, encoded by a housekeeping gene, synthesized by all nucleated cells, filtered freely by the glomerulus, almost entirely reabsorbed and catabolized by proximal tubules [
      • Abrahamson M
      • Olafsson I
      • Palsdottir A
      • Ulvsback M
      • Lundwall A
      • Jensson O
      • et al.
      Structure and expression of the human cystatin C gene.
      ]. Thus, GFR is the main determinant of cystatin C concentration. Previous studies demonstrated that cystatin C concentration is less influenced by age, sex, muscle mass or diet than creatinine, but it could also be affected by factors other than GFR, including obesity, thyroid disorders, inflammation, tobacco consumption, and steroid therapy [
      • Abrahamson M
      • Olafsson I
      • Palsdottir A
      • Ulvsback M
      • Lundwall A
      • Jensson O
      • et al.
      Structure and expression of the human cystatin C gene.
      ,
      • Stevens LA
      • Schmid CH
      • Greene T
      • Li L
      • Beck GJ
      • Joffe MM
      • et al.
      Factors other than glomerular filtration rate affect serum cystatin C levels.
      ]. In addition, a standardized calibration [isotope dilution mass spectrometry (IDMS)-traceable assays for serum creatinine] for serum cystatin C measurement has not been widely applied in global performance. Thus, standardization is required to improve the cystatin C based eGFR equations in clinical practice [
      • Delanaye P
      • Mariat C.
      The applicability of eGFR equations to different populations.
      ].
      There is a significant gap between creatinine-based and cystatin C-based calculated eGFR using the CKD-EPI equation. Several clinical trials have compared the validity of cystatin C and creatinine in CKD-EPI equations for eGFR, but the results remain a matter of debate. Some studies revealed that the cystatin C-based CKD-EPI equation (CKD-EPIcys) was more accurate for GFR estimation and clinical decision making, and a better risk predictor for cardiovascular disease (CVD) and mortality than the creatinine-based CKD-EPI equation (CKD-EPIcrea), or the combined creatinine/cystatin C based CKD-EPI equation (CKD-EPIcrea/cys) [
      • Ying X
      • Jiang Y
      • Qin G
      • Qian Y
      • Shen X
      • Jiang Z
      • et al.
      Association of body mass index, waist circumference, and metabolic syndrome with serum cystatin C in a Chinese population.
      ,
      • Salminen M
      • Laine K
      • Korhonen P
      • Wasen E
      • Vahlberg T
      • Isoaho R
      • et al.
      Biomarkers of kidney function and prediction of death from cardiovascular and other causes in the elderly: A 9-year follow-up study.
      ,
      • van Deventer HE
      • Paiker JE
      • Katz IJ
      • George JA
      A comparison of cystatin C- and creatinine-based prediction equations for the estimation of glomerular filtration rate in black South Africans.
      ,
      • Schottker B
      • Herder C
      • Muller H
      • Brenner H
      • Rothenbacher D
      Clinical Utility of Creatinine- and Cystatin C-Based Definition of Renal Function for Risk Prediction of Primary Cardiovascular Events in Patients With Diabetes.
      ,
      • Abu-Assi E
      • Raposeiras-Roubin S
      • Riveiro-Cruz A
      • Rodriguez-Girondo M
      • Gonzalez-Cambeiro C
      • Alvarez-Alvarez B
      • et al.
      Creatinine-or cystatin C-based equations to estimate glomerular filtration rate in acute myocardial infarction: A disparity in estimating renal function and in mortality risk prediction.
      ,
      • Helmersson-Karlqvist J
      • Arnlov J
      • Larsson A
      Cystatin C-based glomerular filtration rate associates more closely with mortality than creatinine-based or combined glomerular filtration rate equations in unselected patients.
      ]. Other studies suggested that CKD-EPIcrea/cys performed better than CKD-EPIcrea and CKD-EPIcys in estimating GFR [
      • Chi XH
      • Li GP
      • Wang QS
      • Qi YS
      • Huang K
      • Zhang Q
      • et al.
      CKD-EPI creatinine-cystatin C glomerular filtration rate estimation equation seems more suitable for Chinese patients with chronic kidney disease than other equations.
      ,
      • Pan Y
      • Jiang S
      • Qiu DD
      • Shi JS
      • Zhou ML
      • An Y
      • et al.
      Comparing the GFR estimation equations using both creatinine and cystatin c to predict the long-term renal outcome in type 2 diabetic nephropathy patients.
      ,
      • Masson I
      • Maillard N
      • Tack I
      • Thibaudin L
      • Dubourg L
      • Delanaye P
      • et al.
      GFR Estimation Using Standardized Cystatin C in Kidney Transplant Recipients.
      ,
      • Meeusen JW
      • Rule AD
      • Voskoboev N
      • Baumann NA
      • Lieske JC
      Performance of Cystatin C- and Creatinine-Based Estimated Glomerular Filtration Rate Equations Depends on Patient Characteristics.
      ,
      • Fan L
      • Levey AS
      • Gudnason V
      • Eiriksdottir G
      • Andresdottir MB
      • Gudmundsdottir H
      • et al.
      Comparing GFR Estimating Equations Using Cystatin C and Creatinine in Elderly Individuals.
      ], and could obviate the need to consider ethnicity adjustment [
      • Teo BW
      • Zhang LX
      • Guh JY
      • Tang SCW
      • Jha V
      • Kang DH
      • et al.
      Glomerular Filtration Rates in Asians.
      ]. The Kidney Disease Improving Global Outcome (KDIGO) guidelines of 2012 recommended the use of the combined CKD-EPIcrea/cys in adults whose eGFRcrea is within the range of 45-59 ml/min/1.73 m2 without signs or markers of kidney damage [
      • Members KB.
      KDIGO 2012 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease.
      ]. However, since then, validation studies showed that neither CKD-EPIcys nor CKD-EPIcrea/cys could improve the accuracy of eGFR or the predictive ability for clinical events compared to CKD-EPIcrea, and the use of cystatin C was associated with increased medical cost [
      • Machado JD
      • Camargo EG
      • Boff R
      • Rodrigues LD
      • Camargo JL
      • Soares AA
      • et al.
      Combined creatinine-cystatin C CKD-EPI equation significantly underestimates measured glomerular filtration rate in people with type 2 diabetes mellitus.
      ]. Therefore, CKD-EPIcrea has remained the preferred equation [
      • Machado JD
      • Camargo EG
      • Boff R
      • Rodrigues LD
      • Camargo JL
      • Soares AA
      • et al.
      Combined creatinine-cystatin C CKD-EPI equation significantly underestimates measured glomerular filtration rate in people with type 2 diabetes mellitus.
      ,
      • Barr ELM
      • Maple-Brown LJ
      • Barzi F
      • Hughes JT
      • Jerums G
      • Ekinci EI
      • et al.
      Comparison of creatinine and cystatin C based eGFR in the estimation of glomerular filtration rate in Indigenous Australians: The eGFR Study.
      ,
      • Shardlow A
      • McIntyre NJ
      • Fraser SDS
      • Roderick P
      • Raftery J
      • Fluck RJ
      • et al.
      The clinical utility and cost impact of cystatin C measurement in the diagnosis and management of chronic kidney disease: A primary care cohort study.
      ,
      • Lim WH
      • Lewis JR
      • Wong G
      • Turner RM
      • Lim EM
      • Thompson PL
      • et al.
      Comparison of estimated glomerular filtration rate by the chronic kidney disease epidemiology collaboration (CKD-EPI) equations with and without Cystatin C for predicting clinical outcomes in elderly women.
      ,
      • Iliadis F
      • Didangelos T
      • Ntemka A
      • Makedou A
      • Moralidis E
      • Gotzamani-Psarakou A
      • et al.
      Glomerular filtration rate estimation in patients with type 2 diabetes: creatinine- or cystatin C-based equations?.
      ]. However, there has been no published systematic review comparing the CKD-EPIcrea, CKD-EPIcys, or CKD-EPIcrea/cys equations for GFR estimation. This study aimed to conduct a meta-analysis comparing the bias and accuracy of eGFR using these equations and to evaluate whether cystatin C is a more cost-effective biomarker than creatinine in assessing eGFR.

      2. Materials and Methods

      2.1 Literature search strategy and study selection

      Pubmed, Web of Science, EMBASE, and the Cochrane Library were searched from inception until December 1st, 2019 for studies comparing mGFR using reference methods (Table S1) with simultaneous eGFR using CKD-EPI formulas calculated from serum creatinine (IDMS calibration standardization) and cystatin C (Table S2). The search title/abstract/keywords were: “Chronic Kidney Disease Epidemiology Collaboration or CKD-EPI” AND “creatinine” AND “cystatin C”. We included studies that recruited participants >18 years old, excluded studies that recruited highly selected patients (e.g., critical illness, AKI, severe edema, and ascites), but not those prevalent in primary health care, such as CKD, CVD, diabetes or hypertension. The studies that were included reported at least one of the following items: (1) mean bias, mean difference between mGFR and eGFR; (2) accuracy, percentage of eGFR values within 30% of mGFR (P30) [
      • McFadden EC
      • Hirst JA
      • Verbakel JY
      • McLellan JH
      • Hobbs FDR
      • Stevens RJ
      • et al.
      Systematic Review and Metaanalysis Comparing the Bias and Accuracy of the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration Equations in Community-Based Populations.
      ]; (3) mean and standard deviation (SD) of mGFR and eGFR; (4) median and interquartile range (IQR) of mGFR and eGFR; (5) coefficient of correlation (R) between mGFR and eGFR.

      2.2 Data extraction and quality assessment

      Two reviewers (LS, LXZ) independently selected studies and extracted data using a standardized form, with any disagreements resolved by a third reviewer (SBN). Extracted items were recruitment setting, the number of participants, age, sex, comorbid conditions, mean (SD; standard deviation) of mGFR and eGFR, mean bias, P30 (proportion of eGFR within 30% of mGFR), and other measures of accuracy. The risk of bias was evaluated by a modified tool for quality assessment of diagnostic studies (QUADAS-2), with the applicability of 4 domains: patient selection, index test, reference standard, and flow and timing (Table S3) [
      • Whiting PF
      • Rutjes AWS
      • Westwood ME
      • Mallett S
      • Deeks JJ
      • Reitsma JB
      • et al.
      QUADAS-2: A Revised Tool for the Quality Assessment of Diagnostic Accuracy Studies.
      ,
      • Tampi RR
      • Tampi DJ
      • Balachandran S
      • Srinivasan S
      Antipsychotic use in dementia: a systematic review of benefits and risks from meta-analyses.
      ].

      2.3 Data synthesis and analysis

      Using random-effects model, we analyzed the within-study comparisons of (a) difference in bias among CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys in studies that compared with mGFR, and (b) difference in accuracy among CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, both stratified into subgroups by levels of mGFR (<60 ml/min/1.73m2 and ≥60 ml/min/1.73m2), age (<70 years and ≥70 years), and race/ethnicity (Asian and non-Asian). The difference in bias was calculated by the differences in mean absolute bias of eGFR using CKD-EPIcrea, CKD-EPIcys, CKD-EPIcrea/cys equations. Accuracy was expressed as P30, the difference in mean accuracy was calculated as the differences in accuracy between eGFR by subtracting one equation from another. Subgroup analyses were stratified by clinically relevant categories of mGFR (<60 and ≥60mL/min/1.73 m2).
      Heterogeneity was quantified by the I2 statistic [
      • Higgins JPT
      • Thompson SG.
      Quantifying heterogeneity in a meta-analysis.
      ]. If there was high heterogeneity, sensitivity analyses were performed to confirm the validity and stability of the results. All data were pooled using random-effects models. Studies were ordered in forest plots by mean mGFR of included participants (low to high). When the SD could not be calculated from interquartile range (IQR), standard error (SE), or confidence interval (CI), the mean SD was imputed from studies in which it could be calculated. All analyses were carried out using R software, version 3.5.1 (https://www.r-project.org). A P value <0.05 was considered statistically significant.

      3. Results

      In total, 1033 references were initially identified; 383 full-text articles were reviewed after duplicates and abstract reviews, then 348 articles were excluded, of which, 126 articles were excluded due to different study types. Most of the excluded studies were not conducted to compare mGFR with simultaneous eGFR, and some of them were animal experiments. Forty-nine articles involved patients with highly selected diseases, such as AKI, critical illness, advanced liver failure, and ascites. Finally, 35 studies with 23,667 participants met the inclusion criteria (Fig. 1). The characteristics of the included studies are presented in Table S1. The quality assessment of studies was based on QUADAS-2 (Table S3). For the domain of “patient selection”, the bias was variable, because the descriptions of recruitment processes were inconsistent. For the domains of “index test”, “reference standard,” and “flow and timing”, no studies were assessed as high risk of bias, therefore the majority of included studies were assessed as low risk and high quality.
      Fig 1
      Fig. 1Flowchart of study selection. IDMS, isotope dilution mass spectrometry.

      3.1 Difference in bias among CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys for eGFR

      A total of 32 studies (21,899 participants) reported mean bias; 16 studies reported mGFR <60 ml/min/1.73m2, and 17 studies reported mGFR ≥60 ml/min/1.73m2 (Fig. 2). The difference in the bias of eGFR was lowest using CKD-EPIcys, which was 4.84 mL/min/1.73 m2 lower than using CKD-EPIcrea, and 1.50 mL/min/1.73 m2 lower than using CKD-EPIcrea/cys. In the subgroup of mean mGFR <60 mL/min/1.73 m2, CKD-EPIcys eGFR had significantly lower bias than both CKD-EPIcrea eGFR and CKD-EPIcrea/cys eGFR, the difference in the bias of CKD-EPIcys eGFR was 6.56 mL/min/1.73 m2 lower than CKD-EPIcrea eGFR, and 2.17 mL/min/1.73 m2 lower than CKD-EPIcrea/cys eGFR. However, there was no significant difference in the subgroup of mean mGFR ≥60 mL/min/1.73 m2. Similarly, the difference in the bias of CKD-EPIcys eGFR was lower than CKD-EPIcrea eGFR and CKD-EPIcrea/cys eGFR in the subgroup of non-elderly (<70 years), and had no significant difference in the subgroup of elderly (≥70 years) (Fig. S1). In addition, CKD-EPIcys was the best in Asian populations but was not significantly better than CKD-EPIcrea/cys in non-Asian populations (Fig. S2).
      Fig 2
      Fig. 2Difference in mean bias from CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, and pooled estimate (diamond) stratified into subgroups of low and high mGFR using random-effects models. CKD-EPIcys had the lowest bias, and CKD-EPIcrea/cys had lower bias than CKD-EPIcrea. CI, confidence interval.To compare the difference in absolute mean of eGFR using these three equations, we analyzed 28 studies with 17,377 participants, of which, 15 studies reported mean mGFR <60 mL/min/1.73 m2, 14 studies reported mean mGFR ≥60 mL/min/1.73 m2. The absolute mean value of CKD-EPIcys eGFR was lower than CKD-EPIcrea and CKD-EPIcrea/cys, regardless of mean mGFR <60 mL/min/1.73 m2 (4.69 mL/min/1.73 m2 lower than CKD-EPIcrea, and 2.21 mL/min/1.73 m2 lower than CKD-EPIcrea/cys) or ≥60 mL/min/1.73 m2 (5.13 mL/min/1.73 m2 higher than CKD-EPIcys, and 3.85 mL/min/1.73 m2 higher than CKD-EPIcrea/cys). Meanwhile, the absolute mean value was not significant difference between CKD-EPIcrea eGFR and CKD-EPIcrea/cys eGFR in general and subgroup analyses (Fig. S3).

      3.2 Difference in accuracy among CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys for eGFR

      Accuracy estimates (P30) were reported in 22 studies of 17,408 participants, of which, 11 studies reported mean mGFR <60 mL/min/1.73 m2, and 12 studies reported mean mGFR > 60 mL/min/1.73 m2. Fig. 3 showed that mean P30 was best for CKD-EPIcrea/cys, in that P30 of CKD-EPIcrea/cys eGFR was 7.50% higher than CKD-EPIcrea eGFR and 3.21% higher than CKD-EPIcys eGFR. Further, the P30 of CKD-EPIcys was 4.48% higher than CKD-EPIcrea. Similar trends were observed in subgroups of low mGFR (<60ml/min/1.73m2) and high mGFR (≥60ml/min/1.73m2) (Fig. 3), non-elderly (<70 years) and elderly (≥70 years) (Fig. S4), as well as Asian and non-Asian populations (Fig. S5). However, significant superiority was only detected in comparing CKD-EPIcrea/cys eGFR and CKD-EPIcrea eGFR. Moreover, the mean P30 was 64%, 73%, and 73% in CKD-EPIcrea eGFR, CKD-EPIcys eGFR, and CKD-EPIcrea/cys eGFR, respectively. CKD-EPIcys achieved the highest P30 (67%) in the subgroup of mean mGFR <60 mL/min/1.73 m2, while the P30 of CKD-EPIcrea/cys was highest (80%) for mean mGFR ≥60 mL/min/1.73 m2 (Fig. 4).
      Fig 3
      Fig. 3Difference in mean accuracy (P30) from CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, stratified into subgroups of low and high mGFR using random-effects models. CKD-EPIcrea/cys was the most accurate with the highest P30. P30, the proportion of eGFR results within 30% of mGFR result; CI, confidence interval.
      Fig 4
      Fig. 4Mean accuracy (P30) of eGFR using CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, stratified into subgroups of low and high mGFR using random-effects models. The mean P30 was 64%, 73%, and 73% in CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, respectively, and CKD-EPIcrea/cys achieved highest P30 (80%) in the subgroup of mean mGFR ≥60 mL/min/1.73 m2. P30, the proportion of eGFR results within 30% of mGFR result; CI, confidence interval.

      3.3 Correlation of mGFR and eGFR using CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys

      Four studies with 2,107 participants reported the Pearson correlation coefficient (R, which is a measure of the strength of the correlation) between mGFR and eGFR using CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys. Fisher's z-transformed R was applied to narrow the bias from the small size of included studies and participants in the analysis [

      Cooper H, Hedges LV, Valentine JC. The handbook of research synthesis and meta-analysis: Russell Sage Foundation; 2009.

      ]. Fig. 5 showed that both Pearson R and Fisher's z-transformed R were the highest between mGFR and CKD-EPIcrea/cys eGFR, which suggested that the correlation between mGFR and eGFR using CKD-EPIcrea/cys was the best of the three equations. The eGFR using CKD-EPIcrea and CKD-EPIcys had similar Pearson R and Fisher's z-transformed R with mGFR.
      Fig 5
      Fig. 5Pearson correlation coefficient (R) and Fisher's z-transformed R between mGFR and eGFR using CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys. CKD-EPIcrea/cys performed better than CKD-EPIcrea and CKD-EPIcys, with the highest Pearson R and Fisher's z-transformed R. COR, correlation coefficient; ZCOR, Fisher's z-transformed correlation coefficient; CI, confidence interval.

      4. Discussion

      Estimation of GFR is necessary for an accurate evaluation of renal function in detecting and staging CKD, and may also significantly impact appropriate drug dosing, disease intervention, and risk stratification in clinical practice [
      • Members KB.
      KDIGO 2012 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease.
      ]. CKD-EPI equations have been proved to perform better than MDRD Study equation across diverse populations, not only in bias and accuracy but also in risk prediction of GFR decline and all-cause mortality [
      • McFadden EC
      • Hirst JA
      • Verbakel JY
      • McLellan JH
      • Hobbs FDR
      • Stevens RJ
      • et al.
      Systematic Review and Metaanalysis Comparing the Bias and Accuracy of the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration Equations in Community-Based Populations.
      ,
      • Matsushita K
      • Mahmoodi BK
      • Woodward M
      • Emberson JR
      • Jafar TH
      • Jee SH
      • et al.
      Comparison of Risk Prediction Using the CKD-EPI Equation and the MDRD Study Equation for Estimated Glomerular Filtration Rate.
      ]. In populations prevalent in primary health care, we found that eGFR using CKD-EPIcys had less bias than using CKD-EPIcrea and CKD-EPIcrea/cys, and estimates from CKD-EPIcrea/cys were more accurate than those from CKD-EPIcrea. CKD-EPIcys performed best regarding bias among the three equations, particularly in subgroup analyses of low mGFR (<60 mL/min/1.73 m2), non-elderly (<70 years), and Asia populations.
      The reduced accuracy of eGFR could be caused by high variability even when the bias is low. The KDIGO guidelines have recommended that the target of P30 should be greater than 90%. However, most GFR estimating equations have not achieved this requirement [
      • Lamb EJ
      • Stevens PE
      Estimating and measuring glomerular filtration rate: methods of measurement and markers for estimation.
      ]. Our results showed that CKD-EPIcrea/cys equation did improve accuracy compared with CKD-EPIcys and CKD-EPIcrea across a broad range of populations, and the significant difference could be detected in subgroups of high mGFR (≥60 mL/min/1.73 m2) and non-Asia populations, meanwhile, CKD-EPIcrea/cys gave the highest P30 in populations with high mGFR (≥60 mL/min/1.73 m2). Moreover, the eGFR using CKD-EPIcrea/cys had a stronger correlation with mGFR than using CKD-EPIcys and CKD-EPIcrea. However, none of the three equations achieved the goal of P30 >90%. Therefore, further improvement in estimating GFR is still needed.
      As above mentioned, tubular secretion of creatinine and muscle mass are different across diverse ethnicities. Cystatin C is less affected by ethnic variation than creatinine [
      • Delanaye P
      • Mariat C.
      The applicability of eGFR equations to different populations.
      ]. Ethnic coefficients for black and non-black have already been added in the CKD-EPI equations. Creatinine-based eGFR equations may also require an ethnicity coefficient in Asian populations, while CKD-EPIcrea/cys could obviate this need [
      • Teo BW
      • Zhang LX
      • Guh JY
      • Tang SCW
      • Jha V
      • Kang DH
      • et al.
      Glomerular Filtration Rates in Asians.
      ]. Therefore, we further compared the difference in mean bias and accuracy from CKD-EPIcrea, CKD-EPIcys, and CKD-EPIcrea/cys, stratified studies into subgroups of Asian and non-Asian populations. Consistent with previous reports, our findings demonstrated that eGFR using cystatin C based equations had less bias and more accuracy in both Asian and non-Asian populations, which could improve the precision of CKD classification, and reduce the need for ethnic adjustment.
      There are limitations to the current study. First, the comparisons are limited by the ways in which the studies were conducted and the methods of data reporting in the included studies. The participants in the majority of studies were not recruited from community settings, therefore the results could have high clinical and statistical heterogeneity. To accommodate this, sensitivity analyses were conducted by removing one of the studies in order to estimate whether the results could be markedly affected by a single study. The sensitivity analyses confirmed the validity and stability of all the results. Second, the data on other racial groups, such as Hispanic Whites and non-Hispanic Whites, were not available from our included studies, thus, further subgroup analyses of races could not be performed. Third, reference methods for mGFR were different (Table S1), and their effect on the values of measured GFR was unknown. Additionally, the number of studies including R was small, and the findings on the R analysis would need to be confirmed with more studies in the future.

      5. Conclusion

      Cystatin C based CKD-EPI equations gave less bias and more accurate estimates of mGFR. Among CKD-EPIcrea, CKD-EPIcys and CKD-EPIcrea/cys, CKD-EPIcys had the least bias, and CKD-EPIcrea/cys achieved the best accuracy. Further research on the equations of eGFR is still needed to achieve more precise GFR estimation, more variables and coefficients could be added in CKD-EPI equations, and novel equations could be established in the future.

      6. Funding

      This study was supported by the National Natural Science Foundation of China, 81600540 to LS, Natural Science Foundation of Jiangsu Province, BK20150224 to LS, Science and Technology Foundation of Xuzhou City, KC16SL119 and KC17175 to LS, National Institutes of Health/National Center for Advancing Translational Sciences, UL1TR001881 to SBN, National Institutes of Health/National Institute of Minority Health and Health Disparities, U54MD008149 to SBN and SKS.

      Declaration of Competing Interest

      The authors have declared no conflict of interest

      References

        • Stevens LA
        • Padala S
        • Levey AS
        Advances in glomerular filtration rate-estimating equations.
        Curr Opin Nephrol Hypertens. 2010; 19: 298-307
        • Lewis JR
        • Lim W
        • Dhaliwal SS
        • Zhu K
        • Lim EM
        • Thompson PL
        • et al.
        Estimated glomerular filtration rate as an independent predictor of atherosclerotic vascular disease in older women.
        BMC Nephrol. 2012; 13: 58
        • MacKinnon HJ
        • Wilkinson TJ
        • Clarke AL
        • Gould DW
        • O'Sullivan TF
        • Xenophontos S
        • et al.
        The association of physical function and physical activity with all-cause mortality and adverse clinical outcomes in nondialysis chronic kidney disease: a systematic review.
        Ther Adv Chronic Dis. 2018; 9: 209-226
        • Machado JD
        • Camargo EG
        • Boff R
        • Rodrigues LD
        • Camargo JL
        • Soares AA
        • et al.
        Combined creatinine-cystatin C CKD-EPI equation significantly underestimates measured glomerular filtration rate in people with type 2 diabetes mellitus.
        Clin Biochem. 2018; 53: 43-48
        • Matsushita K
        • Selvin E
        • Bash LD
        • Astor BC
        • Coresh J
        Risk Implications of the New CKD Epidemiology Collaboration (CKD-EPI) Equation Compared With the MDRD Study Equation for Estimated GFR: The Atherosclerosis Risk in Communities (ARIC) Study.
        Am J Kidney Dis. 2010; 55: 648-659
        • Levey AS
        • Stevens LA
        • Schmid CH
        • Zhang YP
        • Castro AF
        • Feldman HI
        • et al.
        A New Equation to Estimate Glomerular Filtration Rate.
        Ann Intern Med. 2009; 150: 604-612
        • McFadden EC
        • Hirst JA
        • Verbakel JY
        • McLellan JH
        • Hobbs FDR
        • Stevens RJ
        • et al.
        Systematic Review and Metaanalysis Comparing the Bias and Accuracy of the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration Equations in Community-Based Populations.
        Clin Chem. 2018; 64: 475-485
        • Matsushita K
        • Mahmoodi BK
        • Woodward M
        • Emberson JR
        • Jafar TH
        • Jee SH
        • et al.
        Comparison of Risk Prediction Using the CKD-EPI Equation and the MDRD Study Equation for Estimated Glomerular Filtration Rate.
        JAMA. 2012; 307: 1941-1951
        • Shemesh O
        • Golbetz H
        • Kriss JP
        • Myers BD
        Limitations of creatinine as a filtration marker in glomerulopathic patients.
        Kidney Int. 1985; 28: 830-838
        • Delanaye P
        • Mariat C.
        The applicability of eGFR equations to different populations.
        Nat Rev Nephrol. 2013; 9: 513-522
        • Abrahamson M
        • Olafsson I
        • Palsdottir A
        • Ulvsback M
        • Lundwall A
        • Jensson O
        • et al.
        Structure and expression of the human cystatin C gene.
        Biochem J. 1990; 268: 287-294
        • Stevens LA
        • Schmid CH
        • Greene T
        • Li L
        • Beck GJ
        • Joffe MM
        • et al.
        Factors other than glomerular filtration rate affect serum cystatin C levels.
        Kidney Int. 2009; 75: 652-660
        • Ying X
        • Jiang Y
        • Qin G
        • Qian Y
        • Shen X
        • Jiang Z
        • et al.
        Association of body mass index, waist circumference, and metabolic syndrome with serum cystatin C in a Chinese population.
        Medicine (Baltimore). 2017; 96: e6289
        • Salminen M
        • Laine K
        • Korhonen P
        • Wasen E
        • Vahlberg T
        • Isoaho R
        • et al.
        Biomarkers of kidney function and prediction of death from cardiovascular and other causes in the elderly: A 9-year follow-up study.
        European Journal of Internal Medicine. 2016; 33: 98-101
        • van Deventer HE
        • Paiker JE
        • Katz IJ
        • George JA
        A comparison of cystatin C- and creatinine-based prediction equations for the estimation of glomerular filtration rate in black South Africans.
        Nephrol Dial Transplant. 2011; 26: 1553-1558
        • Schottker B
        • Herder C
        • Muller H
        • Brenner H
        • Rothenbacher D
        Clinical Utility of Creatinine- and Cystatin C-Based Definition of Renal Function for Risk Prediction of Primary Cardiovascular Events in Patients With Diabetes.
        Diabetes Care. 2012; 35: 879-886
        • Abu-Assi E
        • Raposeiras-Roubin S
        • Riveiro-Cruz A
        • Rodriguez-Girondo M
        • Gonzalez-Cambeiro C
        • Alvarez-Alvarez B
        • et al.
        Creatinine-or cystatin C-based equations to estimate glomerular filtration rate in acute myocardial infarction: A disparity in estimating renal function and in mortality risk prediction.
        Int J Cardiol. 2013; 168: 4300-4301
        • Helmersson-Karlqvist J
        • Arnlov J
        • Larsson A
        Cystatin C-based glomerular filtration rate associates more closely with mortality than creatinine-based or combined glomerular filtration rate equations in unselected patients.
        Eur J Prev Cardiol. 2016; 23: 1649-1657
        • Chi XH
        • Li GP
        • Wang QS
        • Qi YS
        • Huang K
        • Zhang Q
        • et al.
        CKD-EPI creatinine-cystatin C glomerular filtration rate estimation equation seems more suitable for Chinese patients with chronic kidney disease than other equations.
        BMC Nephrol. 2017; 18: 226
        • Pan Y
        • Jiang S
        • Qiu DD
        • Shi JS
        • Zhou ML
        • An Y
        • et al.
        Comparing the GFR estimation equations using both creatinine and cystatin c to predict the long-term renal outcome in type 2 diabetic nephropathy patients.
        Journal of Diabetes and Its Complications. 2016; 30: 1478-1487
        • Masson I
        • Maillard N
        • Tack I
        • Thibaudin L
        • Dubourg L
        • Delanaye P
        • et al.
        GFR Estimation Using Standardized Cystatin C in Kidney Transplant Recipients.
        Am J Kidney Dis. 2013; 61: 279-284
        • Meeusen JW
        • Rule AD
        • Voskoboev N
        • Baumann NA
        • Lieske JC
        Performance of Cystatin C- and Creatinine-Based Estimated Glomerular Filtration Rate Equations Depends on Patient Characteristics.
        Clin Chem. 2015; 61: 1265-1272
        • Fan L
        • Levey AS
        • Gudnason V
        • Eiriksdottir G
        • Andresdottir MB
        • Gudmundsdottir H
        • et al.
        Comparing GFR Estimating Equations Using Cystatin C and Creatinine in Elderly Individuals.
        J Am Soc Nephrol. 2015; 26: 1982-1989
        • Teo BW
        • Zhang LX
        • Guh JY
        • Tang SCW
        • Jha V
        • Kang DH
        • et al.
        Glomerular Filtration Rates in Asians.
        Adv Chronic Kidney Dis. 2018; 25: 41-48
        • Members KB.
        KDIGO 2012 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease.
        Kidney Int Suppl. 2013; 3: 1-150
        • Barr ELM
        • Maple-Brown LJ
        • Barzi F
        • Hughes JT
        • Jerums G
        • Ekinci EI
        • et al.
        Comparison of creatinine and cystatin C based eGFR in the estimation of glomerular filtration rate in Indigenous Australians: The eGFR Study.
        Clin Biochem. 2017; 50: 301-308
        • Shardlow A
        • McIntyre NJ
        • Fraser SDS
        • Roderick P
        • Raftery J
        • Fluck RJ
        • et al.
        The clinical utility and cost impact of cystatin C measurement in the diagnosis and management of chronic kidney disease: A primary care cohort study.
        PLoS Med. 2017; 14e1002400
        • Lim WH
        • Lewis JR
        • Wong G
        • Turner RM
        • Lim EM
        • Thompson PL
        • et al.
        Comparison of estimated glomerular filtration rate by the chronic kidney disease epidemiology collaboration (CKD-EPI) equations with and without Cystatin C for predicting clinical outcomes in elderly women.
        PLoS One. 2014; 9e106734
        • Iliadis F
        • Didangelos T
        • Ntemka A
        • Makedou A
        • Moralidis E
        • Gotzamani-Psarakou A
        • et al.
        Glomerular filtration rate estimation in patients with type 2 diabetes: creatinine- or cystatin C-based equations?.
        Diabetologia. 2011; 54: 2987-2994
        • Whiting PF
        • Rutjes AWS
        • Westwood ME
        • Mallett S
        • Deeks JJ
        • Reitsma JB
        • et al.
        QUADAS-2: A Revised Tool for the Quality Assessment of Diagnostic Accuracy Studies.
        Ann Intern Med. 2011; 155: 529-536
        • Tampi RR
        • Tampi DJ
        • Balachandran S
        • Srinivasan S
        Antipsychotic use in dementia: a systematic review of benefits and risks from meta-analyses.
        Ther Adv Chronic Dis. 2016; 7: 229-245
        • Higgins JPT
        • Thompson SG.
        Quantifying heterogeneity in a meta-analysis.
        Stat Med. 2002; 21: 1539-1558
      1. Cooper H, Hedges LV, Valentine JC. The handbook of research synthesis and meta-analysis: Russell Sage Foundation; 2009.

        • Lamb EJ
        • Stevens PE
        Estimating and measuring glomerular filtration rate: methods of measurement and markers for estimation.
        Curr Opin Nephrol Hypertens. 2014; 23: 258-266