Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

PERSpECTIvES

Corrected: Author Correction

continued to be published, notably the Mayo


OPINION
Clinic Quadratic (MCQ) equation in 2004
(ref.3), the CKD Epidemiology Collaboration

Estimated GFR: time for a critical formula (CKD-​EPI) in 2009 (ref.4), and
an eGFR equation based on normalized
appraisal values of serum creatinine in 2016 (ref.14).
In 1985, cystatin C was first proposed as a
marker of renal function15. Since then,
Esteban Porrini, Piero Ruggenenti, Sergio Luis-​Lima, Fabiola Carrara, more than 15 eGFR formulae based on
Alejandro Jiménez, Aiko P.J. de Vries, Armando Torres, Flavio Gaspari cystatin C levels have been published
and Giuseppe Remuzzi (Supplementary Table 1).
In addition, several eGFR formulae have
Abstract | Since 1957 , over 70 equations based on creatinine and/or cystatin C combined creatinine and cystatin C16–20,
levels have been developed to estimate glomerular filtration rate (GFR). However, used multiple equations21,22, or included
whether these equations accurately reflect renal function is debated. In this biomarkers (such as β-​trace protein or
β2-microglobulin23) to increase the accuracy
Perspectives article, we discuss >70 studies that compared estimated GFR (eGFR) and precision of GFR estimation. The most
with measured GFR (mGFR), involving ~40,000 renal transplant recipients and recent equations available were published
patients with chronic kidney disease (CKD), type 2 diabetes mellitus or polycystic in 2018: the full age spectrum (FAS)
kidney disease. Their results show that eGFR often differed from mGFR by ±30% or formulae are based on either creatinine or
more, that eGFR values incorrectly staged CKD in 30–60% of patients, and that cystatin C levels, or both21. Finally, some
eGFR equations were developed in specific
eGFR and mGFR gave different rates of GFR decline. Errors were unpredictable,
populations, such as renal transplant
and comparable for equations based on creatinine and/or cystatin C. We argue, recipients24, patients with type 2 diabetes
therefore, that the persistence of these errors (despite intensive research) suggests mellitus (T2DM)25,26, and specific ethnic
that the problem lies with using creatinine and/or cystatin C as markers of renal groups, including African American18,27,
function, rather than with the mathematical methods used for GFR estimation. Chinese20,28–33, Japanese34 or Thai35 patients.
Most modern formulae (algorithms)
used to calculate eGFR (such as the MCQ
A reliable determination of renal function (Supplementary Table 1). These equations equation) are based on serum creatinine
is fundamental in many clinical situations, have increased in mathematical complexity and/or cystatin C levels, age, weight and sex.
including the evaluation of patients with over time, from simple ratios between a Typically, these formulae were developed
renal diseases, staging of chronic kidney constant and creatinine level2, to expressions in large populations covering the whole
disease (CKD), predicting the risk of disease involving exponentials or logarithms spectrum of renal function3,4,14,17,18,36 and,
progression, monitoring renal function combined with conditional parameters therefore, are considered more reliable
over time, determining the need for dialysis and corrective factors for sex, ethnicity or than early equations such as the MDRD,
therapy, screening living kidney donors and renal function3–6. which was derived from studies of patients
enabling dose adjustment of nephrotoxic The evolution of GFR estimation with CKD. However, the reliability of these
agents (drugs or contrast media) in patients methods can be considered from an equations remains a matter of debate.
with impaired renal function. In addition, historical perspective. The first formula Several studies performed in different
the accurate and precise evaluation based on serum levels of creatinine used clinical settings have criticized the
of glomerular filtration rate (GFR) is to calculate estimated GFR (eGFR) was performance of eGFR in reflecting real renal
particularly important in clinical studies published in 1957 and was mainly intended function37–44. Moreover, without detracting
that include renal function as a primary to facilitate dose adjustments of medications from the valuable work dedicated to
outcome measure. cleared by the kidneys, such as digitalis and developing these equations, the observation
Serum creatinine level is the most widely antineoplastic drugs7. Other equations based that new eGFR formulae continue to be
used marker of renal function in routine on serum creatinine levels followed8–11, and published implies that a definitive equation
clinical practice1. Creatinine clearance over in 1976 Cockcroft and Gault published an has not yet been developed. In fact, the
a 24 h period (24 h CrCl) is considered a eGFR formula that became very popular performance of modern eGFR formulae
surrogate marker of GFR. However, among physicians12. The eGFR formula (when assessed using appropriate statistical
measurement of 24 h CrCl is burdensome developed in the Modification of Diet in methods) has not improved substantially on
and frequently compromised by incomplete Renal Disease (MDRD) study, published that of the first equation described in 1957
data on urinary volume. To overcome in 2000, became the reference equation — even with the use of complex algorithms
these limitations, more than 70 equations for several years13. However, additional and cystatin C45,46, as we discuss in further
have been developed to estimate GFR creatinine-​based eGFR formulae have detail herein.


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 177
Perspectives

In this Perspectives article, we critically creatinine, leading to underestimation creatinine level and GFR can be observed
appraise the results of over 70 published of GFR. irrespective of whether creatinine is
studies that compared the performance of Furthermore, secretion and reabsorption measured with the IDMS (Fig. 1) or Jaffé
any eGFR formulae against any measured of creatinine by renal tubular cells and (Fig. 2) methods.
GFR (mGFR) reference method in renal extrarenal clearance also influence serum
transplant recipients and patients with creatinine levels. Under normal conditions, Cystatin C
CKD, T2DM or polycystic kidney disease. tubular secretion accounts for about 10% Cystatin C is an inhibitor of various cysteine
The results of studies that assessed a large of total urinary creatinine excretion, but proteinases that is encoded by CST3, a
number of patients or eGFR formulae, this percentage increases in parallel with regulatory gene expressed in nucleated
used appropriate statistics, or evaluated decreases in GFR1,47, reaching 80–100% in cells66. Cystatin C has a constant rate of
the agreement between eGFR decline and patients with advanced CKD. Thus, during production and is freely filtered across
mGFR decline, are included in Tables 1–4. the course of CKD, increased tubular glomeruli66,67. However, it is completely
Those that did not fulfil these criteria are secretion of creatinine limits the increase reabsorbed and metabolized by renal tubular
included in Supplementary Tables 2–4. We in serum creatinine level, which masks the cells, which prevents its use in calculations
argue that the persistence of substantial reduction in GFR. Data on the effect of of 24 h urinary clearance66,67. High serum
errors in all GFR estimation formulae tubular reabsorption of creatinine on serum levels of cystatin C can be observed in
developed to date suggests that the problem creatinine levels are limited56–58. In addition, individuals with morbid obesity, T2DM,
lies with serum creatinine and cystatin C, extrarenal clearance of creatinine (which hypertension and metabolic syndrome68.
which are poor markers of real renal includes recycling of creatinine to creatine Thus, changes in cystatin C levels might
function, rather than with the mathematical and degradation of creatine to products reflect the patient’s clinical status, in
approach used to develop the formulae. other than creatinine) increases with the addition to their renal function. A possible
Finally, we discuss the implications of eGFR decrease in renal function, which further explanation for the association between high
errors for clinical practice and research. contributes to the overestimation of GFR cystatin C levels and conditions related to
in patients with advanced renal disease59,60. obesity is that cystatin C is the endogenous
Markers for estimating GFR Similar limitations also apply to 24 h CrCl. inhibitor of cathepsins, which have a role
An ideal marker of GFR must not be bound These factors explain why the in adipogenesis; thus, cystatin C might be
to proteins and must be inert, freely filtered relationship between creatinine and mGFR secreted by adipocytes in a homeostatic
by glomeruli, constantly produced, and is not linear but curvilinear and why a given response to the expansion of adipose tissue68.
eliminated in the urine but not reabsorbed, value of serum creatinine can be associated As with creatinine and GFR, cystatin C
secreted or metabolized by renal cells1. The with a wide range of mGFR values (Fig. 1). has a curvilinear relationship with GFR
rate of plasma clearance from the circulation For example, a serum creatinine level of such that a given value of cystatin C reflects
must equal that of urinary excretion1, and 1.5 mg/dl can be found in patients with a range of GFR values (Fig. 1). For example,
the relationship between its concentrations GFRs between 30 ml/min and 90 ml/ cystatin C levels of 1.5 mg/l are associated
in serum with mGFR must be reciprocal min (Fig. 1), illustrating the limitations of with GFR values of 30–90 ml/min (Fig. 1).
(Supplementary Fig. 1). Unfortunately, serum creatinine levels for estimating GFR. Finally, the low coefficient of variation of
neither creatinine nor cystatin C meets Interestingly, this curvilinear relationship cystatin C could influence the estimation
these conditions47,48. between serum creatinine and GFR of GFR65. To sum up, we consider that the
(and the resulting lack of accuracy and extremely wide margins of error for both
Creatinine precision of eGFR) has been known creatinine-​based and cystatin C-​based
Serum creatinine is not bound to proteins since 1985 (ref.47) (Fig. 2). estimation of GFR make it impossible to
and is freely filtered by the glomeruli1. Errors in the assays used to measure prove that these values can be corrected by
However, synthesis of creatinine from creatinine levels might also influence the mathematical modelling.
creatine is not constant, since it is affected correlation between creatinine and GFR.
by the daily intake of protein and by muscle The Jaffé reaction (a colorimetric method Agreement between eGFR and mGFR
turnover. Meat is the major dietary source of measuring creatinine concentration in Agreement between mGFR and eGFR values
of creatine; meat intake influences the blood and serum) can also be induced by has been evaluated by various statistical
total creatine pool and, therefore, serum organic compounds in plasma other than methods, including correlation coefficients,
creatinine levels49. The major endogenous creatinine, which can lead to overestimation mean and mean percent errors, Student’s
source of creatinine is muscle, and changes of creatinine levels61–63. Modern enzymatic t-​test, linear regression, quadratic mean,
in muscle mass also influence the production assays, in which creatinine levels are and the percentage of estimations falling
of creatinine. Dietary protein deficiency standardized by calibration traceable within defined margins of error (Tables 1–4;
leads to loss of muscle mass, which reduces to isotopic dilution mass spectrometry Supplementary Tables 2–4). All these
levels of creatinine50,51. In other conditions (IDMS) reference measurement procedures, approaches have methodological limitations.
linked to reduced muscle mass (such as overcome this issue and have been proposed For example, Pearson’s correlation
anorexia, cirrhosis or sarcopenia), reduced as a reference method64. In addition, the coefficient does not take into account any
production of creatinine leads to diminished low coefficient of variation of creatinine differences in values between the groups
serum creatinine levels and overestimation measurement could influence the estimation being compared. Thus, correlation can be
of GFR. By contrast, conditions associated of GFR65. However, in our opinion, the good (r > 0.9), even when eGFR values differ
with increased muscle mass or high dietary choice of method for serum creatinine from mGFR values by twofold or more
protein intakes, such as vigorous exercise52,53, measurement does not appreciably affect (Supplementary Fig. 2). The mean and mean
chronic glucocorticoid therapy54 and the relationship between the two variables, percentage errors represent the mean of the
hyperthyroidism55, can increase the levels of as the same curvilinear correlation between difference between eGFR and mGFR for

178 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

Table 1 | Studies that evaluated estimated GFR in patients with CKD


Study Population Gold standard Formulae Statistics and error Comments Refs
Cross-​sectional studies
Selistre 9,430 white adults Inulin Cr: CKD-​EPI, • P30 overall: CKD-​EPI 80, In 20% of patients, eGFR 81

(2016) (including 1,550 renal Schwartz Schwartz 78 error was more than ±30%
transplant recipients) • P30 CKD stage 1: CKD-​EPI of mGFR
95, Schwartz 86
• P30 CKD stage 2: CKD-​EPI
82, Schwartz 86
• P30 CKD stages 3–5: CKD-​
EPI 69, Schwartz 67
• P10 overall: CKD-​EPI 38,
Schwartz 33
Fan (2014) 1,119 white patients NS Cr: CKD-​EPI P30: Overall, 87; CKD stage • In 10–15% of patients, 82

(including 594 with 1–2, 91; CKD stage 3, 87; CKD eGFR error was more
diabetes mellitus) stage 4–5, 75 than ±30% of mGFR
• Similar error in all CKD
Cy: CKD-​EPI P30: Overall, 79; CKD stage stages
1–2, 93; CKD stage 3, 82; CKD
stage 4–5, 72
Cr+Cy: CKD-​EPI P30: Overall, 91; CKD stage
1–2, 96; CKD stage 3, 91; CKD
stage 4–5, 81
Evans (2013) 2,198 white patients Iohexol Cr: MDRD, CG, • P30: MDRD 65, CG 54, CKD-​ In 35–45% of patients, 83

with CKD stage 4 and CKD-​EPI, Rule, EPI 67 , Rule 67 , Lund-​Malmö eGFR error was more than
5 (including 729 with Lund-​Malmö 76 ±30% of mGFR
diabetes mellitus) • L A: MDRD −8 to 14, CG −6
to 18, CKD-​EPI −9 to 14,
Rule −9 to 16; Lund-​Malmö
−7 to 11
Inker (2016) 3,551 black and white Iothalamate Cr: CKD-​EPI P30: CKD-​EPI 84 • In 10–25% of patients, 23

patients (including 273 eGFR error was more


with diabetes mellitus) Cy: CKD-​EPI P30: CKD-​EPI 83 than ±30% of mGFR
Cr+Cy: CKD-​EPI, P30: CKD-​EPI 89, BTP 76, • BTP and B2M
BTP, B2M, BTP-​B2M B2M 82, BTP-​B2M 85, CKD-​ comparable to old
EPI+BTP-​B2M 90 formulae
van Deventer 100 black South Cr-​EDTA
51
Cr: MDRD, CKD-​ P30: MDRD 74, CKD-​EPI 72, In 15–30% of patients, 84

(2011) African patients EPI, van Deventer van Deventer 70 eGFR error was more than
±30% of mGFR
Cy: van Deventer P30: van Deventer 84
Stevens 1,737 black patients Iothalamate Cr: MDRD (with P30: 82 if mGFR <60 ml/min/ • In 10–20% of patients, 85

(2007) corrective factor) 1.73 m2; 88 if mGFR eGFR error was more


>60 ml/min/1.73 m2 than ±30% of mGFR
• Similar P30 in the white
population (not shown)
Feng (2013) 788 Chinese patients 99m
Tc-​DTPA Cr: CG, MDRD • P30: CG 55, MDRD 44 • In 25–55% of patients, 20

• L A: CG −52 to 47 , MDRD eGFR error was more


−60 to 52 than ±30% of mGFR
• Wide L A for all formulae
Cy: CKD-​EPI, CCG, • P30: CKD-​EPI 63, CCG 65, • In 30–50% of patients,
Feng Feng 72 eGFR led to an error in
• L A: CKD-​EPI −25 to 36, CCG the classification of CKD
−34 to 32, Feng −28 to 31
CG, CCG, Feng • P30: CKD-​EPI 63, CCG 59,
Feng 74
• L A: CKD-​EPI −33 to 39, CCG
−43 to 38, Feng −29 to 28
Murata 2,324 white patients Iothalamate Cr: MDRD, P30: MDRD 75, CKD-​EPI 75 • In 25% of patients, eGFR 86

(2011) CKD-​EPI error was more than


±30% of mGFR
• In 50–60% of patients,
eGFR led to an error in
the classification of CKD
Froissart 2,095 white patients Cr-​EDTA
51
Cr: MDRD, CG • P30 overall: MDRD 87 , CG • In 15–20% of patients, 87

(2005) 79 eGFR error was more


• P30 if mGFR >60 ml/min: than ±30% of mGFR
MDRD 92, CG 88 • In 30% of patients, eGFR
• P30 if mGFR <60 ml/min: led to an error in the
MDRD 83, CG 69 classification of CKD


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 179
Perspectives

Table 1 (cont.) | Studies that evaluated estimated GFR in patients with CKD
Study Population Gold standard Formulae Statistics and error Comments Refs
Hojs (2011) 764 white patients Cr-​EDTA
51
• Cr: MDRD, • P30 CKD stage 3: MDRD 78, In 30–50% of patients, 88

CKD-​EPI, CG CG 59, CKD-​EPI 75, Hojs 78 eGFR error was more than
• Cy: Hojs • P30 CKD stage 4: MDRD 53, ±30% of mGFR
CG 44, CKD-​EPI 50, Hojs 70
• P30 CKD stage 5: MDRD 50,
CG 51, CKD-​EPI 44, Hojs 53
Longitudinal studies
Wang (2006) 1,094 black patients 125
I-​iothalamate Cr: AASK Poor concordance (r = 0.60) Poor agreement between 89

eGFR and mGFR


Xie (2008) 542 white patients 125
I-​iothalamate Cr: MDRD Poor correlation between eGFR decline slower than 90

eGFR and mGFR decline mGFR decline


(r = 0.60, r2 = 0.42). Between-​
patient difference between
mGFR slope and eGFR slope:
58% from −2 to +2 ml/min per
year ; 20% from −2 to −4 ml/
min per year or larger
Tent (2012) 65 white patients 125
I-​iothalamate Cr: CKD-​EPI, • Reduced r2 between mGFR • Median 9 mGFRs per 104

MDRD decline and eGFR decline: patient (range 9–16)


MDRD: 0.45, CKD-​EPI: 0.52 • Poor agreement between
• L A: from +3 to −3 eGFR and mGFR
Padala (2012) 3,531 white American 125
I-​iothalamate Cr: CKD-​EPI Mean eGFR decline 15% • All patients had 3–15 91

and African-​American (pooled data from faster in the AASK study and mGFRs
patients four studies), 18% slower in the CSG study. • Poor agreement between
AASK , DCCT, CSG, Comparable results for the eGFR decline and mGFR
MDRD DCCT study decline
Lee (2009) 146 white patients with Cr-​EDTA
51
Cr: MDRD, CG, Bias between formulae and Poor agreement between 92

stage 3–5 CKD MDRD6 mGFR ranged from +30 to eGFR decline and mGFR
99m
Tc-​DTPA
−30 ml/min at baseline, decline
12 and 24 months
AASK , The African American Study of Kidney Disease and Hypertension; BTP, β-​trace protein; B2M, β2-microglobulin; CCC, concordance correlation coefficient;
CCG, Chinese Collaborative Group formula; CG, Cockcroft–Gault; CKD, chronic kidney disease; CKD-​EPI, CKD Epidemiology Collaboration; Cr, creatinine; CrCl,
creatinine clearance; Cy , cystatin C; DTPA , pentetic acid; eGFR , estimated GFR; GFR , glomerular filtration rate; L A , limits of agreement (values in ml/min/1.73 m2
for cross-​sectional studies and in ml/min/1.73 m2 per year for longitudinal studies); mGFR , measured GFR; MDRD, Modification of Diet in Renal Disease; NS, not
specified; P, percentage of estimations included within an error of either ±30% (P30) or ±10% (P10) of measured GFR; TDI, total deviation index.

each patient. For example, in a hypothetical respectively) have also been used, P30 and clinical) within a short period of
group of five patients, eGFR and mGFR deserves special consideration. Notably, P30 time, and the biological variability of GFR;
values might differ by −20, −15, 20, 15 and was defined in the absence of any clinical and second, the clinical relevance of the
−5 ml/min. The mean error is −1 because or statistical rationale for considering it an proposed limits in the evaluation of GFR
numerically identical values of opposite acceptable margin of error. values and GFR decline. The reproducibility
signs (such as −20 and 20) cancel each other However, in our opinion, the limits of of mGFR is 5.0–8.0% for iohexol69,70, 4.0%
out, blunting the deviation of eGFR from ±30% are extremely wide and not acceptable for EDTA71,72 and 8.5% for iothalamate73.
mGFR (Supplementary Fig. 3). In the paired for evaluating GFR estimation formulae. For Moreover, the variability of an external
t-​test, only mean values are compared, and instance, in a patient with mGFR 60 ml/min/ quality assessment scheme for iohexol
the differences between individual data 1.73 m2, a P30 margin of error means (used by clinical laboratories to calibrate
points are, therefore, largely underestimated. that eGFR values of 42–78 ml/min/ their GFR reference measurements) is
To assess the acceptability of a formula 1.73 m2 (Fig. 3a) are considered acceptable. about 8%74. Thus, the variability of all
that estimates GFR, it is necessary to define Furthermore, for most eGFR equations, the GFR measurement methods is 4–8%. This
a priori clinically meaningful margins proportion of measurements falling within range includes both the intrinsic error of
of error, to determine the proportion of the nominally acceptable P30 limit varies the method and the biological variability
estimations that fall within these margins, from 50% to 90% (Tables 1–4).Thus, in of GFR. To ensure that eGFR formulas are
and to define how much disagreement 10–50% of measurements, the error in eGFR reliable enough to replace mGFR, estimated
is tolerable. The parameter used most values is >30%. A more rigorous evaluation values must be comparable to measured
frequently to assess the performance of GFR of eGFR than the P30 is needed. values. Of course, error is present in all
estimation equations is P30, which is defined Two issues must be considered to measurement techniques, but intrinsic
as the percentage of GFR estimations falling establish new limits of acceptability for errors must be minimized. Accordingly, we
within ±30% of mGFR values (Tables 1–4; eGFR: first, the reproducibility of the suggest that an acceptable limit of agreement
Supplementary Tables 2–4). Although reference method used to measure GFR, between eGFR and mGFR should not
other parameters (usually P10 and P20, which is influenced by both the variation in exceed ±10% of mGFR and, importantly,
representing the percentages of estimations measurements made by the same instrument that 90% of estimations should fall within
falling within ±10% and ±20% of mGFR, under similar conditions (technical this margin of error (Fig. 3b). Acceptance of

180 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

Table 2 | Studies that evaluated estimated GFR in patients with type 2 diabetes mellitus
Study Population Gold Formulae Statistics and error Comments Refs
standard
Cross-​sectional studies
Iliadis (2011) 448 white 51
Cr-​EDTA • Cr: MDRD, CKD-​EPI P30: MDRD 79, CKD-​EPI 81, • In 20–80% of patients, 46

patients • Cy: Rule, Perkins, Rule 53, Perkins 21, Arnal 45, eGFR error was more than
Tan, Arnal, Stevens, Stevens (Cy) 54, MacIsaac 46, ±30% of mGFR
Grubb, MacIsaac, Tan 79, Grubb 69, Tidman 40, • Similar bias between
Tidman, Flodin Flodin 43, Stevens (Cr+Cy) 70 Cr-​based and Cy-​based
• Cr+Cy: Stevens formulae
Maple-​Brown (2014) 224 indigenous Iohexol Cr: MDRD, CKD-​EPI, CG P30: MDRD 80, CKD-​EPI 87 , In 15–20% of patients, eGFR 108

Australian CG 71 error was more than ±30%


patients of mGFR
Rigalleau (2006) 200 white 51
Cr-​EDTA Cr: MDRD, CG, MCG L A: Bias NS In 30–50% of patients, 109

patients eGFR led to an error in the


classification of CKD
MacIsaac (2015) 199 white 99m
Tc-​DTPA Cr: MDRD, CKD-​EPI • P30: MDRD 86, CKD-​EPI 90 • In 10–15% of patients, 37

patients • L A: MDRD −34 to 31, eGFR error was more than
CKD-​EPI −30 to 27 ±30% of mGFR
• Wide L A
• In 10–30% of patients,
eGFR led to an error in the
classification of CKD
Inker (2012) 1,726 patientsa NS • Cr: CKD-​EPI P30: ~90 for all three In 10% of patients, eGFR 18

from a • Cy: CKD-​EPI equations error was more than ±30%


multi-​ethnic • Cr-​Cy: CKD-​EPI of mGFR
cohort
Li (2010) 91 Chinese 99m
Tc-​DTPA • Cr: MDRD P30: MDRD 69, Stevens 70, • In 60–80% of patients, 102

patients • Cy: Stevens, Ma, Rule, Ma 61, Rule 47 , MacIsaac 55, eGFR error was more than
MacIsaac, Perkins Perkins 44 ±30% of mGFR
• Similar bias between
Cr-​based and Cy-​based
formulae
Longitudinal studies
Gaspari (2013) 600 white Iohexol Cr: MDRD, CG, CKD-​ GFR at baseline: • 449 patients had multiple 45

patientsa EPI, Rule, Effersøe, Hull, mGFRs


• TDI: from Rule 32 to
Mawer, Gates, Ibrahim,
Jelliffe-2 93; CG 51, CKD-​EPI • Poor agreement between
Bjornsson, Davis-​ eGFR and mGFR
41; MDRD 52; Effersøe 55
Chandler, Jelliffe-1,
• CCC: from Jelliffe-2 0.21 • Hyperfiltration rarely
Jelliffe-2, Walser, detected
to Rule 0.52; CG 0.43,
Edwards-​Whyte
CKD-​EPI 0.43, MDRD 0.38, • Effersøe similar to CG and
Effersøe 0.31 CKD-​EPI
• eGFR decline slower than
GFR decline: mGFR decline
• CCC: from Effersøe −0.21
to Hull 0.36; CKD-​EPI 0.28,
MDRD 0.32, CG 0.35
Rossing (2006) 383 white 51
Cr-​EDTA Cr: MDRD, CG GFR at baseline: • Wide L A for eGFR and 38

patients eGFR decline


• L A µAlb: CG −58 to 31,
MDRD −66 to 20 • eGFR decline slower than
• L A proteinuria: CG −39 to mGFR decline
33; MDRD −47 to 25
GFR decline
(mean ±SD ml/min per year):
• µAlb mGFR: −4.1 ± 4.2, CG
−3.4 ± 3.2, MDRD −2.9 ± 2.8
• Proteinuria mGFR: −5.2
± 4.1, CG −4.6 ± 4.1,
MDRD −4.2 ± 3.8
Wood (2016) 152 white 99m
Tc-​DTPA Cr: CKD-​EPI GFR decline eGFR decline slower than 110

patients (ml/min per year): mGFR −2.6; mGFR decline


CKD-​EPI −1.6
µAlb, microalbuminuria; CCC, concordance correlation coefficient; CG, Cockcroft–Gault; CKD, chronic kidney disease; CKD-​EPI, CKD Epidemiology
Collaboration; Cr, creatinine; CrCl, creatinine clearance; Cy , cystatin C; DTPA , pentetic acid; eGFR , estimated GFR; GFR , glomerular filtration rate; L A , limits of
agreement (values in ml/min/1.73 m2 for cross-​sectional studies and in ml/min/1.73 m2 per year for longitudinal studies); mGFR , measured GFR; MCG, modified
Cockcroft–Gault; MDRD, Modification of Diet in Renal Disease; NS, not specified; P, percentage of estimations included within an error of either ±30% (P30) or
±10% (P10) of measured GFR; TDI, total deviation index. aOnly the patients with diabetes mellitus were considered.


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 181
Perspectives

Table 3 | Studies that evaluated the performance of estimated GFR in patients after renal transplantation
Study Population Gold standard Formulae Statistics and error Comments Refs
Cross-​sectional studies
Luis-​Lima 193 white Iohexol Cr: 34 formulae • CCC: Rowe 0.04, Rule 0.66, CG • For all formulae, eGFR showed 79

(2015) patients 0.70, Effersøe 0.77 , CKD-​EPI 0.79, poor agreement with mGFR
MDRD 0.82, Matsuo 0.84 • In 30–60% of patients, eGFR led
• TDI: Matsuo >49%, MDRD 53%, to an error in the classification of
Effersøe and CKD-​EPI 61%, CG CKD stage
81%, Rule 101%
Cy: 14 formulae • CCC: Perkins 0.67; Rule, Tan,
Hoek 0.90
• TDI: Hoek 34%, Perkins 83%
Cr+Cy: • CCC: Ma 0.85, Stevens 0.90,
3 formulae CKD-​EPI 0.87
• TDI: Ma 49%, Stevens 37%,
CKD-​EPI 47%
Masson 825 white • Inulin Cr: MDRD, • P30: MDRD 80, CKD-​EPI 74 • In 20–25% of patients, eGFR error 115

(2013) patients • 51Cr-​EDTA CKD-​EPI • Similar errors within each CKD was more than ±30% of mGFR
stage • Wide L A for all formulae
• L A: MDRD −23 to 26, CKD-​EPI • In 30% of patients, eGFR led to
−21 to 32 an error in the classification of
CKD
Masson 670 white Inulin Cr: CKD-​EPI P30: 75 In 15–25% of patients, eGFR error 116

(2013) patients was ±30% of mGFR


Cy: CKD-​EPI P30: 81
Cr+Cy: CKD-​EPI P30: 86
Mariat 294 white Inulin Cr: MDRD, CG, • P10: MDRD 30, CG 24, Walser 28, • In 70% of patients, eGFR error 40

(2004) patients Walser, Jelliffe, Jelliffe 32, Nankivell 23, 24 h CrCl 26 was more than ±30% of mGFR
Nankivell, 24 h • L A: CG −35 to 27 , Walser −23 to • Wide L A for all formulae
CrCl 33, MDRD −29 to 28, Jelliffe −26
to 29, Nankivell −37 to 21, 24 h
CrCl −39 to 26
Mariat 284 white Inulin Cr: CG, aMDRD, P30: CG 59, aMDRD 73, MDRD-7 75 • In 25–40% of patients, eGFR 41

(2005) patients MDRD-7 error was more than ±30% of


mGFR
• Wide L A for all formulae
• In 30–40% of patients, eGFR led
to an error in the classification
of CKD
Longitudinal studies
Bosma 798 white 125
I-​iothalamate Cr: MDRD, • P30 at baseline: MDRD 88, • 478 patients had 3 mGFRs in 42

(2005) patients Jelliffe-1, Jelliffe-2, Jelliffe-1 87 , Jelliffe-2 88, Gates 5 years
Gates, Nankivell, 87 , Hull 74, Nankivell 76, CG 76; • In 10–30% of patients, eGFR
CG, Hull, Mawer, Mawer 73, Rule 72, 24 h CrCl 66; error was more than ±30% of
Rule, 24 h CrCl P10: <40% for all equations mGFR
• mGFR decline: −1.9 ± 15 • Poor agreement between eGFR
• eGFR decline: from 0.5 ± 12.0 and mGFR decline
(Jelliffe-1) to −2.3 ± 12.0 (Hull);
P30: <24% for all equations
Gaspari 81 white Iohexol Cr: CG, Nankivell, • P10 at baseline: from • All patients had 3 mGFRs 43

(2004) patients Bjornsson, Davis-Chandler 20 to Walser 46, • In 20–55% of patients, eGFR


Edwards-​White, CG 31, MDRD 44 error was more than ±10% of
Davis-​Chandler, • mGFR decline: −3.0 ml/min/ mGFR
Hull, Jelliffe-1, 1.73 m2 per year • eGFR decline was faster than
Jelliffe-2, Mawer, • eGFR decline: from Walser −5.0 mGFR decline
Walser, MDRD to Davis-Chandler −7.4 ml/
min/1.73 m2 per year
Gera (2007) 684 white 125
I-​iothalamate Cr: MDRD, CG, • mGFR decline: −3.9 ± 13 • 360 patients had >3 mGFRs in 117

patients Rule • eGFR decline:. MDRD 2.2 ± 10, 3 years


CG: −1.1 ± 11, Rule −1.5 ± 12 • eGFR decline was slower than
(% per year) mGFR decline
• Correlation (r2) between MDRD • Poor correlation between eGFR
and mGFR 0.36; between Rule and mGFR
and mGFR 0.49
Fauvel 631 white Inulin Cr: MDRD, CKD-​ L A (% of variation in GFR decline): • All patients had 2–7 mGFRs 118

(2013) patients EPI, CG, 1/Cr from −40% to 40% for all formulae • Extreme variability between
eGFR and mGFR decline

182 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

Table 3 (cont.) | Studies that evaluated the performance of estimated GFR in patients after renal transplantation
Study Population Gold standard Formulae Statistics and error Comments Refs
Hossain 99 white 99m
Tc-​DTPA Cr: MDRD, CG • mGFR decline: −6.0 eGFR decline was faster than 119

(2010) patients • eGFR decline: MDRD −6.8, CG mGFR decline


−6.2 (% per year)
Buron (2011) 1,249 white Inulin Cr: MDRD, CKD-​ • P30 at baseline: MDRD 85, CKD-​ • In 15–35% of patients, eGFR error 120

patients EPI, CG, Nankivell EPI 81, CG 77 , Nankivell 64 was more than ±30% of mGFR
• L A: MDRD −22 to 21, CKD-​EPI • Wide L A for all formulae
−21 to 29, CG −20 to 30, Nankivell • No statistical analysis of GFR
−13 to 35 decline
aMDRD, abbreviated Modification of Diet in Renal Disease; CCC: concordance correlation coefficient; CG: Cockcroft–Gault; CKD, chronic kidney disease;
CKD-​EPI, CKD Epidemiology Collaboration; Cr, creatinine; CrCl, creatinine clearance; Cy , cystatin C; DTPA , pentetic acid; eGFR , estimated GFR; GFR , glomerular
filtration rate; L A , limits of agreement (values in ml/min/1.73 m2 for cross-​sectional studies and in ml/min/1.73 m2 per year for longitudinal studies, except where
stated); mGFR , measured GFR; MCG, modified Cockcroft–Gault; MDRD, Modification of Diet in Renal Disease; P, percentage of estimations included within an
error of either ±30% (P30) or ±10% (P10) of mGFR; TDI, total deviation index.

wider limits of agreement (>10%) would provide confidence intervals that enable P30 of 50% in patients with CKD stages 4–5,
both increase the variability between eGFR generalization of the results. However, these have been reported with this formula88,100.
and mGFR and reduce the reliability of tests have rarely been used in the study of Moreover, the P30 values of formulae
GFR estimations. eGFR45,79,80 (Tables 1–4). developed in populations that include the
Another important point to consider is whole spectrum of GFR values (such as
the large effect of errors in eGFR values Evidence of eGFR errors CKD-​EPI, variants of which are based on
on the calculation of GFR decline. This factor Chronic kidney disease. Over 30 studies, creatinine, cystatin C or both, MCQ and
explains why eGFR declines can be either involving about 30,000 participants, have Cockcroft–Gault) are similar to or even
faster or slower than mGFR declines in evaluated the performance of eGFR in worse than those of the MDRD equation
published studies, a topic that is discussed patients with CKD17,20,23,28–35,81–103 (Table 1; in patients with CKD82,83,86,88 (Table 1).
further in the following sections. So, eGFR Supplementary Table 2). In cross-​sectional The same equation can also yield discrepant
values must be comparable to mGFR values, studies, P30 ranged from 60% to 90%, and results in similar populations; for
to avoid errors in the evaluation of renal P10 averaged 40%. Thus, most eGFR values example, CKD-​EPI had P30s of 67%83
function and in its change over time. showed ±30% discordance with mGFR and 87%82 in two separate studies. Similar
Furthermore, appropriate statistics must values. Importantly, in 10–40% of patients, variability was observed for MDRD86,87 and
be used to assess the agreement between the discordance with mGFR exceeded cystatin C-based formulae17,82,83,97,100.
mGFR and eGFR. Such statistics must ±30%. The margin of error was comparable Interestingly, eGFR has shown wider
simultaneously consider precision and for creatinine-​based and cystatin C-​based margins of error in non-​white than in
accuracy, and must provide confidence equations, showing that cystatin C did white populations. However, the use of
intervals75–77. Appropriate parameters not improve the performance of eGFR ethnicity-​specific corrective factors or
include the limits of agreement78, total calculations17,20,23,82,97. For example, two population-​specific equations did not
deviation index (TDI), concordance different studies observed that CKD-​EPI improve the accuracy or precision of eGFR
correlation coefficient (CCC) and coverage equations using creatinine, cystatin C, or measurements. In African Americans with
probability75–77. The limit of agreement is a both, all exhibited similar P30 values of CKD, eGFRs calculated using the African-​
range that encompasses most differences 83–89%23,97. The limits of agreement between American Study of Hypertension and
between two measurements (the reference eGFR and mGFR were extremely wide Kidney Disease (AASK) formula exhibited
interval is defined as mean ±1.96 × 1 SD). (from −40 ml/min/1.73 m2 to 40 ml/min/ great variability compared with mGFRs25,27.
The narrower these limits are, the better the 1.73 m2)98. The only study that used statistics In patients with eGFRs <60 ml/min/1.73 m2,
agreement between the measurements. of agreement showed poor concordance an MDRD variant adapted to African
CCC combines elements of accuracy (CCC <0.75) between mGFRs and eGFRs Americans and the CKD-​EPI equation4 both
and precision; a score >0.90 (range 0–1) calculated using the MDRD, CKD-​EPI and showed similar P30 values of ~82%85. In
reflects optimal concordance between Cockcroft–Gault equations93. black South Africans, the standard MDRD
measurements. TDI is a measure that Agreement between eGFR and mGFR exhibited a higher P30 than the adapted
captures a large proportion of data within was poor for every stage of CKD81,82,88,97. formula84 (Table 1). Similarly, adapted
a boundary representing the allowable P30 values tended to be higher in CKD equations or population-​specific formulae
difference between two measurements75. stage 1–2 than in CKD stage 3–5. This error did not improve the accuracy of eGFRs
For example, a TDI of 60% represents poor markedly affected the classification of CKD in Chinese or Japanese patients20,29–35,90.
agreement (because it means that 90% of stage, resulting in about one in three patients Finally, equations based on β-​trace protein
eGFR values fall within ±60% of mGFR, being incorrectly classified on the basis of or β2-microglobulin did not outperform
which is a very large margin of error)75. eGFR into a lower or higher CKD stage than previous formulae23.
The ideal situation would be a TDI of <10%, they would be by mGFR20,86,101. The MDRD Few studies have compared the
meaning that 90% of eGFR values fall within equation is considered appropriate for use in performance of different eGFR formulae
±10% of mGFR, a much smaller margin patients with CKD, because it was developed in the monitoring of renal function over
of error (Fig. 3b). Coverage probability mostly in individuals with this condition. time, an issue of particular relevance in
values range from 0 to 1; this statistic However, P30 values of 92% and 88% patients with progressive diseases. From a
estimates whether a given TDI is less than a (obtained in patients with mGFRs above and methodological point of view, it is important
prespecified percentage75. All these statistics below 60 ml/min, respectively87), and a lower to note that GFR decline is frequently


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 183
Perspectives

Table 4 | Studies that evaluated the performance of estimated GFR in autosomal dominant polycystic kidney disease
Study Population Gold standard Formulae Statistics and error Comments Refs
Cross-​sectional studies
Orskov 101 white Cr-​EDTA
51
Cr: CG, P30: CG 90, MDRD 83, CKD-​EPI 90; • In 10–25% of patients, eGFR 136

(2009) patients MDRD, MDRD-​Cy 97 error was more than ±30%


CKD-​EPI Cy: of mGFR
MDRD-​Cy • Similar error in eGFR
whether mGFR was above
or below 60 ml/min
Longitudinal studies
Spithoven 121 white 125
I-​iothalamate Cr: MDRD, GFR at baseline: • In 1–3% of patients, eGFR 137

(2013) patients CKD-​EPI error was more than ±30%


• P30: MDRD 97%, CKD-​EPI 99%; P10:
of mGFR
MDRD 50%, CKD-​EPI 54%
• In 46–50% of patients, eGFR
• L A: MDRD −15 to 28; CKD-​EPI −15 to 22
error was more than ±10%
GFR decline (ml/min per year): of mGFR
• mGFR −2.8 ± 3.4; CKD-​EPI −3.2 ± 3.3; • Wide L A between mGFR
MDRD −2.9 ± 3.2 and eGFR decline
• L A: MDRD −5 to 5; CKD-​EPI −5 to 5
Ruggenenti 111 white Iohexol Cr: MDRD, GFR at baseline: • Poor agreement between 44

(2012) patients CKD-​EPI eGFR and mGFR decline


• L A: MDRD −30.6 to 19.5; CKD-​EPI −21.3 to 27
• mGFR decline faster than
GFR decline (ml/min per year): eGFR decline
• mGFR −8 ± 10; CKD-​EPI −5 ± 9; MDRD −4.5 ± 10
• L A: MDRD −22 to 29; CKD-​EPI −21 to 28
CG, Cockcroft–Gault; CKD-​EPI, Chronic Kidney Disease–Epidemiology Collaboration; Cr, creatinine; Cy , cystatin C; eGFR , estimated GFR; GFR , glomerular
filtration rate; L A , limits of agreement (values in ml/min/1.73 m2 for cross-​sectional studies and in ml/min/1.73 m2 per year for longitudinal studies); mGFR ,
measured GFR; MDRD, Modification of Diet in Renal Disease; P, percentage of estimations included within an error of either ±30% (P30) or ±10% (P10) of mGFR .

evaluated using linear regression analysis, in the design of clinical trials investigating undiagnosed by eGFR. In addition, a major
because the reliability of this calculation is the prevention of renal function loss. adverse renal event was diagnosed in 23
highly dependent on having a large number Very few studies have analysed the patients by eGFR but not detected by mGFR.
of GFR determinations. When GFR decline effect of GFR decline on important clinical Therefore, eGFR proved to be inaccurate in
is evaluated with only 2–3 data points, as it outcomes such as cardiovascular disease or the diagnosis of major adverse renal events.
generally is in published studies, the result is end-​stage renal disease (ESRD). In one By contrast, in a study that evaluated
highly dependent on the last recorded value study of patients with GFR <30 ml/min, the effect of GFR decline (both eGFR-​
of GFR. This factor has to be considered mGFR was a better predictor of all-​cause based and mGFR-​based) on the risk of
when interpreting the results of available mortality than CKD-​EPI-based eGFR: cardiovascular events, death and ESRD107,
studies, which have consistently shown an improvement in GFR of ~7 ml/min was mGFR declines over a 2-year period did
poor agreement between eGFR-​based and associated with reductions in all-​cause not outperform eGFR declines for these
mGFR-based rates of decline (Table 1). mortality of 29% when mGFR was used outcomes. Methodological limitations might
Mean eGFR decline was 15% or 18% versus 9% when eGFR was used105. This explain these intriguing and counterintuitive
faster than mGFR decline in two different disparity clearly indicates that errors in results, including the determination of GFR
studies91. In another study, MDRD-​based eGFR calculations affect predictions of the decline (both eGFR and mGFR) using only
eGFR decline was poorly correlated consequences of CKD. Moreover, this study two measurements in 24 months, which
(r = 0.60, r2 = 0.42) with mGFR decline90. used a simplified mGFR technique based on as we noted above is an unreliable method
Accordingly, in 58% of these patients, eGFR a single measurement of plasma clearance of measuring the rate of decline in renal
decline differed from mGFR decline by of iohexol, a method shown elsewhere to function. In addition, eGFR (not mGFR)
between −2 ml/min/1.73 m2 per year and be inaccurate in 25% of patients106. The values were used to define ESRD and to
+2 ml/min/1.73 m2 per year, and another use of a more accurate GFR measurement prompt the initiation of dialysis, which leads
20% of patients showed differences between method might further increase the observed to bias in the results favouring eGFR-​based
−2 ml/min/1.73 m2 per year and −4 ml/min/ difference between mGFR-​based and eGFR-​ measurements. Finally, this study was not
1.73 m2 per year or more90. Comparable based predictions of all-​cause mortality designed to evaluate the non-​inferiority
results were observed in other studies, which in patients with CKD. In a subanalysis of of either method with regard to outcome
revealed either poor concordance (r = 0.60) the AASK study89, the incidence of major prediction. Therefore, the results of this
between eGFR decline and mGFR decline89, adverse renal events (defined as either a study must be interpreted cautiously.
or wide limits of agreement (ranging from 50% reduction in GFR or ESRD) differed
+3 ml/min/1.73 m2 per year to −3 ml/min/ according to whether eGFR or mGFR was Type 2 diabetes mellitus. Several studies,
1.73 m2 per year)104. Thus, the rate of used. A total of 280 events were detected involving about 3,500 patients with T2DM,
change in eGFR is not a reliable method using mGFR, of which eGFR only detected have evaluated the performance of eGFR
for detecting changes in renal function 223 (80%). Therefore, 57 of the patients versus mGFR over a wide spectrum
over time. This factor must be taken into identified by mGFR as having experienced of renal function: normoalbuminuria,
consideration both in clinical practice and a major adverse renal event would remain microalbuminuria, overt proteinuria,

184 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

a 7.0 Fig. 1 | Relationship between serum levels of


6.5 creatinine or serum cystatin C and measured
6.0 glomerular filtration rate in patients with and
Serum creatinine (mg/dl)

5.5
without renal disease. The cause of renal impair-
5.0
4.5
ment in these two cohorts of patients was type 2
4.0 diabetes mellitus, renal transplantation, autoso-
3.5 mal dominant polycystic kidney disease or
3.0 chronic kidney disease. Measured glomerular fil-
2.5 tration rate (mGFR) was assessed by plasma clear-
2.0 ance of iohexol. a | The curvilinear relationship
1.5 between serum creatinine level and mGFR in
1.0 3,146 patients with renal impairment. b | In these
0.5
3,146 patients, serum creatinine 1.5 mg/dl is asso-
0.0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
ciated with mGFRs of 30–90 ml/min. c | The curvi-
mGFR (ml/min) linear relationship between serum cystatin C level
b 7.0 and mGFR in 597 patients with and without renal
6.5 impairment. d | In these 597 patients, serum
6.0 cystatin C 1.5 mg/l is associated with mGFRs of
Serum creatinine (mg/dl)

5.5 30–90 ml/min. Data for this figure were provided


5.0 by the Laboratory of the Aldo e Cele Daccó Center
4.5
for Rare Diseases, Mario Negri Institute, Bergamo,
4.0
3.5
Italy and the Laboratory of Renal Function,
3.0 Hospital Universitario de Canarias, University of La
2.5 Laguna, Tenerife, Spain.
2.0
1.5
In 600 patients with T2DM and
1.0
0.5 normoalbuminuria, microalbuminuria,
0.0 normal renal function or hyperfiltration45,
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 15 creatinine-​based eGFR formulae all
mGFR (ml/min) showed very poor concordance with mGFR
c 6.5
(CCC <0.52). In addition, the average TDI
6.0
5.5 was 50%, indicating that 90% of eGFRs
5.0 fell within a margin of error of ±50%
Serum cystatin C (mg/l)

4.5 compared with mGFR. In addition, 70%


4.0 of the patients with mGFR-​confirmed
3.5 hyperfiltration were not identified by any
3.0 eGFR formula. In patients with T2DM and
2.5
overt nephropathy or CKD, Cockcroft–
2.0
Gault and MDRD eGFR formulae showed
1.5
1.0
wide limits of agreement with mGFR (from
0.5
−47 ml/min to 33 ml/min)38. This error led
0.0 to the misclassification of CKD stage in
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 35% of patients, when staging was based
mGFR (ml/min) on eGFRs derived by the Cockcroft–Gault,
d 6.5
6.0
MDRD and MCQ equations109. Similar
5.5 results were observed in patients with
5.0 T2DM and CKD46,110,111. Thus, compared
Serum cystatin C (mg/l)

4.5 with mGFR, eGFR formulae have failed to


4.0 identify hyperfiltration as well as normal
3.5 kidney function or various degrees of
3.0 renal insufficiency in patients with T2DM
2.5
(reviewed elsewhere114).
2.0
Cystatin C-​based eGFR formulae have
1.5
1.0 shown no improvements in precision or
0.5 accuracy versus serum-​creatinine-based
0.0 formulae in patients with T2DM. Indeed,
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 many cystatin C-​based formulae have
mGFR (ml/min) lower P30 values than those for MDRD or
CKD-EPI formulae, although a few have P30
normal renal function, hyperfiltration from mGFRs by ±30% or more. The limits values comparable to those of creatinine-​
and CKD18,37,38,45,46,102,108–110 (Table 2; of agreement between eGFR and mGFR based formulae (Table 2). Also, the three
Supplementary Table 3). In cross-​sectional were wide and sometimes reached extreme versions of CKD-​EPI (those based on
studies, P30 was 40–90% and P10 was values; for example, from −50 ml/min to creatinine, cystatin C, or both markers) all
8–48%, indicating that most eGFRs differed 35 ml/min for the MDRD109 (Table 2). have comparable P30 values of about 90%,


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 185
Perspectives

a 8.5 estimations fell within a margin of error of


8.0
60–70% with regard to mGFR. The error for
7.5
7.0
24 h CrCl was similar to that for eGFR42,130–132.
6.5 Furthermore, CKD staging based on eGFR
Serum creatinine (mg/dl)

6.0 values led to incorrect classification in


5.5 30–60% of patients79,132–135. The use of eGFR
5.0
values for CKD staging purposes has a major
4.5
4.0 confounding effect in this population.
3.5 Studies that compared creatinine-​based
3.0 and cystatin C-​based formulae yielded
2.5 contradictory results. One report suggested
2.0
1.5
that P30 values were comparable between
1.0 the three different CKD-​EPI formulae116.
0.5 By contrast, several cystatin C-​based
0.0 equations performed better than creatinine-​
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
based formulae, demonstrated by a reduction
mGFR (ml/min)
b 8.5
in TDI from 70% to 40%79. However, this
8.0 numerical change, though substantial from
7.5 a mathematical point of view, is not sufficient
7.0 to indicate a clinically relevant improvement
6.5 in agreement between eGFR and mGFR.
Serum creatinine (mg/dl)

6.0
5.5
A TDI of 40% still represents a very wide
5.0 margin of error that cannot be accepted in
4.5 day-​to-day clinical practice or research.
4.0 Longitudinal studies have shown opposite
3.5 results: in some publications, eGFR decline
3.0
2.5
was faster, and in others it was slower, than
2.0 mGFR decline. One study reported an mGFR
1.5 decline of ~2 ml/min per year, but most
1.0 eGFR equations showed slower declines42.
0.5
Another report showed that eGFR decline
0.0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 was 30–50% slower than mGFR decline117.
mGFR (ml/min) Finally, eGFR decline measured using
creatinine-​based eGFR formulae was twofold
Fig. 2 | Relationship between serum creatinine level and measured glomerular filtration rate. faster than the mGFR decline43. Thus, eGFR
Measured glomerular filtration rate (mGFR) was assessed by plasma clearance of 99Tc-​labelled pentetic has poor reliability for monitoring kidney
acid (DTPA). a | The curvilinear relationship between serum creatinine level and mGFR in 171 patients function over time in this population.
with renal disease. b | In these 171 patients, serum creatinine 1.5 mg/dl is associated with mGFRs of
20–70 ml/min. Adapted with permission from ref.47, Elsevier.
Autosomal dominant polycystic kidney
disease. Few studies have tested the
and similar results have also been observed reduced statistical power to detect potential performance of eGFR formulae in patients
in other studies111 (Table 2). So, as observed benefits of treatments designed to prevent with autosomal dominant polycystic
for patients with CKD, the use of cystatin C renal function loss in patients with T2DM. kidney disease (ADPKD)44,136,137 (Table 4).
did not improve the accuracy of eGFR Two reports showed that most eGFRs
formulae in patients with T2DM. Renal transplantation. The performance of (calculated using MDRD, Cockcroft–Gault
Decline in renal function was slower eGFR formulae in renal transplant recipients and CKD-EPI equations) fell within a
when determined by eGFR than by mGFR. has been evaluated in more than 30 studies, margin of error of ±30%136,137. In longitudinal
For example, in one longitudinal study including ~6,000 patients40–43,79,115–135 (Table 3; studies, eGFR decline showed wide limits of
in patients with T2DM, the mean mGFR Supplementary Table 4). In cross-​sectional agreement with mGFR decline137. Moreover,
decline was −3.37 ml/min/1.73 m2 per year, studies, P30 was 30–90% and P10 was eGFR decline was 50% slower than mGFR
whereas the mean eGFR decline was 50–60% 8–48%. Thus, most eGFR formulae erred decline: −8 ml/min per year versus −4 ml/min
slower (from −1 ml/min/1.73 m2 per year to by ±30% compared to mGFR, and errors per year44. This difference indicates that eGFR
−1.50 ml/min/1.73 m2 per year) with various were frequently even greater. The limits formulae fail to detect patients who progress
creatinine-​based formulae45. Similar results of agreement between eGFR and mGFR rapidly towards CKD in this population.
were observed for the CKD-​EPI110, MDRD38 were wide, from −30 ml/min to 30 ml/min
and Cockcroft–Gault38 equations, in patients on average40,115,122,131. This error was similar Flaws of eGFR in clinical trials
with normoalbuminuria, microalbuminuria for creatinine-​based and cystatin C-​based In clinical trials of interventions designed
or overt proteinuria38,45. The fact that eGFR equations116,127,131–135. In 193 patients evaluated to slow the rate of decline in renal function
decline is slower than mGFR decline limits using 51 eGFR formulae, all equations over time, the use of a reliable tool to
the use of eGFR formulae in clinical practice. showed poor concordance with mGFR determine GFR is imperative. The fact
Moreover, clinical trials that use eGFR to (CCC 0.04–0.90)79. The average TDI was that chronic nephropathies are slowly
measure renal function decline will have a about 60–70%, indicating that 90% of progressive must be taken into account.

186 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

a b Several different methods have been used


mGFR mGFR for measuring GFR: the clearance of inulin,
ml/min 30 36 42 60 78 84 90 ml/min 42 48 54 60 66 72 78 51
Cr-​EDTA,99Tc-​pentetic acid (DTPA), iohexol
90%
or iothalamate143. All these methods have been
P30 (%) –50 –40 –30 30 40 50 P10 (%) –30 –20 –10 10 20 30
criticized as impractical, difficult, burdensome
and time-consuming, which has limited and
Fig. 3 | The clinical relevance of margins of error in parameters used to assess agreement
even discouraged their use in clinics and
between estimated and measured glomerular filtration rates. a | P30 is defined as the percentage research. However, these limitations mostly
of estimated glomerular filtration rates (eGFRs) that lie within ±30% of the measured glomerular fil- relate to measuring the clearance of inulin,
tration rate (mGFR). In a patient with mGFR 60 ml/min, eGFRs of 42–78 ml/min are within the P30 which requires continuous marker infusion
limit and are therefore considered to be in good agreement with mGFR . b | P10 is defined as the per- and urine collection. In a review of studies
centage of eGFR values that lie within ±10% of mGFR . In a patient with mGFR 60 ml/min, eGFRs of that compared clearance of 51Cr-​EDTA,
54–66 ml/min fall within the P10 limit and are considered to be in good agreement with mGFR . To be 99
Tc-​DTPA, iohexol or iothalamate with
clinically meaningful, 90% of eGFRs should fall within this range. inulin clearance, the authors concluded that
the evidence suggests that renal clearance
of iothalamate, renal or plasma clearance of
The 2012 Kidney Disease: Improving Agency recommended that mGFR should be EDTA, and plasma clearance of iohexol are
Global Outcomes (KDIGO) clinical practice prioritized over eGFR whenever the precise reliable alternatives to inulin clearance in the
guideline on CKD defined rapid progression determination of GFR is considered essential measurement of GFR144. This conclusion is
to CKD as a GFR decline faster than (that is, in prospective studies) to avoid the especially important because inulin is now
−5 ml/min per year138. Clinical interventions great variability in eGFR140. very expensive. Additionally, 51Cr-​EDTA
that achieve a 40–50% reduction in GFR and 99Tc-​DTPA have the disadvantage of
decline (for example, from −5.0 ml/min mGFR in clinical practice requiring administration of a radiolabelled
per year to −2.0 or −2.5 ml/min per year) A new approach to the evaluation of renal substance. However, iothalamate and iohexol
are considered to have proven efficacy in function in day-​to-day clinical practice are non-​radioactive markers and clearance
preventing renal function loss in high- is needed. According to the evidence we of these agents is simple and practical to
risk patients. Although this difference discuss here, the development of further measure in the clinic69,70. In fact, measuring
in the rate of decline is clinically important, eGFR equations similar to those already the plasma clearance of iohexol is safe145 and
the change is numerically small, so very published might not be the correct strategy. only requires a single intravenous injection
precise and accurate tools are needed However, the incorporation of biomarkers of of the marker (5 ml in 2 min) with minimal
to detect it. The margin of error of all renal damage into existing equations might blood sampling over the subsequent 4–8 h
eGFR formulae is wide, and errors are improve the precision and accuracy of eGFR. depending on the patient’s renal function69,70.
unpredictable and occur randomly. Thus, Another possibility could be that mGFR
in serial eGFR values, variations of ±30% could be used in circumstances in which a Glossary
from mGFR severely affect the calculation reliable determination of renal function is
of GFR decline, especially when only a few needed. No international recommendations Accuracy
The degree of closeness of the determined value (in this
measurements are obtained. in this field are available. We speculate —
case, estimated glomerular filtration rate) to the true
Several studies in renal transplant based on the available literature — that value (measured glomerular filtration rate) under
recipients and patients with CKD, T2DM or mGFR could be implemented to evaluate prescribed conditions. Accuracy is also sometimes
ADPKD showed important discrepancies in renal function, its evolution over time and termed trueness.
renal function decline as assessed by eGFR CKD staging in patients with established
Bias
and mGFR. In fact, the rate of change in renal disease, and to prevent or avoid the The difference between an estimated value (such as
renal function using eGFR formulae proved toxic effects of drugs or procedures involving estimated glomerular filtration rate) and a true value
to be 50–100% faster or slower than with contrast media. These circumstances might (measured glomerular filtration rate), which is also
mGFR38,42–45,90,91,110,117. This variability can include, for example: the evaluation of living termed error. A statistic is biased if it is calculated in
such a way that it is systematically different from the
mask the effects of interventions that slow kidney donors before nephrectomy141,142;
parameter being estimated.
GFR decline. For example, data from the evaluation of the progression of renal
ALADIN study139 showed a 50% reduction function over time in patients with Coefficient of variation
in GFR decline versus placebo in patients established CKD20,79,86,101,109,132–135; dose The variation obtained when measurements are
with ADPKD receiving somatostatin adjustment of potentially nephrotoxic drugs repeated under the same conditions. A low value
indicates that the technique is both accurate and
(−2 ml/min per year versus −4 ml/min per (such as chemotherapeutic agents) according precise.
year; P = 0.03). However, recalculation of to renal function; evaluation of patients
GFR decline using the abbreviated MDRD with CKD scheduled to undergo procedures Precision
equation led to slower rates of decline in that require the use of contrast media; to The closeness of agreement (that is, the degree of
scatter) in a series of determinations (that is, estimated
both groups, particularly in the patients guide therapeutic decisions based on renal
glomerular filtration rate values) obtained from multiple
receiving somatostatin, which masked the function, such as the initiation of dialysis sampling of the same homogenous sample under the
difference between groups (−3.0 ml/min per in patients with advanced (stage 4–5) CKD; prescribed conditions.
year versus −2.8 ml/min; P not significant) to investigate the possibility of performing
for somatostatin versus placebo nephron-​sparing surgery in patients with Reproducibility
The precision of results compared between two
(P.R., unpublished observations). CKD scheduled to undergo nephrectomy; laboratories. Reproducibility also refers to the precision
In the light of the evidence from and evaluation of patients with CKD in of a particular method when used under the same
published studies, the European Medicines whom urine collection is not possible. operating conditions over a short period of time.


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 187
Perspectives

Importantly, plasma clearance of inulin and assess renal function in health and disease, as 15. Grubb, A., Simonsen, O., Sturfelt, G., Truedsson, L.
& Thysell, H. Serum concentration of cystatin C, factor
iohexol result in comparable mGFR values well as in clinical practice and research. D and β2-microglobulin as a measure of glomerular
when multiple blood samples are taken146. Whenever feasible, we recommend the filtration rate. Acta Med. Scand. 218, 499–503
(1985).
Moreover, iohexol-based GFR measurement direct measurement of GFR in situations 16. Ma, Y. C. et al. Improved GFR estimation by combined
is not a particularly expensive procedure; it where careful monitoring of kidney function creatinine and cystatin C measurements. Kidney Int.
72, 1535–1542 (2007).
costs about €100–200 (refs69,70). and/or its changes over time are required at 17. Stevens, L. A. et al. Estimating GFR using serum cystatin
If methodological complexity is claimed the individual level. In the era of precision C alone and in combination with serum creatinine:
a pooled analysis of 3,418 individuals with CKD. Am. J.
as an argument against the clinical use of medicine, the ongoing use of unreliable Kidney Dis. 51, 395–406 (2008).
mGFR, the most appropriate approach must eGFRs rather than mGFR, a key marker of 18. Inker, L. A. et al. Estimating glomerular filtration rate
from serum creatinine and cystatin C. N. Engl. J. Med.
be to simplify the reference method without kidney function, remains a paradox. 367, 20–29 (2012).
losing its accuracy and precision. One group 19. Schaeffner, E. S. et al. Two novel equations to estimate
Esteban Porrini1,2*, Piero Ruggenenti3,4, kidney function in persons aged 70 years or older.
simplified the measurement of plasma Sergio Luis-Lima1,2, Fabiola Carrara4, Ann. Intern. Med. 157, 471–481 (2012).
clearance of iohexol by replacing venous Alejandro Jiménez1, Aiko P. J. de Vries5, 20. Feng, J. F. et al. Multicenter study of creatinine- and/or
cystatin C-​based equations for estimation of
blood samples with dried capillary blood Armando Torres1,2, Flavio Gaspari4 and glomerular filtration rates in Chinese patients with
samples deposited on filter paper147. The Giuseppe Remuzzi3,4,6 chronic kidney disease. PLOS ONE 8, e57240 (2013).
21. Pottel, H. et al. Estimating glomerular filtration rate
dried blood sampling approach is simple and 1
Hospital Universitario de Canarias, Tenerife, Spain.
for the full age spectrum from serum creatinine and
safe, and increases patient comfort owing to 2
University of La Laguna, ITB: Instituto de Tecnologías cystatin C. Nephrol. Dial. Transplant. 32, 497–507
(2017).
the use of a painless finger prick to collect Biomédicas, Tenerife, Spain.
22. Björk, J. et al. Comparison of glomerular filtration rate
capillary blood. This sampling method also 3
Nephrology and Dialysis Unit, Azienda Socio Sanitaria estimating equations derived from creatinine and
Territoriale Papa Giovanni XXIII, Bergamo, Italy. cystatin C: validation in the age, gene/environment
reduces the number of venipunctures as well susceptibility-​Reykjavik elderly cohort. Nephrol. Dial.
as the cost of iohexol clearance measurement
4
Istituto di Ricerche Farmacologiche Mario Negri Transplant. 33, 1380–1388 (2018).
IRCCS, Centro di Ricerche Cliniche per le Malattie Rare 23. Inker, L. A. et al. GFR estimation using β-​trace protein
(since tubes, syringes and sample “Aldo e Cele Daccò”, Ranica, Bergamo, Italy. and β2-microglobulin in CKD. Am. J. Kidney Dis. 67,
centrifugation are not needed). Moreover, 40–48 (2016).
5
Leiden University Medical Centre, Department of 24. Nankivell, B. J., Gruenewald, S. M., Allen, R. D.
dried blood on filter paper is stable at room Medicine, Division of Nephrology, Leiden, Netherlands. & Chapman, J. R. Predicting glomerular filtration rate
temperature and does not need to be stored 6
Department of Biochemical and Clinical Sciences ‘
after kidney transplantation. Transplantation 59,
1683–1689 (1995).
in the freezer. Agreement between the dried L. Sacco’, University of Milan, Milan, Italy. 25. Ibrahim, H. et al. An alternative formula to the
blood sampling approach and the standard *e-​mail: esteban.l.porrini@gmail.com Cockcroft–Gault and the modification of diet in renal
diseases formulas in predicting GFR in individuals with
method for measuring plasma clearance of https://doi.org/10.1038/s41581-018-0080-9 type 1 diabetes. J. Am. Soc. Nephrol. 16, 1051–1060
iohexol was excellent (TDI 9%), suggesting Published online 5 December 2018 (2005).
26. MacIsaac, R. J. et al. Estimating glomerular filtration
that the two methods are interchangeable147. rate in diabetes: a comparison of cystatin-​C− and
Thus, the dried blood sampling approach 1. Perrone, R. D., Madias, N. E. & Levey, A. S. Serum creatinine-​based methods. Diabetologia 49,
represents a valuable simplification of a creatinine as an index of renal function: new insights
into old concepts. Clin. Chem. 38, 1933–1953 (1992).
1686–1689 (2006).
27. Lewis, J. et al. Comparison of cross-​sectional renal
reference method for measuring GFR that 2. Kaji, D., Strauss, I. & Kahn, T. Serum creatinine in function measurements in African Americans with
could help to promote the use of mGFR in patients with spinal cord injury. Mt. Sinai J. Med. 57,
160–164 (1990).
hypertensive nephrosclerosis and of primary formulas
to estimate glomerular filtration rate. Am. J. Kidney
clinical practice and research. 3. Rule, A. D. et al. Using serum creatinine to estimate Dis. 38, 744–753 (2001).
Finally, we consider that the suggestion glomerular filtration rate: accuracy in good health and 28. Xie, P., Huang, J. M., Li, Y., Liu, H. J. & Qu, Y.
in chronic kidney disease. Ann. Intern. Med. 141, The modified CKD-​EPI equation may be not more
that GFR measurement is time-​consuming 929–937 (2004). accurate than CKD-​EPI equation in determining
cannot defend the continued use of 4. Levey, A. S. et al. A new equation to estimate glomerular filtration rate in Chinese patients with chronic
glomerular filtration rate. Ann. Intern. Med. 150, kidney disease. J. Nephrol. 30, 397–402 (2017).
suboptimal and unreliable measures of 604–612 (2009). 29. Yang, M. et al. Performance of the creatinine and
kidney function. No one would today 5. Björk, J. et al. Prediction of relative glomerular cystatin C-​based equations for estimation of GFR in
filtration rate in adults: new improved equations based Chinese patients with chronic kidney disease.
challenge the role of morphological and on Swedish Caucasians and standardized plasma-​ Clin. Exp. Nephrol. 21, 236–246 (2017).
functional evaluations in the diagnosis creatinine assays. Scand. J. Clin. Lab. Invest. 67, 30. Changjie, G. et al. Evaluation of glomerular filtration
678–695 (2007). rate by different equations in Chinese elderly with
and monitoring of cardiovascular and 6. Björk, J., Grubb, A., Sterner, G. & Nyman, U. Revised chronic kidney disease. Int. Urol. Nephrol. 49,
gastroenterological diseases that require equations for estimating glomerular filtration rate 133–141 (2017).
based on the Lund–Malmö Study cohort. Scand. J. 31. Guo, X. et al. Improved glomerular filtration rate
time-​consuming and costly procedures, Clin. Lab. Invest. 71, 232–239 (2011). estimation using new equations combined with
such as angiography, colonoscopy, 7. Effersøe, P. Relationship between endogenous 24-hour standardized cystatin C and creatinine in Chinese adult
creatinine clearance and serum creatinine chronic kidney disease patients. Clin. Biochem. 47,
magnetic resonance neurography or concentration in patients with chronic renal disease. 1220–1226 (2014).
histological evaluations. Acta Med. Scand. 156, 429–434 (1957). 32. Li, J. T. et al. Relative performance of two equations
8. Edwards, K. D. & Whyte, H. M. Plasma creatinine level for estimation of glomerular filtration rate in a Chinese
and creatinine clearance as tests of renal function. population having chronic kidney disease. Chin. Med.
Conclusions Australas. Ann. Med. 8, 218–224 (1959). J. 125, 599–603 (2012).
9. Jelliffe, R. W. Estimation of creatinine clearance when 33. Liu, X. et al. Comparison of prediction equations to
Our review of comparative studies reveals urine cannot be collected. Lancet 1, 975–976 (1971). estimate glomerular filtration rate in Chinese patients
that the margin of error for all eGFR 10. Mawer, G. E., Lucas, S. B., Knowles, B. R. with chronic kidney disease. Intern. Med. J. 42,
& Stirland, R. M. Computer-​assisted prescribing of e59–e67 (2012).
formulae is unexpectedly and unacceptably kanamycin for patients with renal insufficiency. 34. Matsuo, S. et al. Revised equations for estimated GFR
wide across the whole spectrum of kidney Lancet 1, 12–15 (1972). from serum creatinine in Japan. Am. J. Kidney Dis. 53,
11. Jelliffe, R. W. Letter: creatinine clearance: bedside 982–992 (2009).
function, from hyperfiltration and normal estimate. Ann. Intern. Med. 79, 604–605 (1973). 35. Praditpornsilpa, K. et al. The need for robust
renal function to moderate and advanced 12. Cockcroft, D. W. & Gault, M. H. Prediction of validation for MDRD-​based glomerular filtration rate
creatinine clearance from serum creatinine. Nephron estimation in various CKD populations. Nephrol. Dial.
CKD. The performance of eGFR formulae 16, 31–41 (1976). Transplant. 26, 2780–2785 (2011).
(in terms of recognized metrics such as 13. Levey, A. S., Greene, T., Kusek, J. W. & Beck, G. J. 36. Grubb, A. et al. Generation of a new cystatin C-​based
A simplified equation to predict glomerular filtration rate estimating equation for glomerular filtration rate by
P30 values) has not improved over the from serum creatinine [abstract]. J. Am. Soc. Nephrol. use of 7 assays standardized to the international
past 60 years, and no improvements can 11, 115A (2000). calibrator. Clin. Chem. 60, 974–986 (2014).
14. Pottel, H. et al. An estimated glomerular filtration 37. MacIsaac, R. J. et al. The Chronic Kidney Disease-​
be attributed to the use of cystatin C-​based rate equation for the full age spectrum. Nephrol. Dial. Epidemiology Collaboration (CKD-​EPI) equation does
formulae. Thus, eGFR is an unreliable tool to Transplant. 31, 798–806 (2016). not improve the underestimation of glomerular

188 | MARCH 2019 | volume 15 www.nature.com/nrneph


Perspectives

filtration rate (GFR) in people with diabetes and 63. Peake, M. & Whiting, M. Measurement of serum 87. Froissart, M., Rossert, J., Jacquot, C., Paillard, M. &
preserved renal function. BMC Nephrol. 16, 198 creatinine — current status and future goals. Houillier, P. Predictive performance of the modification
(2015). Clin. Biochem. Rev. 27, 173–184 (2006). of diet in renal disease and Cockcroft–Gault equations
38. Rossing, P., Rossing, K., Gaede, P., Pedersen, O. 64. Myers, G. L. et al. Recommendations for improving for estimating renal function. J. Am. Soc. Nephrol. 16,
& Parving, H. H. Monitoring kidney function in type 2 serum creatinine measurement: a report from the 763–773 (2005).
diabetic patients with incipient and overt diabetic Laboratory Working Group of the National Kidney 88. Hojs, R., Bevc, S., Ekart, R., Gorenjak, M. & Puklavec, L.
nephropathy. Diabetes Care 29, 1024–1030 (2006). Disease Education Program. Clin. Chem. 52, 5–18 Kidney function estimating equations in patients with
39. Fontsere, N. et al. Are prediction equations for (2006). chronic kidney disease. Int. J. Clin. Pract. 65, 458–464
glomerular filtration rate useful for the long-​term 65. Delanaye, P., Cavalier, E., Depas, G., Chapelle, J. P. (2011).
monitoring of type 2 diabetic patients? Nephrol. Dial. & Krzesinski, J. M. New data on the intraindividual 89. Wang, X. et al. Validation of creatinine-​based
Transplant. 21, 2152–2158 (2006). variation of cystatin C. Nephron. Clin. Pract. 108, estimates of GFR when evaluating risk factors
40. Mariat, C. et al. Assessing renal graft function in c246–c248 (2008). in longitudinal studies of kidney disease. J. Am.
clinical trials: can tests predicting glomerular filtration 66. Abrahamson, M. et al. Structure and expression of the Soc. Nephrol. 17, 2900–2909 (2006).
rate substitute for a reference method? Kidney Int. human cystatin C gene. Biochem. J. 268, 287–294 90. Xie, D. et al. A comparison of change in measured and
65, 289–297 (2004). (1990). estimated glomerular filtration rate in patients with
41. Mariat, C. et al. Predicting glomerular filtration rate in 67. Grubb, A. Diagnostic value of analysis of cystatin C nondiabetic kidney disease. Clin. J. Am. Soc. Nephrol.
kidney transplantation: are the K/DOQI guidelines and protein HC in biological fluids. Clin. Nephrol. 38 3, 1332–1338 (2008).
applicable? Am. J. Transplant. 5, 2698–2703 (2005). (Suppl. 1), S20–S27 (1992). 91. Padala, S. et al. Accuracy of a GFR estimating
42. Bosma, R. J., Doorenbos, C. R., Stegeman, C. A., 68. de Vries, A. P. & Rabelink, T. J. A possible role of equation over time in people with a wide range of
van der Heide, J. J. & Navis, G. Predictive performance cystatin C in adipose tissue homeostasis may impact kidney function. Am. J. Kidney Dis. 60, 217–224
of renal function equations in renal transplant kidney function estimation in metabolic syndrome. (2012).
recipients: an analysis of patient factors in bias. Am. J. Nephrol. Dial. Transplant. 28, 1628–1630 (2013). 92. Lee, D., Levin, A., Roger, S. D. & McMahon, L. P.
Transplant. 5, 2193–2203 (2005). 69. Delanaye, P. et al. Iohexol plasma clearance for Longitudinal analysis of performance of estimated
43. Gaspari, F. et al. Performance of different prediction measuring glomerular filtration rate in clinical practice glomerular filtration rate as renal function declines in
equations for estimating renal function in kidney and research: a review. Part 1: how to measure chronic kidney disease. Nephrol. Dial. Transplant. 24,
transplantation. Am. J. Transplant. 4, 1826–1835 glomerular filtration rate with iohexol? Clin. Kidney J. 109–116 (2009).
(2004). 9, 682–699 (2016). 93. Hossain, F., Kendrick-​Jones, J., Ma, T. M.
44. Ruggenenti, P. et al. Measuring and estimating GFR 70. Delanaye, P. et al. Iohexol plasma clearance for & Marshall, M. R. The estimation of glomerular
and treatment effect in ADPKD patients: results and measuring glomerular filtration rate in clinical practice filtration rate in an Australian and New Zealand
implications of a longitudinal cohort study. PLOS ONE and research: a review. Part 2: why to measure cohort. Nephrology (Carlton) 17, 285–293 (2012).
7, e32533 (2012). glomerular filtration rate with iohexol? Clin. Kidney J. 94. Ebert, N. et al. Cystatin C standardization decreases
45. Gaspari, F. et al. The GFR and GFR decline cannot be 9, 700–704 (2016). assay variation and improves assessment of
accurately estimated in type 2 diabetics. Kidney Int. 71. Chantler, C. & Barratt, T. M. Estimation of glomerular glomerular filtration rate. Clin. Chim. Acta 456,
84, 164–173 (2013). filtration rate from plasma clearance of 51-chromium 115–121 (2016).
46. Iliadis, F. et al. Glomerular filtration rate estimation in edetic acid. Arch. Dis. Child. 47, 613–617 (1972). 95. Poggio, E. D., Wang, X., Greene, T., Van Lente, F.
patients with type 2 diabetes: creatinine- or cystatin 72. Brochner-​Mortensen, J. & Rodbro, P. Selection of & Hall, P. M. Performance of the modification of diet
C-​based equations? Diabetologia 54, 2987–2994 routine method for determination of glomerular in renal disease and Cockcroft–Gault equations in the
(2011). filtration rate in adult patients. Scand. J. Clin. Lab. estimation of GFR in health and in chronic kidney
47. Shemesh, O., Golbetz, H., Kriss, J. P. & Myers, B. D. Invest. 36, 35–43 (1976). disease. J. Am. Soc. Nephrol. 16, 459–466
Limitations of creatinine as a filtration marker in 73. Hall, P. M. & Rolin, H. Iothalamate clearance and its (2005).
glomerulopathic patients. Kidney Int. 28, 830–838 use in large-​scale clinical trials. Curr. Opin. Nephrol. 96. Bevc, S., Hojs, R., Ekart, R., Gorenjak, M. & Puklavec, L.
(1985). Hypertens. 4, 510–513 (1995). Simple cystatin C formula compared to serum
48. Laterza, O. F., Price, C. P. & Scott, M. G. Cystatin C: 74. Equalis. External quality assessment (EQA) schemes: creatinine-​based formulas for estimation of glomerular
an improved estimator of glomerular filtration rate? Iohexol (024). equalis https://www.equalis.se/en/ filtration rate in patients with mildly to moderately
Clin. Chem. 48, 699–707 (2002). products-​and-services/external-​quality-assessment-​ impaired kidney function. Kidney Blood Press. Res. 35,
49. Crim, M. C., Calloway, D. H. & Margen, S. Creatine eqa/eqa-​schemes/g-​l/iohexol-024/ (2014). 649–654 (2012).
metabolism in men: urinary creatine and creatinine 75. Lin, L., Hedayat, A. & Wu, W. Statistical Tools for 97. Bevc, S., Hojs, R., Ekart, R., Gorenjak, M. & Puklavec, L.
excretions with creatine feeding. J. Nutr. 105, Measuring Agreement (Springer Science+Business Simple cystatin C formula compared to sophisticated
428–438 (1975). Media, 2012). CKD-​EPI formulas for estimation of glomerular filtration
50. Heymsfield, S. B., Arteaga, C., McManus, C., Smith, J. 76. Lin, L. I. A concordance correlation coefficient to rate in the elderly. Ther. Apher. Dial. 15, 261–268
& Moffitt, S. Measurement of muscle mass in humans: evaluate reproducibility. Biometrics 45, 255–268 (2011).
validity of the 24-hour urinary creatinine method. (1989). 98. Hojs, R., Bevc, S., Ekart, R., Gorenjak, M. & Puklavec, L.
Am. J. Clin. Nutr. 37, 478–494 (1983). 77. Lin, L., Hedayat, A., Sinha, B. & Yang, M. Statistical Serum cystatin C as an endogenous marker of renal
51. Bleiler, R. E. & Schedl, H. P. Creatinine excretion: methods in assessing agreement: models, issues, and function in patients with mild to moderate impairment
variability and relationships to diet and body size. tools. J. Am. Stat. Assoc. 97, 257–270 (2002). of kidney function. Nephrol. Dial. Transplant. 21,
J. Lab. Clin. Med. 59, 945–955 (1962). 78. Bland, J. M. & Altman, D. G. Statistical methods for 1855–1862 (2006).
52. Irving, R. A., Noakes, T. D., Irving, G. A. assessing agreement between two methods of clinical 99. Barroso, S., Martinez, J. M., Martin, M. V., Rayo, I.
& Van Zyl-Smit, R. The immediate and delayed measurement. Lancet 1, 307–310 (1986). & Caravaca, F. Accuracy of indirect estimates of renal
effects of marathon running on renal function. J. Urol. 79. Luis-​Lima, S. et al. Estimated glomerular filtration rate function in advanced chronic renal failure patients.
136, 1176–1180 (1986). in renal transplantation: the nephrologist in the mist. Nefrologia 26, 344–350 (2006).
53. Rennie, M. J. et al. Effect of exercise on protein Transplantation 99, 2625–2633 (2015). 100. Lopes, M. B. et al. Estimation of glomerular filtration
turnover in man. Clin. Sci. 61, 627–639 (1981). 80. Ahlstrom, M. G., Kjaer, A., Gerstoft, J. & Obel, N. rate from serum creatinine and cystatin C in
54. Horber, F. F., Scheidegger, J. & Frey, F. J. Agreement between estimated and measured renal octogenarians and nonagenarians. BMC Nephrol. 14,
Overestimation of renal function in glucocorticosteroid function in an everyday clinical outpatient setting of 265 (2013).
treated patients. Eur. J. Clin. Pharmacol. 28, human immunodeficiency virus-​infected individuals. 101. Brown, M. A. et al. Inaccuracies in estimated
537–541 (1985). Nephron 136, 318–327 (2017). glomerular filtration rate in one Australian renal
55. Friedman, R. B., Anderson, R. E., Entine, S. M. 81. Selistre, L. et al. Comparison of the Schwartz and centre. Nephrology (Carlton) 16, 486–494 (2011).
& Hirshberg, S. B. Effects of diseases on clinical CKD-​EPI equations for estimating glomerular filtration 102. Li, H. X., Xu, G. B., Wang, X. J., Zhang, X. C. & Yang, J. M.
laboratory tests. Clin. Chem. 26, 1D–476D rate in children, adolescents, and adults: a Diagnostic accuracy of various glomerular filtration rates
(1980). retrospective cross-​sectional study. PLOS Med. 13, estimating equations in patients with chronic kidney
56. Miller, B. F., Leaf, A., Mamby, A. R. & Miller, Z. e1001979 (2016). disease and diabetes. Chin. Med. J. 123, 745–751
Validity of the endogenous creatinine clearance as a 82. Fan, L. et al. Glomerular filtration rate estimation (2010).
measure of glomerular filtration rate in the diseased using cystatin C alone or combined with creatinine as 103. El-​Minshawy, O. M. & El-​Bassuoni, E. Validation of
human kidney. J. Clin. Invest. 31, 309–313 (1952). a confirmatory test. Nephrol. Dial. Transplant. 29, El-​Minia equation for estimation of glomerular
57. Baldwin, D. S., Sirota, J. H. & Villarreal, H. Diurnal 1195–1203 (2014). filtration rate in different stages of chronic kidney
variations of renal function in congestive heart failure. 83. Evans, M. et al. Glomerular filtration rate-​estimating disease. Iran. J. Kidney Dis. 6, 262–268 (2012).
Proc. Soc. Exp. Biol. Med. 74, 578–581 (1950). equations for patients with advanced chronic kidney 104. Tent, H. et al. Performance of MDRD study and
58. Chesley, L. C. Renal excretion at low urine volumes disease. Nephrol. Dial. Transplant. 28, 2518–2526 CKD-EPI equations for long-​term follow-​up of
and the mechanism of oliguria. J. Clin. Invest. 17, (2013). nondiabetic patients with chronic kidney disease.
591–597 (1938). 84. van Deventer, H. E., Paiker, J. E., Katz, I. J. Nephrol. Dial. Transplant. 27 (Suppl. 3), 89–95 (2012).
59. Jones, J. D. & Burnett, P. C. Creatinine metabolism in & George, J. A. A comparison of cystatin C− and 105. Methven, S., Gasparini, A., Carrero, J. J., Caskey, F. J.
humans with decreased renal function: creatinine creatinine-based prediction equations for the estimation & Evans, M. Routinely measured iohexol glomerular
deficit. Clin. Chem. 20, 1204–1212 (1974). of glomerular filtration rate in black South Africans. filtration rate versus creatinine-​based estimated
60. Mitch, W. E., Collier, V. U. & Walser, M. Creatinine Nephrol. Dial. Transplant. 26, 1553–1558 (2011). glomerular filtration rate as predictors of mortality in
metabolism in chronic renal failure. Clin. Sci. 58, 85. Stevens, L. A. et al. Evaluation of the modification of patients with advanced chronic kidney disease: a
327–335 (1980). diet in renal disease study equation in a large diverse Swedish Chronic Kidney Disease Registry cohort study.
61. Miller, W. G. et al. Creatinine measurement: state of population. J. Am. Soc. Nephrol. 18, 2749–2757 Nephrol. Dial. Transplant. 32 (Suppl. 2), 170–179
the art in accuracy and interlaboratory harmonization. (2007). (2017).
Arch. Pathol. Lab. Med. 129, 297–304 (2005). 86. Murata, K. et al. Relative performance of the MDRD and 106. Gaspari, F. et al. Glomerular filtration rate determined
62. Panteghini, M. Enzymatic assays for creatinine: time CKD-​EPI equations for estimating glomerular filtration from a single plasma sample after intravenous iohexol
for action. Clin. Chem. Lab. Med. 46, 567–572 rate among patients with varied clinical presentations. injection: is it reliable? J. Am. Soc. Nephrol. 7,
(2008). Clin. J. Am. Soc. Nephrol. 6, 1963–1972 (2011). 2689–2693 (1996).


NAtuRe Reviews | NePhRoloGy volume 15 | MARCH 2019 | 189
Perspectives

107. Ku, E. et al. Change in measured GFR versus eGFR and 125. Maillard, N. et al. Cystatin C-​based equations in renal 142. The Renal Association. Clinical practice guideline post-​
CKD outcomes. J. Am. Soc. Nephrol. 27, 2196–2204 transplantation: moving toward a better glomerular operative care in the kidney transplant recipient. renal
(2016). filtration rate prediction? Transplantation 85, https://renal.org/wp-​content/uploads/2017/06/FINAL-​
108. Maple-​Brown, L. J. et al. Performance of formulas for 1855–1858 (2008). Post-Operative-​Care-Guideline.pdf (2017).
estimating glomerular filtration rate in indigenous 126. Poge, U. et al. MDRD equations for estimation of GFR 143. Schaeffner, E. Determining the glomerular filtration
Australians with and without type 2 diabetes: the in renal transplant recipients. Am. J. Transplant. 5, rate — an overview. J. Ren. Nutr. 27, 375–380
eGFR Study. Diabet. Med. 31, 829–838 (2014). 1306–1311 (2005). (2017).
109. Rigalleau, V. et al. A simplified Cockcroft–Gault 127. Poge, U. et al. Cystatin C-​based calculation of 144. Soveri, I. et al. Measuring GFR: a systematic review.
formula to improve the prediction of the glomerular glomerular filtration rate in kidney transplant Am. J. Kidney Dis. 64, 411–424 (2014).
filtration rate in diabetic patients. Diabetes Metab. recipients. Kidney Int. 70, 204–210 (2006). 145. Gaspari, F. et al. Safety of iohexol administration to
32, 56–62 (2006). 128. Poge, U., Gerhardt, T., Stoffel-​Wagner, B., Sauerbruch, T. measure glomerular filtration rate in different patient
110. Wood, A. J. et al. Estimating glomerular filtration rate: & Woitas, R. P. Validation of the CKD-​EPI formula in populations: a 25-year experience. Nephron 140,
performance of the CKD-​EPI equation over time in patients after renal transplantation. Nephrol. Dial. 1–8 (2018).
patients with type 2 diabetes. J. Diabetes Transplant. 26, 4104–4108 (2011). 146. Sterner, G. et al. Determining ‘true’ glomerular
Complications 30, 49–54 (2016). 129. Poggio, E. D. et al. Assessing glomerular filtration filtration rate in healthy adults using infusion of inulin
111. Beauvieux, M. C. et al. New predictive equations rate by estimation equations in kidney transplant and comparing it with values obtained using other
improve monitoring of kidney function in patients with recipients. Am. J. Transplant. 6, 100–108 clearance techniques or prediction equations.
diabetes. Diabetes Care 30, 1988–1994 (2007). (2006). Scand. J. Urol. Nephrol. 42, 278–285 (2008).
112. Silveiro, S. P. et al. Chronic Kidney Disease 130. Raju, D. L., Grover, V. K. & Shoker, A. Limitations of 147. Luis-​Lima, S. et al. Iohexol plasma clearance simplified
Epidemiology Collaboration (CKD-​EPI) equation glomerular filtration rate equations in the renal by dried blood spot testing. Nephrol. Dial. Transplant.
pronouncedly underestimates glomerular filtration transplant patient. Clin. Transplant. 19, 259–268 33, 1597–1603 (2018).
rate in type 2 diabetes. Diabetes Care 34, (2005).
2353–2355 (2011). 131. Risch, L. & Huber, A. R. Assessing glomerular filtration Acknowledgements
113. Perkins, B. A. et al. Detection of renal function decline rate in renal transplant recipients by estimates derived The authors acknowledge research support from the
in patients with diabetes and normal or elevated GFR from serum measurements of creatinine and cystatin DIABESITY working group of the ERA-​EDTA, the IMBRAIN
by serial measurements of serum cystatin C C. Clin. Chim. Acta 356, 204–211 (2005). (CIBICAN) project (FP7-RE6-POT-2012-CT2012-31637-
concentration: results of a 4-year follow-​up study. 132. White, C. et al. Estimating glomerular filtration rate IMBRAIN) funded under the 7th Framework Programme
J. Am. Soc. Nephrol. 16, 1404–1412 (2005). in kidney transplantation: a comparison between (capacities); Instituto de Salud Carlos III (ISCIII) grants
114. Bjornstad, P., Cherney, D. Z. & Maahs, D. M. Update serum creatinine and cystatin C-​based methods. PI13/00342 and PI16/01814 to E.P. and A.T., REDINREN
on estimation of kidney function in diabetic kidney J. Am. Soc. Nephrol. 16, 3763–3770 (2005). RD16/0009 and PI10/02428 grants to E.P. and A.T.; and
disease. Curr. Diab. Rep. 15, 57 (2015). 133. White, C. et al. Chronic kidney disease stage in renal funding from the IRSIN (Instituto Reina Sofia de Investigacion)
115. Masson, I. et al. MDRD versus CKD-​EPI equation to transplantation classification using cystatin C and and FEDER (both to A.T.). S.L.L. is a research fellow
estimate glomerular filtration rate in kidney transplant creatinine-​based equations. Nephrol. Dial. Transplant. supported by ISCIII grant CM15/00214 for Río Hortega spe-
recipients. Transplantation 95, 1211–1217 (2013). 22, 3013–3020 (2007). cialized health-​c are post-​t raining contracts. E.P. is a
116. Masson, I. et al. GFR estimation using standardized 134. Yeo, Y. et al. Suitability of the IDMS-​traceable MDRD researcher supported by the ISCIII Ramón y Cajal Programme
cystatin C in kidney transplant recipients. Am. J. equation method to estimate GFR in early and Fundación Caja Canarias grant DIAB05. The authors
Kidney Dis. 61, 279–284 (2013). postoperative renal transplant recipients. Nephron thank F. G. Rinne for preparation of the figures, N. N. Mena
117. Gera, M. et al. Assessment of changes in kidney Clin. Pract. 114, c108–c117 (2010). for performing the iohexol method in the Laboratory of Renal
allograft function using creatinine-​based estimates 135. Zahran, A., Qureshi, M. & Shoker, A. Comparison Function, and M. L. McLean for technical assistance.
of glomerular filtration rate. Am. J. Transplant. 7, between creatinine and cystatin C-​based GFR
880–887 (2007). equations in renal transplantation. Nephrol. Dial. Author contributions
118. Fauvel, J. P., Hadj-​Aissa, A., Buron, F., Morelon, E. Transplant. 22, 2659–2668 (2007). E.P., F.C. and F.G. researched data for the article, made sub-
& Ducher, M. Performance of estimated glomerular 136. Orskov, B. et al. Estimating glomerular filtration rate stantial contributions to discussions of its content, wrote the
filtration rates to monitor change in renal function in using the new CKD-​EPI equation and other equations manuscript and reviewed or edited the manuscript before
kidney transplant recipients. Nephrol. Dial. Transplant. in patients with autosomal dominant polycystic kidney submission. P.R., A.d.V. and G.R. made substantial contri­
28, 3096–3100 (2013). disease. Am. J. Nephrol. 31, 53–57 (2010). butions to discussions of the article content, wrote the
119. Hossain, M. A., Attia, A. & Shoker, A. Measurement 137. Spithoven, E. M. et al. Tubular secretion of creatinine manuscript and reviewed or edited the manuscript before sub-
error in estimated GFR slopes across transplant in autosomal dominant polycystic kidney disease: mission. S.L.-L. researched data for the article, contributed
chronic kidney disease stages. Am. J. Nephrol. 31, consequences for cross-​sectional and longitudinal substantially to discussions of its content, and reviewed or
151–159 (2010). performance of kidney function estimating equations. edited the manuscript before submission. A.J. researched data
120. Buron, F. et al. Estimating glomerular filtration rate in Am. J. Kidney Dis. 62, 531–540 (2013). for the article and contributed substantially to discussions of
kidney transplant recipients: performance over time of 138. [No authors listed.] Chapter 2: definition, its content. A.T. substantially contributed to discussions of the
four creatinine-​based formulas. Transplantation 92, identification, and prediction of CKD progression. article content.
1005–1011 (2011). Kidney Int. Suppl. 3, 63–72 (2013).
121. Attia, A., Zahran, A. & Shoker, A. Comparison of 139. Caroli, A. et al. Effect of longacting somatostatin Competing interests
equations to estimate the glomerular filtration rate in analogue on kidney and cyst growth in autosomal The authors declare no competing interests.
post-​renal transplant chronic kidney disease patients. dominant polycystic kidney disease (ALADIN):
Saudi J. Kidney Dis. Transpl. 23, 453–460 (2012). a randomised, placebo-​controlled, multicentre trial. Publisher’s note
122. Goerdt, P. J., Heim-​Duthoy, K. L., Macres, M. & Lancet 382, 1485–1495 (2013). Springer Nature remains neutral with regard to jurisdictional
Swan, S. K. Predictive performance of renal function 140. European Medicines Agency. Guideline on the clinical claims in published maps and institutional affiliations.
estimate equations in renal allografts. Br. J. investigation of medicinal products to prevent
Clin. Pharmacol. 44, 261–265 (1997). development/slow progression of chronic renal Reviewer information
123. Harzallah, K. et al. Creatinine clearance estimation insufficiency. EMA http://www.ema.europa.eu/docs/en_ Nature Reviews Nephrology thanks E. Cavalier, L. Dubourg
after kidney transplantation: an analysis of three GB/document_library/Scientific_guideline/2016/10/ and the other anonymous reviewer(s) for their contribution to
methods. Transplant. Proc. 39, 2571–2573 WC500214980.pdf (2016). the peer review of this work.
(2007). 141. British Transplantation Society. Guidelines for living
124. Kukla, A. et al. GFR-​estimating models in kidney donor kidney transplantation. BTS https://bts.org.uk/ Supplementary information
transplant recipients on a steroid-​free regimen. wp-​content/uploads/2018/07/FINAL_LDKT-​guidelines_ Supplementary information is available for this paper at
Nephrol. Dial. Transplant. 25, 1653–1661 (2010). June-2018.pdf (2018). https://doi.org/10.1038/s41581-018-0080-9.

190 | MARCH 2019 | volume 15 www.nature.com/nrneph

You might also like