See Clinical Research on Page 786
In the United States, 37 million adults (15%) have predialysis chronic kidney disease
(CKD) defined by albuminuria (urine albumin:creatinine ratio ≥30 mg/g) or decreased
estimated glomerular filtration rate (eGFR <60 ml/min per 1.73 m2) by the Chronic
Kidney Disease Epidemiology Collaboration (CKD-EPI) equation.
1
Nearly half (16.3 million) of those with predialysis CKD are 65 years or older. National
surveys use single random measurements of kidney markers and therefore overestimate
CKD prevalence by 2 to 3 percentage points, largely due to variability in albuminuria.S1
Conversely, commonly used eGFR equations, such as the CKD-EPI or Modification of Diet
in Renal Disease (MDRD), generally underestimate true glomerular filtration rate (GFR),
but may overestimate it in the elderly with low muscle mass.
2
Precision of eGFR can be extremely important for appropriate patient management, where
diagnosis of CKD, drug dosing, and proper health care utilization often depend on
accurate estimates of kidney function. Few aspects of clinical care among older adults
have stirred more controversy than the accuracy and precision of GFR obtained from
estimating equations.
In this issue, Scarr et al.
3
assessed the performance of eGFRs by serum creatinine, cystatin C, and β2-microglobulin
(β2M) as compared with measured GFR (mGFR) in older adults with and without type 1
diabetes in the Canadian Study of Longevity in Type 1 Diabetes. Seventy-five participants
with type 1 diabetes and a disease duration ≥50 years were age- and sex-matched with
75 controls without diabetes, and eGFR was calculated using the MDRD; CKD-EPI, creatinine
(CKD-EPIcr); CKD-EPI, cystatin C (CKD-EPIcys); CKD-EPIcr-cys; and β2M equations. GFR
was also measured by the mean of 2 plasma inulin clearances. Measures of bias, precision,
and accuracy were used to evaluate the performance of the eGFR equations overall and
within subgroups. Bias was defined as the mean difference between eGFR and mGFR and
precision was estimated from the SD of the bias. Accuracy, a marker of both bias and
precision, was calculated as the percentage of eGFRs outside a 30% or 20% range of
the mGFR (1 − P30 and 1 − P20), respectively. The 95% confidence intervals around
these measures were calculated using a bootstrap method. The authors observed that
although no participants had mGFR <60 ml/min per 1.73 m2, 6% of participants were
classified as having eGFR <60 ml/min per 1.73 m2 by the MDRD and CKD-EPIcr equations,
30% by CKD-EPIcys, 12% by CKD-EPIcr-cys, and 9% by β2M. All eGFRs significantly underestimated
mGFR, with greater bias observed for cystatin-based equations than the other equations.
Although bias was lowest for β2M (1.9 ml/min per 1.73 m2, P = 0.61), this marker was
also associated with the lowest precision. For all eGFR equations, negative bias was
greater and accuracy was lower among participants with higher mGFR.
An unexpected finding observed by Scarr et al.
3
was that cystatin C did not improve eGFR performance over serum creatinine alone.
Indeed, accuracy was higher for creatinine-based eGFR (MDRD 32.4%, CKD-EPIcr 37.4%)
than for eGFR computed using any other markers they tested (CKD-EPIcr-cys 52.5%, β2M
52.5%, and CKD-EPIcys 69.1%, P < 0.05 for all comparisons). Performance metrics were
similar for creatinine- or cystatin-based eGFRs, regardless of diabetes, whereas β2M
eGFR showed significantly greater bias, lower precision, and lower accuracy in participants
with type 1 diabetes than in controls. In subgroup analyses, cystatin-based eGFRs
showed greater negative bias, that is, underestimation of measured GFR, and lower
accuracy in older participants and in women.
The Kidney Disease: Improving Global Outcomes guidelinesS2 recommend that GFR estimates
based on cystatin C be used to confirm the presence of CKD in adults with eGFRcr of
45 to 59 ml/min per 1.73 m2 but without evidence of kidney damage, or in patients
in whom an accurate determination of GFR is required and measurement of GFR with an
exogenous marker is not feasible. This recommendation heeded ongoing controversy over
the application of an arbitrary and isolated threshold of GFR to define CKD. An eGFRcys
or eGFRcr-cys below 60 ml/min per 1.73 m2 would confirm the presence of CKD diagnosed
by a creatinine-based equation, whereas a value above this cutoff would refute this
diagnosis. The rationale for this practice was based on a significant body of literature
indicating that cystatin C improved accuracy of GFR estimation and had greater predictive
power for clinical outcomes, including end-stage kidney disease, mortality, and cardiovascular
disease events, leading to better CKD classification than creatinine-based eGFR. The
guideline made no specific suggestions or recommendations for use of these markers
in the elderly. However, because serum cystatin C is less dependent on muscle mass
than serum creatinine and is virtually completely cleared from the circulation by
glomerular filtration with subsequent proximal tubular uptake and degradation, it
is generally considered as an ideal alternate marker of kidney function, particularly
in older individuals. How then do we explain the findings by Scarr et al.?
3
Several factors need to be considered when addressing this question. First, significant
interlaboratory variability in serum creatinine measurement persists even after the
introduction of isotope dilution mass spectrometry standardization.
4
Studies analyzing patient samples across the range of serum creatinine concentrations
find that Jaffe assays, as used by Scarr et al.,
3
yield higher creatinine values than enzymatic assays, leading to a more frequent diagnosis
of CKD. In one study in which serum creatinine concentration was measured by both
methods in the same samples,
4
60% of eGFRs based on creatinine measured by the Jaffe method yielded eGFR values <60
ml/min per 1.73 m2, whereas only 39% of eGFRs based on creatinine measured by the
enzymatic method did so. In addition, serum concentrations of bilirubin >0.5 mg/dl
or of glucose >90 mg/dl were shown to increase interlaboratory variability and differences
between Jaffe and enzymatic results, particularly when the creatinine concentration
is low, as in elderly persons with low muscle mass. And cystatin C concentrations
also vary substantially between assays despite claims of calibration traceability
to the ERM-DA471/IFCC reference material.
5
Second, a non-normal distribution of the differences between eGFR methods may adversely
affect performance comparisons.
6
,
S3 If the differences are not normally distributed, transformation of the original
data may be helpful, or nonparametric tests with confidence limits computed using
bootstrap procedures that account for non-normality should be considered.
Third, eGFR equations (like all estimating equations) perform best in the cohorts
from which they are developed. To date, few studies have explored the performance
of eGFRs in elderly individuals of diverse racial/ethnic background and a wide spectrum
of kidney function, and even fewer include frail and hospitalized patients, in whom
these estimates are least reliable but most needed. Newer equations developed specifically
for this age category, such as the Berlin Initiative Study creatinine equation, the
Berlin Initiative Study creatinine and cystatin C equation,S4 or the Full Age Spectrum
equation,
7
appear to offer better accuracy than either MDRD or CKD-EPI, but these equations have
not been externally validated. Even as newer equations are improving on previous ones,
the applicability of a single equation to all situations may ultimately require a
trade-off between the cost of the filtration markers and the accuracy and precision
of the estimate. Foreseeable alternatives are novel, rapid, and affordable methods
of GFR measurement in humans. Transcutaneous GFR measurements using exogenous fluorescent
marker fluorescein isothiocyanate–sinistrin have been studied in animal models, allowing
GFR assessment in real time without serial blood or urine sampling.
8
Fluorescence technologies in humans are being developed using 2-compartment GFR measurements
that allow rapid direct quantification of GFR and renal reserve. GFR measurement by
iohexol clearance using dried capillary blood spots may also be a useful option when
an accurate measurement is required.
9
The findings reported by Scarr et al.,
3
and the inconsistencies they and others have noted among studies of eGFR performance
generally, remind us of the many challenges associated with estimating and interpreting
GFR in elderly persons and in other high-risk groups in whom precise and accurate
estimates of kidney function are increasingly needed for optimal clinical management.
Identifying appropriate filtration markers and estimating equations for these important
subgroups first requires that we address inconsistencies across studies caused by
differences in study design; statistical approach; laboratory assays; and specimen
collection, handling, and storage. Perhaps through standardized reporting requirements
for eGFR performance comparisons, we can more confidently identify the best approaches
for evaluating kidney function in these groups. The study by Scarr et al.
3
moves us closer to that goal.
Disclosure
The authors declared no competing interests.