Estimated Glomerular Filtration Rate Is a Poor Predictor of Concentration for a Broad Range of Uremic Toxins

Abstract

Summary

Background and objectives

The degree of chronic kidney disease (CKD) is currently expressed in terms of GFR, which can be determined directly or estimated according to different formulas on the basis of serum creatinine and/or cystatin C measurements (estimated GFR [eGFR]). The purpose of this study was to investigate whether eGFR values are representative for uremic toxin concentrations in patients with different degrees of CKD.

Design, setting, participants, & measurements

Associations between eGFR based on serum cystatin C and different uremic solutes (mol wt range 113 to 240 D; determined by colorimetry, HPLC, or ELISA) were evaluated in 95 CKD patients not on dialysis (CKD stage 2 to 5). The same analysis was also applied for six other eGFR formulas.

Results

There was a substantial disparity in fits among solutes. In linear regression, explained variance of eGFR was extremely low for most solutes, with eGFR > 0.4 only for creatinine. The other eGFR formulations gave comparably disappointing results with regard to their association to uremic solutes. Relative similarity in R² values per solute for the different eGFR values and the strong disparity in values between solutes suggest that the differences in R² are mainly due to discrepancies in solute handling apart from GFR.

Conclusions

eGFR is poorly associated with concentrations of all studied uremic toxins in patients with different degrees of CKD, correlates differently with each individual solute, and can thus not be considered representative for evaluating the accumulation of solutes in the course of CKD.

Introduction

As renal function gradually deteriorates, many different complications develop, affecting the general clinical condition and survival prognosis of patients with chronic kidney disease (CKD). This evolution parallels the retention of uremic solutes, which to a large extent are held responsible for this functional deterioration (1,2).

Severity of CKD is nowadays usually expressed in terms of GFR. GFR can be determined accurately with several labor-intensive and expensive methods (3), which for those reasons are usually reserved for selected patients at selected time points. Far more frequently estimated values, such as those calculated by the Modification of Diet in Renal Disease (MDRD) formula, are used, which are mostly calculated based on serum creatinine and anthropometric data (4); the current CKD classification system is based on this methodology (5,6). This approach has the advantage of being easily applicable for a low cost and without implying major workload (7). Because creatinine concentration may be affected by other factors than GFR, alternative calculation formulas that are based partly or entirely on cystatin C have been developed (8–11).

GFR or its surrogates are generally used at this time to estimate kidney function, which in its turn is the main element affecting the concentration of metabolites that are held responsible for the functional disturbances of uremia. However, over time it became more and more clear that the concentration of these solutes can be influenced by many factors. These could be, as also summarized in the review by Stevens and Levey (12), renal tubular secretion (indoxyl sulfate [IS], hippurate) (13), enzymatic metabolism (asymmetric dimethylarginine [ADMA]) (14), intestinal secretion/absorption (uric acid) (15), generation by intestinal flora (indoles, phenols) (16), and/or dietary habits or changes in distribution volume. Therefore, the question can be raised whether GFR predicts elimination and concentration of those solutes and thus their effect on the uremic status in a sufficiently robust manner.

Several studies already investigated in how far estimated GFR (eGFR) is an indicator for real GFR (12,17,18). However, a major question that, to the best of our knowledge, has not yet been answered is whether these eGFR values accurately reflect the concentration of the other different uremic retention solutes that are accumulated in progressive stages of CKD. Several of those retention compounds have recently been linked by in vitro or in vivo studies to mechanisms of vascular damage (19–25), whereas in observational trials they were also linked to worse clinical outcomes (26–28).

This question becomes more relevant in view of the findings of the recent randomized controlled Initiating Dialysis Early and Late (IDEAL) trial (29,30), in which approximately 75% of CKD patients targeted to start dialysis at low eGFR (5 to 7 ml/min per 1.73 m²) had to be started earlier. In most, this was attributed to uremic symptoms, suggesting that eGFR might be not very representative for the uremic status.

In the study presented here, the association between eGFR and a broad set of uremic solutes was evaluated in a group of patients with different degrees of kidney failure (i.e., CKD stage 2 to 5) not on dialysis.

Materials and Methods

Patient Selection

This is a planned subanalysis of a larger study undertaken over an 18-month period (January 2006 to June 2007), including a total of 150 Caucasian-prevalent CKD patients from the Nephrology Department at Amiens University Hospital, in which uremic retention molecules were evaluated for their relation with clinical outcomes (19,20). All patients gave their written informed consent. The study was approved by the local ethical committee and was performed in accordance with the principles of the Declaration of Helsinki.

Included patients had to be over the age of 40 years with a confirmed diagnosis of CKD, which was defined as having two previously estimated creatinine clearances with an interval of 3 to 6 months and values <90 ml/min per 1.73m² as calculated according to the Cockcroft–Gault formula (31). Exclusion criteria included the presence of chronic inflammatory disease, atrial fibrillation, complete heart block, abdominal aorta aneurysm, an aortic and/or femoral artery prosthesis, primary hyperparathyroidism, kidney transplantation, on dialysis, and any acute cardiovascular event in the 3 months before screening for inclusion. From the 140 patients (in CKD stage 1 to 5D) who met the inclusion criteria, 45 were not included because they were treated with hemodialysis because this strategy has an effect on solute removal apart from the natural removal routes and, because of this reason, should also affect the GFR estimation. Hence, in the analysis presented here, 95 patients met the inclusion criteria.

Sampling and Laboratory

Blood samples of all patients were taken in the morning from 9:00 a.m. onward. Blood samples were immediately centrifuged, frozen, and stored at −80°C.

Serum creatinine (SCrea; mol wt = 113 D) was measured colorimetrically by standard laboratory methods. Serum cystatin C (CysC) levels were determined by immunonephelometry (N Latex Cystatin C, Dade Behring, Marburg, Germany).

Different other solutes, most of them protein bound, were determined by HPLC. These included uric acid (UA; mol wt = 168 D), hippuric acid (HA; mol wt = 179 D, protein binding [PB] = ±50%), 3-carboxy-4-methyl-5-propyl-2-furanpropionic acid (CMPF; mol wt = 240 D, PB = ±100%), IS (mol wt = 213 D, PB = ±90%), indole acetic acid (IAA; mol wt = 175 D, PB = ±65%), and p-cresylsulfate (pCS; mol wt = 187 D, PB = ±95%). To determine the total fraction, serum samples were first deproteinized by heat denaturation (32) and analyses were performed by reverse-phase HPLC. IS and IAA (excitation wavelength = 280 nm; emission wavelength = 340 nm) and pCS (excitation wavelength = 265 nm; emission wavelength = 290 nm) were determined by fluorescence analysis, and HA and CMPF were analyzed by UV detection at 254 nm (33). Free fractions were determined according to Fagugli et al. (33). HPLC data that were beneath the detection limit (0.01 mg/dl for pCS, 0.02 mg/dl for IAA and IS, and 0.1 for HA) were not included.

Two ELISA kits manufactured by DLD Diagnostika GmbH (Hamburg, Germany) were used for measuring ADMA (mol wt = 202 D) and symmetric dimethylarginine (SDMA; mol wt = 202 D) after acylation.

eGFR Calculation

To assess correlation of uremic solute concentration with eGFR, seven different formulas to estimate GFR were used, of which two were based on SCrea alone: the MDRD-eGFR = 175 × SCrea^−1.154 × age^−0.203 × (0.742 if female) × (1.21 if black) (4); and the Cockcroft–Gault eGFR = [(140 − age) × weight] × 0.85 (if female)/(SCrea × 72] (31). One formula, the epi-GFR, was based on SCrea and CysC: eGFR = 177.6 × SCrea^−0.65 × CysC^−0.57 × age^−0.20 × 0.82 (if female) (8). Four formulas were based on CysC alone: eGFR = [78 × (1/CysC)] + 4 (10); log(eGFR) = 1.962 + [1.123 × log(1/CysC)] (9); eGFR = 66.8 × (CysC)^−1.3 (11); and eGFR = 127.7 × CysC^−1.17 × age^−0.13 × 0.91 (if female) × 1.06 (if black) (8).

Comparison of Creatinine Determinations by Colorimetry and HPLC

In the function of the results obtained, in which some of the differences could be attributed to creatinine being determined colorimetrically whereas most other determinations were made by HPLC, we measured SCrea colorimetrically and by HPLC and compared the results obtained with both methods.

Statistical Analyses

Data are expressed as mean ± SD. Pearson correlations and linear regressions were performed on semilogarithmic concentrations as a function of eGFR. Agreement between SCrea measured colorimetrically and by HPLC was investigated by linear regression analysis and Bland–Altman graphs. P ≤0.05 was considered statistically significant. All statistical analyses were performed using PASW Statistics 18 (SPSS, Inc., Chicago, IL) for Windows (Microsoft Corporation, Redmond, WA).

Results

Table 1 shows the demographic and clinical characteristics of the 95 included patients. With respect to CKD stage, 11.5% were classified as stage 2, 39.0% as stage 3, 39.0% as stage 4, and 10.5% as stage 5 (5).

Table 1.

Main demographic and clinical characteristics of the study population (n = 95)

	CKD Stage					Pa
	2 to 5	2	3	4	5
Number, n (%)	95 (100)	11 (11.5)	37 (39.0)	37 (39.0)	10 (10.5)
eGFR (ml/min per 1.73 m2)	35 ± 18	69 ± 8	43 ± 9	22 ± 4	11 ± 3	<0.001
Age (years)	68 ± 12	65 ± 8	69 ± 12	65 ± 13	66 ± 15	0.07
Male gender, n (%)	59 (62)	9 (82)	24 (65)	22 (60)	4 (40)	0.39
Diabetes mellitus, n (%)	45 (47)	4 (36)	19 (51)	18 (49)	4 (40)	0.50
BMI, kg/m2	29 ± 7	26 ± 5	29 ± 6	31 ± 7	28 ± 7	0.28
PAS, mmHg	153 ± 25	145 ± 27	153 ± 22	156 ± 29	143 ± 17	0.45
PAD, mmHg	81 ± 12	85 ± 9	80 ± 11	83 ± 12	75 ± 16	0.21
Cholesterol, mmol/L	5.0 ± 1.1	5.4 ± 0.7	4.6 ± 1.1	5.3 ± 1.1	4.6 ± 0.5	0.02
Triglycerides, mmol/L	1.9 ± 1.4	1.7 ± 0.9	1.6 ± 0.7	2.4 ± 1.9	2.2 ± 1.2	0.06
Calcium, mmol/L	2.3 ± 0.1	2.3 ± 0.1	2.3 ± 0.1	2.3 ± 0.2	2.3 ± 0.2	0.96
Phosphate, mmol/L	1.2 ± 0.3	0.9 ± 0.3	1.1 ± 0.2	1.4 ± 0.3	1.5 ± 0.5	<0.001
Albumin, g/L	38.9 ± 6.4	40.6 ± 8.8	38.4 ± 5.7	39.9 ± 5.8	33.8 ± 6.7	0.07
Hemoglobin, g/L	12.5 ± 1.7	14.0 ± 1.2	12.7 ± 1.5	12.0 ± 1.6	10.9 ± 1.4	<0.001
Vitamin D supplement, n (%)	17 (18)	1 (0.1)	5 (0.1)	6 (0.2)	5 (50)	0.06
Antihypertensives, n (%)	90 (94)	10 (83)	37 (100)	35 (94)	8 (80)	0.08
LLT, n (%)	65 (70)	8 (67)	27 (73)	24 (65)	5 (100)	0.50

Data are expressed as mean ± SD or number for binary variables, with percentages in parentheses. DM, diabetes mellitus; BMI, body mass index; PAS, pulmonary artery systolic pressure; PAD, pulmonary artery diastolic pressure; LLT, lipid-lowering therapy; eGFR, estimated GFR; CKD, chronic kidney disease.

^aANOVA P value comparing stage 2 to 5.

The distribution of serum levels of each evaluated uremic retention solute (mean ± SD) in function of the CKD stages is given in Table 2.

Table 2.

Serum levels of diverse uremic toxins by CKD stage (mg/dl)

	CKD Stage
	2	3	4	5
Creatinine	1.14 ± 0.18	1.52 ± 0.32	2.85 ± 0.96	5.30 ± 1.93
SDMA	0.049 ± 0.017	0.071 ± 0.037	0.115 ± 0.075	0.208 ± 0.065
ADMA	0.057 ± 0.008	0.065 ± 0.010	0.069 ± 0.016	0.076 ± 0.015
Uric acid	7.17 ± 2.33	8.07 ± 1.89	8.33 ± 2.13	8.98 ± 2.66
IS
total	0.15 ± 0.06	0.26 ± 0.14	0.46 ± 0.28	0.92 ± 0.65
free	0.032 ± 0.005	0.03 ± 0.004	0.04 ± 0.01	0.06 ± 0.03
HA
total	0.45 ± 0.16	0.43 ± 0.25	0.59 ± 0.43	0.91 ± 0.46
free	0.28 ± 0.12	0.26 ± 0.12	0.35 ± 0.19	0.53 ± 0.24
pCS
total	0.59 ± 0.33	1.03 ± 0.67	1.68 ± 1.34	2.91 ± 1.67
free	0.028 ± 0.032	0.03 ± 0.03	0.08 ± 0.08	0.19 ± 0.12
IAA
total	0.06 ± 0.03	0.09 ± 0.04	0.10 ± 0.04	0.14 ± 0.89
free	0.023 ± 0.002	0.030 ± 0.004	0.030 ± 0.007	0.030 ± 0.009
Total CMPFa	0.31 ± 0.13	0.34 ± 0.15	0.43 ± 0.31	0.42 ± 0.31

Data are expressed as mean ± SD. SDMA, symmetric dimethylarginine; ADMA, asymmetric dimethylarginine; IS, indoxyl sulfate; HA, hippuric acid; pCS, p-cresylsulfate; IAA, indole acetic acid; CMPF, 3-carboxy-4-methyl-5-propyl-2-furanpropionic acid.

^aOnly total concentration is given because the free concentration was below the detection limit.

Linear regression analyses with the CysC-based eGFR formula developed by Stevens et al. (8) as the independent variable and the natural logarithm of the concentration of each uremic retention solute as the dependent variable are illustrated in Figure 1 and Table 3 by their respective R² values. We focused first on the formula developed by Stevens et al. (8) in view of its development in a large and representative population and because it is not based on SCrea, which was one of the dependent variables under study.

Figure 1.

Coefficient of determination (R²) of the linear regression analysis for different natural logarithmic uremic retention solute concentrations (total and free concentrations) as a function of the eGFR as calculated according to Stevens et al. (8). The limits of R² = 0.4 and 0.2 are indicated as dashed lines. The association between eGFR and the natural logarithm of uremic retention solutes was significant (P < 0.05), except for total CMPF.

Table 3.

Linear regression analyses (R² values) between the natural logarithmic functions of different uremic retention solute concentrations as a function of the eGFR as calculated according to four different methods

		MDRD	C-G	epi-GFR	Stevens
SCrea		0.833	0.737	0.737	0.570
SDMA		0.237	0.300	0.309	0.320
ADMA		0.108	0.149	0.167	0.193
UA		0.054	0.062	0.058	0.062
IS	Total	0.416	0.426	0.426	0.361
HA	Total	0.107	0.141	0.130	0.117
pCS	Total	0.150	0.129	0.119	0.085
IAA	Total	0.123	0.134	0.132	0.115
CMPF	Total	0.035	0.016	0.014	0.009
IS	Free	0.265	0.289	0.282	0.257
HA	Free	0.122	0.160	0.147	0.134
pCS	Free	0.203	0.175	0.164	0.124
IAA	Free	0.218	0.257	0.253	0.238

C-G, Cockcroft-Gault; SCrea, serum creatinine; MDRD, Modification of Diet in Renal Disease. All correlations were significant (P <0.05) except total CMPF. R² values <0.2 are in bold.

The association with SCrea was the most significant (R² = 0.570). When considering the other solutes, R² was found <0.2 for ADMA, free HA, free pCS, total HA, total IAA, total pCS, UA, and CMPF; and only for total and free IS, SDMA, and free IAA was R² in the range of 0.2 to 0.4. No solutes other than SCrea showed an R² > 0.4; the highest other R² values were at least 40% lower than those for creatinine (R² = 0.570 versus R² = 0.361 for SCrea and total IS, respectively). The variation in serum concentration of the tested uremic solutes is thus only to a limited and variable extent explained by variations in eGFR.

The distribution of the serum levels of four studied uremic solutes as a function of eGFR is illustrated in Figure 2, which illustrates how the cloud of individual points diverges from the regression line, even with the solute concentration expressed as natural logarithms. As compared with creatinine (Figure 2A), this divergence is more pronounced for all other molecules, with total IS as second best (Figure 2B) and total CMPF and UA as least correlated solutes (Figure 2, C and D, respectively). In addition, except for creatinine, an overlap in the values was found (i.e., that similar concentration values can be found irrespective of the CKD subgroup) (Figure 2, B through D), pointing again to the large variability in the relation between concentrations and eGFR.

Figure 2.

Relationships (inverse correlation for all) between the natural logarithmic uremic retention solute concentrations and the eGFR according to Stevens et al. (8): (A) SCrea, (B) total IS, (C) total CMPF, and (D) UA. The figure shows the regression lines (solid lines) and the 95% confidence interval (dashed lines).

All other eGFR formulas gave similar results as the ones obtained with the formula of Stevens et al., as illustrated in Table 3 (eGFR formulas based on CysC alone and those other than Stevens et al. were not shown in the table). The sole exception was the relation with creatinine, which, as expected, gave better R² values for creatinine-based formulations.

Because SCrea was determined colorimetrically whereas most other determinations were made by HPLC, and because one reason for the lower degree of association of most other solutes as compared with creatinine could be differences in the variability of results (e.g., if colorimetric determinations were less variable than HPLC measurements), we compared SCrea as determined by both approaches on the same samples. We found SD values of 71% (colorimetry) and 68% (HPLC) of the means, pointing to a similar variability. Thus, it is more likely that differences in association are attributable to something other than technical factors interfering with determination. In addition, both sets of creatinine data were significantly correlated (R = 0.997; P < 0.001) (Figure 3A, broken line is the identity line), but, as expected, a Bland–Altman plot (Figure 3B, the average difference is indicated by the thick line and the ±1.96 SD as dotted lines) shows that the values obtained with HPLC were lower than with colorimetry because of interference of noncreatinine chromogens with the colorimetric method.

Figure 3.

Results of statistical analysis of creatinine values. (A) Bland–Altman graph with bias (average difference) as thick line and 1.96SD as dotted lines. (B) Creatinine values as measured with colorimetry and those as measured with HPLC (R = 0.997; P < 0.001). The broken line is the identity line.

Discussion

The study presented here was undertaken to evaluate the relationship between eGFR as calculated with different currently applied formulas and the concentration of various uremic retention products, several of which have been linked to inflammation, vascular disease, and mortality (19–24,27,28,34–37).

This study demonstrated a main finding of substantial disparity among the evaluated uremic retention solutes in the coefficients of regression (R²) between eGFR and the natural logarithms of solute concentrations. R² was even below 0.2 for ADMA, free HA, free pCS, total HA, total IAA, total pCS, UA, and CMPF. eGFR thus poorly reflects accumulation of other uremic retention products than creatinine. In addition, individual data are substantially more scattered around the regression line (Figure 2), especially for all other molecules than creatinine.

UA, ADMA, pCS, and IS have repeatedly been associated with vascular damage and mortality in renal and general populations (19,25,27,28,36–40). UA, ADMA, and total and free pCS are all characterized by surprisingly weak relationships with eGFR (R² values in the range of 0.05 to 0.20), suggesting that other factors than GFR affect their concentration and that eGFR may not adequately reflect GFR. Although R² was slightly higher for free and total IS, another factor related to cardiovascular damage (22,27), the role of eGFR also seems very limited, with R² values of approximately 0.26 to 0.42, suggesting again that the change in solute concentration during progression of renal failure to a large extent (>50%) is influenced by other mechanisms than changes in eGFR. Even if the relationship with eGFR is better for creatinine than for all other compounds, it still remains mediocre. Of note, of all uremic retention solutes considered, creatinine is the one that has least of all been linked to biochemical/biologic, and thus toxic, effects at concentrations observed in uremia.

In our analysis, seven different formulas based on SCrea and/or CysC were used, each derived in different groups of patients. The Cockcroft–Gault (31) and MDRD formulas (4) were chosen because of their frequent use in the past or at present. In addition, they were both included because one is claimed to be more reliable in the low GFR ranges and the other in the high ones. The epi-GFR formula (8) was developed most recently and is frequently used in the context of studies. In addition to these formulas that are partly or entirely based on SCrea, four other formulas were chosen that are purely based on CysC, a marker less dependent on non-GFR factors than creatinine. In this context, the formula of Stevens et al. (8) was used as the key formula because it is based on a large group of U.S. and European patients (3418 total). The other CysC-based formulas used in the study presented here were based on smaller groups and sometimes nonspecific subgroups of the CKD population: 25 French adult transplants (Le Bricon et al. [10]), 536 Canadian children with renal pathologies (Filler et al. [9]), and 204 adult U.S. CKD patients (Rule et al. [11]). Nevertheless, all CysC-based formulas, including the one proposed by Stevens et al., gave comparable results (data not shown). In addition, the data based on Stevens et al. were largely comparable to those obtained with the creatinine-based formulations and the epi-GFR formula based on SCrea and CysC (Table 3).

Although most formulas were derived in patient populations including a minority of diabetics (4,8–11,31), 47% of our study population consisted of patients with diabetes mellitus. However, the R² values for the linear regression between the natural logarithm of the concentrations and the eGFR in this subgroup did not differ substantially from those as found in the entire population (data not shown), again with predominant differences among solutes for the same eGFR value and comparable results among eGFR formulas for the same solute. Hence, the conclusions of our study are identical when we only focus on the nondiabetic group (n = 50).

Our findings also shed more light on the recently published findings of the IDEAL trial, in which use of eGFR as an index for starting dialysis was apparently not adequate enough to guide postponing the start of dialysis (29,30). On the contrary, 75.9% of the patients randomized to start late had to be started before the target eGFR was reached, mostly because of uremic symptoms. Thus, eGFR does not seem to allow waiting to start dialysis and cannot offer a minimum threshold until which the start of dialysis can be postponed. The clinical condition and the underlying biochemical changes seem to overrule eGFR when it comes to deciding to enroll patients into chronic dialysis. Part of this failure can be attributed to the imperfect correlation between the eGFR and the true GFR. The R² of 0.737 to 0.833 (Table 3) observed in our study between SCrea and creatinine-based eGFR is probably the reflection of this discrepancy. However, this does not explain the important differences in R² values among compounds. If the discrepancies between true GFR and eGFR were the only determinant in the disappointing correlation between eGFR and uremic toxin concentration, then all R² values would be more or less in the same range. Now they range between 0.035 and 0.833 (MDRD—Table 3) or between 0.009 and 0.570 (Stevens et al.—Table 3).

A strength of our study is that the analysis is done for seven different formulas of eGFR calculation, for a wide range of eGFR values covering CKD classes 2 to 5, and for several uremic toxins that have been associated with (cardio)vascular damage and/or mortality. Although our results cannot be generalized for all available eGFR formulations and all identified uremic toxins, our study clearly shows that eGFR is not predictive for the concentration of several uremic toxins that have been related to cardiovascular morbidity and mortality.

A shortcoming of this study is its transversal nature with only one set of values per patient at one single moment. However, the type of analysis presented here starting at CKD stage 2 would in a longitudinal setting have taken substantially more time and would have necessitated the inclusion of a much larger population, whereas the mechanisms potentially affecting solute concentration are likely the same, irrespective of whether the approach for evaluation is longitudinal or transversal. Likewise, the use of eGFR to represent kidney function admittedly gives only an approximate value for glomerular filtration because it is based on correlation analyses and would have been measured more exactly by direct assessment methods such as EDTA clearance. We preferred to use methods that are applied on a day-to-day basis. Nevertheless, the differences in correlations are so striking that these findings very likely can be extrapolated to GFR in general, and we may accept that GFR as such is far from entirely representative for the behavior of concentration of uremic retention solutes with moreover considerable individual differences among solutes.

In conclusion, we observed that several currently used eGFR formulas for evaluating renal dysfunction in clinical practice in patients with different degrees of CKD poorly reflect the accumulation of a broad range of uremic retention solutes that have been linked to toxicity, especially in the cardiovascular system.

Disclosures

None.