Title: Glomerular filtration rate estimated by cystatin C among different clinical presentations
Abstract: Glomerular filtration rate (GFR) estimates from serum creatinine has not been generalizable across all populations. Cystatin C has been proposed as an alternative marker for estimating GFR. The objective of this study was to compare cystatin C with serum creatinine for estimating GFR among different clinical presentations. Cystatin C and serum creatinine levels were obtained from adult patients (n=460) during an evaluation that included a GFR measurement by iothalamate clearance. Medical records were abstracted for clinical presentation (healthy, native chronic kidney disease or transplant recipient) at the time of GFR measurement. GFR was modeled using the following variables: cystatin C (or serum creatinine), age, gender and clinical presentation. The relationship between cystatin C and GFR differed across clinical presentations. At the same cystatin C level, GFR was 19% higher in transplant recipients than in patients with native kidney disease (P<0.001). The association between cystatin C and GFR was stronger among native kidney disease patients than in healthy persons (P<0.001 for statistical interaction). Thus, a cystatin C equation was derived using only patients with native kidney disease (n=204). The correlation with GFR (r2=0.853) was slightly higher than a serum creatinine equation using the same sample (r2=0.827), the Modification of Diet in Renal Disease equation (r2=0.825) or the Cockcroft–Gault equation (r2=0.796). Averaged estimates between cystatin C and serum creatinine equations further improved correlation (r2=0.891). Cystatin C should not be interpreted as purely a marker of GFR. Other factors, possibly inflammation or immunosuppression therapy, affect cystatin C levels. While recognizing this limitation, cystatin C may improve GFR estimates in chronic kidney disease patients. Glomerular filtration rate (GFR) estimates from serum creatinine has not been generalizable across all populations. Cystatin C has been proposed as an alternative marker for estimating GFR. The objective of this study was to compare cystatin C with serum creatinine for estimating GFR among different clinical presentations. Cystatin C and serum creatinine levels were obtained from adult patients (n=460) during an evaluation that included a GFR measurement by iothalamate clearance. Medical records were abstracted for clinical presentation (healthy, native chronic kidney disease or transplant recipient) at the time of GFR measurement. GFR was modeled using the following variables: cystatin C (or serum creatinine), age, gender and clinical presentation. The relationship between cystatin C and GFR differed across clinical presentations. At the same cystatin C level, GFR was 19% higher in transplant recipients than in patients with native kidney disease (P<0.001). The association between cystatin C and GFR was stronger among native kidney disease patients than in healthy persons (P<0.001 for statistical interaction). Thus, a cystatin C equation was derived using only patients with native kidney disease (n=204). The correlation with GFR (r2=0.853) was slightly higher than a serum creatinine equation using the same sample (r2=0.827), the Modification of Diet in Renal Disease equation (r2=0.825) or the Cockcroft–Gault equation (r2=0.796). Averaged estimates between cystatin C and serum creatinine equations further improved correlation (r2=0.891). Cystatin C should not be interpreted as purely a marker of GFR. Other factors, possibly inflammation or immunosuppression therapy, affect cystatin C levels. While recognizing this limitation, cystatin C may improve GFR estimates in chronic kidney disease patients. Ideally, glomerular filtration rate (GFR) should be determined with a method that is convenient, inexpensive, and accurate. GFR can be estimated from serum creatinine using the Modification of Diet in Renal Disease (MDRD) or Cockcroft–Gault equations.1.Levey A.S. Coresh J. Balk E. et al.National kidney foundation practice guidelines for chronic kidney disease: evaluation, classification, and stratification.Ann Intern Med. 2003; 139: 137-147Crossref PubMed Scopus (3462) Google Scholar, 2.Cockcroft D.W. Gault M.H. Prediction of creatinine clearance from serum creatinine.Nephron. 1976; 16: 31-41Crossref PubMed Scopus (12499) Google Scholar However, these equations have not been generalizable across all clinical presentations. For example, the MDRD equation, derived with chronic kidney disease (CKD) patients, underestimated GFR in healthy persons by 29%.3.Rule A.D. Larson T.S. Bergstralh E.J. et al.Using serum creatinine to estimate glomerular filtration rate: accuracy in good health and in chronic kidney disease.Ann Intern Med. 2004; 141: 929-937Crossref PubMed Scopus (876) Google Scholar This occurred in part because the strength of association between serum creatinine and GFR is much less among healthy persons compared to that of patients with CKD.3.Rule A.D. Larson T.S. Bergstralh E.J. et al.Using serum creatinine to estimate glomerular filtration rate: accuracy in good health and in chronic kidney disease.Ann Intern Med. 2004; 141: 929-937Crossref PubMed Scopus (876) Google Scholar, 4.Lin J. Knight E. Hogan M. Singh A. A comparison of prediction equations for estimating glomerular filtration rate in adults without kidney disease.J Am Soc Nephrol. 2003; 14: 2573-2580Crossref PubMed Scopus (362) Google Scholar, 5.Poggio E.D. Wang X. Greene T. et al.Performance of the modification of diet in renal disease and Cockcroft–Gault equations in the estimation of GFR in health and in chronic kidney disease.J Am Soc Nephrol. 2005; 16: 459-466Crossref PubMed Scopus (579) Google Scholar The MDRD equation also overestimates the strength of association between GFR and serum creatinine in type I diabetes pateints with normal serum creatinine levels.6.Ibrahim H. Mondress M. Tello A. et al.An alternative formula to the Cockcroft–Gaultand the modification of diet in renal disease formulas in predicting GFR in individuals with Type 1 diabetes.J Am Soc Nephrol. 2005; 16: 1051-1060Crossref PubMed Scopus (118) Google Scholar Studies have suggested that serum cystatin C may have advantages over serum creatinine for estimating GFR. Cystatin C is a low molecular weight basic protein (13 kDa) that is freely filtered and metabolized after tubular re-absorption.7.Abrahamson M. Olafsson I. Palsdottir A. et al.Structure and expression of the human cystatin C gene.Biochem J. 1990; 268: 287-294Crossref PubMed Scopus (490) Google Scholar, 8.Tenstad O. Roald A.B. Grubb A. Aukland K. Renal handling of radiolabelled human cystatin C in the rat.Scand J Clin Lab Invest. 1996; 56: 409-414Crossref PubMed Scopus (367) Google Scholar Studies have reported that cystatin C is less influenced by age, gender and muscle mass than serum creatinine.9.Finney H. Newman D.J. Price C.P. Adult reference ranges for serum cystatin C. creatinine and predicted creatinine clearance.Ann Clin Biochem. 2000; 37: 49-59Crossref PubMed Scopus (199) Google Scholar, 10.Norlund L. Fex G. Lanke J. et al.Reference intervals for the glomerular filtration rate and cell-proliferation markers: serum cystatin C and serum beta 2-microglobulin/cystatin C-ratio.Scand J Clin Lab Invest. 1997; 57: 463-470Crossref PubMed Scopus (129) Google Scholar, 11.Vinge E. Lindergard B. Nilsson-Ehle P. Grubb A. Relationships among serum cystatin C, serum creatinine, lean tissue mass and glomerular filtration rate in healthy adults.Scand J Clin Lab Invest. 1999; 59: 587-592Crossref PubMed Scopus (196) Google Scholar However, in at least one general population study, cystatin C was found to be influenced by several factors including age, gender, body size, cigarette smoking and C-reactive protein independent of creatinine clearance.12.Knight E.L. Verhave J.C. Spiegelman D. et al.Factors influencing serum cystatin C levels other than renal function and the impact on renal function measurement.Kidney Int. 2004; 65: 1416-1421Abstract Full Text Full Text PDF PubMed Scopus (720) Google Scholar Another study suggested that cystatin C was not simply a marker of GFR because it predicted future cardiovascular events independent of estimated creatinine clearance.13.Koenig W. Twardella D. Brenner H. Rothenbacher D. Plasma concentrations of cystatin C in patients with coronary heart disease and risk for secondary cardiovascular events: more than simply a marker of glomerular filtration rate.Clin Chem. 2005; 51: 321-327Crossref PubMed Scopus (241) Google Scholar A potential problem with these studies was the evaluation of cystatin C with serum creatinine-based equations or creatinine clearance instead of with measured GFR. The primary objective of this study was to assess the relationship between cystatin C and GFR (iothalamate clearance) among clinical presentations that commonly lead to a GFR measurement. The secondary objective was to compare a cystatin C equation to a serum creatinine equation for estimating GFR. Table 1 presents demographic characteristics and laboratory measurements, overall and by clinical presentation. GFR estimated by the abbreviated MDRD equation or the Cockcroft–Gault equation was similar to GFR measured by iothalamate clearance except in the healthy group. Transplant recipients were classified by kidney graft alone (n=103, 50%), liver graft (n=55, 27%), heart graft (n=30, 15%), pancreas graft (n=15, 7%) and lung graft (n=3, 1%). Patients with non-kidney grafts could be further classified as those who had an iothalamate clearance as part of routine follow-up versus those who had it for CKD, but this distinction was not important in multivariable models.Table 1Descriptive characteristics and laboratory measurements, overall and by clinical presentationOverallHealthy (potential donors)Native kidney diseaseKidney transplant recipientOther transplant recipientSample size46050204103103Female, n (%)193 (42)34 (68)92 (45)38 (37)29 (28)Caucasian, n (%)446 (97)50 (100)197 (97)100 (97)99 (96)Diabetic, n (%)74 (16)0 (0)37 (18)19 (18)18 (17)Age (years)51±1541±1155±1649±1351±12Height (cm)171±10168±9171±10171±10171±10Weight (kg)84±2080±1786±2383±2084±17Body mass index (kg/m2)29±628±629±728±528±6Iothalamate GFR (ml/min/1.73 m2)57±29101±1651±2952±1854±22MDRD GFRaMDRD GFR=186 × SCr-1.154 × age-0.203 (0.742 if female) (1.21 if black).1, 14 (ml/min/1.73m2)55±2486±1350±2551±1557±21C-G GFRbC-G GFR=((140-age) weight(kg)/(SCr × 72)) (0.85 if female) (1.73/body surface area) × 0.84.2, 14 (ml/min/1.73 m2)55±2587±1949±2650±1555±19Serum creatinine (mg/dl)1.6±0.90.9±0.21.8±1.21.5±0.41.5±0.6Cystatin C (mg/l)bC-G GFR=((140-age) weight(kg)/(SCr × 72)) (0.85 if female) (1.73/body surface area) × 0.84.2, 141.5±0.80.8±0.21.6±0.91.6±0.61.6±0.7Entries are mean±s.d. or count (percent) where appropriate.a MDRD GFR=186 × SCr-1.154 × age-0.203 (0.742 if female) (1.21 if black).1.Levey A.S. Coresh J. Balk E. et al.National kidney foundation practice guidelines for chronic kidney disease: evaluation, classification, and stratification.Ann Intern Med. 2003; 139: 137-147Crossref PubMed Scopus (3462) Google Scholar, 14.Levey A.S. Bosch J.P. Lewis J.B. et al.A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group.Ann Intern Med. 1999; 130: 461-470Crossref PubMed Scopus (12277) Google Scholarb C-G GFR=((140-age) weight(kg)/(SCr × 72)) (0.85 if female) (1.73/body surface area) × 0.84.2.Cockcroft D.W. Gault M.H. Prediction of creatinine clearance from serum creatinine.Nephron. 1976; 16: 31-41Crossref PubMed Scopus (12499) Google Scholar, 14.Levey A.S. Bosch J.P. Lewis J.B. et al.A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group.Ann Intern Med. 1999; 130: 461-470Crossref PubMed Scopus (12277) Google Scholar Open table in a new tab Entries are mean±s.d. or count (percent) where appropriate. The relationship between cystatin C and measured GFR (Figure 1a), and between serum creatinine and measured GFR (Figure 1b) is shown on a logarithmic scale. Natural logarithmic (ln) GFR was regressed on ln analyte for each clinical presentation. There was a stronger regression fit between ln GFR and ln cystatin C than between ln GFR and ln serum creatinine for the native kidney disease group (r2=0.853 versus 0.770), the transplant recipient group (r2=0.768 versus 0.671), and the healthy group (r2=0.382 versus 0.123). There was an upward shift in the regression line between GFR and cystatin C among transplant recipients compared to those with native kidney disease (Figure 1a); this was not observed between GFR and serum creatinine (Figure 1b). The healthy group had a more gradual slope than the native kidney disease group when GFR was regressed on either cystatin C or serum creatinine. Table 2 presents the effect of predictor variables on GFR in a multivariable model based on cystatin C versus a similar model based on serum creatinine. Both a 50% increase in cystatin C and a 50% increase in serum creatinine were associated with a 39% decrease in GFR. Age was a stronger predictor of GFR in a model based on serum creatinine than in a model based on cystatin C (P<0.002 for both). At the same serum creatinine level, female subjects had a 23% lower GFR than male subjects (P<0.001), but there was no difference at the same cystatin C level (P=0.18). At the same serum creatinine level, healthy persons had a 15% higher GFR than patients with native kidney disease (P=0.002), but there was no difference at the same cystatin C level (P=0.28). At the same cystatin C level, transplant recipients (kidney, liver, heart, pancreas or lung) had a 19% higher GFR (P<0.001) than patients with native kidney disease. At the same serum creatinine level, liver transplant recipients had a 10% lower GFR (P=0.007) than patients with native kidney disease, but there was no difference with non-liver transplant recipients (P=0.36).Table 2Effect of predictor variables on GFR in a model based on cystatin C versus a similar model based on serum creatinine (sample size=460)% Difference in measured GFR (95% confidence interval)Cystatin C modelSerum creatinine modelCystatin C (per 50% increase)-38.8 (-40.2 to -37.3)—Serum creatinine (per 50% increase)—-39.0 (-40.5 to -37.6)Age (per 50% increase)-4.8 (-7.5 to -1.9)-8.2 (-11.0 to -5.3)Male (n=267)RefRefFemale (n=193)-3.1 (-7.3 to 1.4)-23.2 (-26.9 to -19.3)Native kidney disease (n=204)RefRefHealthy (n=50)4.7 (-3.6 to 13.7)14.8 (5.5 to 24.9)Kidney recipient (n=103)18.3 (11.8 to 13.7)2.5 (-3.5 to 9.0)Liver recipient (n=55)19.1 (11.0 to 27.8)-9.9 (-16.3 to -2.9)Heart recipient (n=30)25.5 (14.4 to 37.6)3.8 (-5.9 to 14.5)Other organ recipientaPancreas or lung. (n=18)13.7 (1.4 to 27.3)-2.9 (-14.0 to 9.6)Model fit (r2)bModel fit or coefficient of determination (r2) is the proportion of between patient variability in measured GFR explained by the model.0.8520.832Root mean square error (%)cRoot mean square error is an average error between measured GFR and estimated GFR given as a percentage.26.528.4a Pancreas or lung.b Model fit or coefficient of determination (r2) is the proportion of between patient variability in measured GFR explained by the model.c Root mean square error is an average error between measured GFR and estimated GFR given as a percentage. Open table in a new tab As seen graphically by the different slopes in Figure 1, there was a statistical interaction between serum analyte and healthy versus native kidney disease status in the prediction of GFR (P<0.001 for both cystatin C and serum creatinine). To understand the potential importance of these statistical interactions, ln GFR was modeled separately in the healthy group and the native kidney disease group (Table 3). In the healthy group, a difference in cystatin C or serum creatinine was less strongly associated with a difference in GFR compared to that in the native kidney disease group. Furthermore, in the healthy group compared to the native kidney disease group, age and gender were stronger predictors of GFR in a cystatin C model, but weaker predictors of GFR in a serum creatinine model. For both cystatin C and serum creatinine models, the root mean square error was approximately half for the healthy group compared to the native kidney disease group. Although subtle, there was also a statistical interaction between cystatin C and transplant recipient versus native kidney disease status in the prediction of GFR (P=0.02). A 50% increase in cystatin C was associated with a 37.1% decrease in GFR for the transplant recipient group compared to a 40.3% decrease in GFR for the native kidney disease group.Table 3Effect of predictor variables on GFR in cystatin C models compared to serum creatinine models by clinical presentation (native kidney disease, n=204; healthy, n=50)% Difference in measured GFR (95% confidence interval)Cystatin C ModelsSerum creatinine modelsClinical presentationNative kidney diseaseHealthyNative kidney diseaseHealthyCystatin C (per 50% increase)-40.3 (-42.1 to -38.4)-20.2 (-26.0 to -14.0)——Serum creatinine (per 50% increase)——-39.0 (-41.0 to -36.9)-18.4 (-29.3 to -5.8)Age (per 50% increase)-4.3 (-8.5 to 0.2)-8.0 (-12.8 to -3.1)-11.4 (-15.6 to -7.0)-6.6 (-12.5 to -0.3)MaleRefRefRefRefFemale1.4 (-5.7 to 9.1)-4.9 (-12.0 to 2.8)-26.2 (-32.1 to -19.7)-10.9 (-21.8 to 1.6)Model fit (r2)aModel fit or coefficient of determination (r2) is the proportion of between patient variability in measured GFR explained by the model.0.8560.4930.8270.237Root mean square error (%)bRoot mean square error is an average error between measured GFR and estimated GFR given as a percentage.29.013.132.816.3a Model fit or coefficient of determination (r2) is the proportion of between patient variability in measured GFR explained by the model.b Root mean square error is an average error between measured GFR and estimated GFR given as a percentage. Open table in a new tab As shown in Figure 1a and Tables 2 and 3, the relationship between cystatin C and GFR varied across clinical presentation. Thus, we developed cystatin C and serum creatinine equations using only the native kidney disease group (equations (1) and (2), n=204). For the cystatin C model, gender was not statistically significant (P=0.71) and age was only borderline significant (P=0.06). Including these variables only increased the r2 by 0.003. Thus, the final cystatin C equation did not include age or gender. A separate cystatin C equation was also developed for transplant recipients (equation (3), n=206). For convenience, Table 4 provides the corresponding cystatin C levels for each stage of CKD.1.Levey A.S. Coresh J. Balk E. et al.National kidney foundation practice guidelines for chronic kidney disease: evaluation, classification, and stratification.Ann Intern Med. 2003; 139: 137-147Crossref PubMed Scopus (3462) Google Scholar Native CKDGFR=66.8×(cystatinC)−1.30(1) Native CKDGFR=273×(serum creatinine)−1.22×age−0.299×0.738(if female)(2) Transplant recipientGFR=76.6×(cystatinC)−1.16(3) The correlation of the cystatin C equation (equation (1), r2=0.853) was higher than the serum creatinine equation derived using the same sample (equation (2), r2=0.827), although this did not reach statistical significance (P=0.15).15.Meng X. Rosenthal R. Rubin D. Comparing correlated correlation coefficients.Psychol Bull. 1992; 111: 172-175Crossref Scopus (1809) Google Scholar It was also higher than the MDRD equation (r2=0.825) and the Cockcroft–Gault equation (r2=0.796). The cystatin C equation also appeared to perform slightly better than a serum creatinine equation when applied to independent datasets approximated with bootstrapping. The r2 adjusted for optimism was 0.852 for the cystatin C equation (equation (1)) and was 0.821 for the serum creatinine equation (equation (2)). The cystatin C equation derived with transplant recipients (equation (3)) had an r2 of 0.768 and an r2 adjusted for optimism of 0.766.Table 4Cystatin C levels for determining stage of CKDCystatin C levelStageaGFR estimates and CKD stage will be inaccurate if there is a calibration difference with the Dade-Behring BN II Nephelometer assay used in this study.DescriptionGFR rangeaGFR estimates and CKD stage will be inaccurate if there is a calibration difference with the Dade-Behring BN II Nephelometer assay used in this study. (ml/min/1.73 m2)Native kidney diseasebUsing the prediction equation: GFR=66.8 (cystatin C)-1.30.Transplant recipientcUsing the prediction equation: GFR=76.6 (cystatin C)-1.16.1Normal or increased GFR≥90≤0.80≤0.872Mildly decreased GFR60 to 890.80 to 1.090.87 to 1.233Moderately decreased GFR30 to 591.10 to 1.861.24 to 2.244Severely decreased GFR15 to 291.87 to 3.172.25 to 4.105Kidney Failure<15>3.17>4.10a GFR estimates and CKD stage will be inaccurate if there is a calibration difference with the Dade-Behring BN II Nephelometer assay used in this study.b Using the prediction equation: GFR=66.8 (cystatin C)-1.30.c Using the prediction equation: GFR=76.6 (cystatin C)-1.16. Open table in a new tab A composite equation for the native kidney disease group based on cystatin C, serum creatinine, age and gender had the best model fit (r2=0.892, equation not shown). This was significantly higher than the serum creatinine equation (P<0.0001).15.Meng X. Rosenthal R. Rubin D. Comparing correlated correlation coefficients.Psychol Bull. 1992; 111: 172-175Crossref Scopus (1809) Google Scholar However, a similar model fit (r2=0.891) was obtained regressing measured GFR on the mean of the cystatin C equation (equation (1)) and the serum creatinine equation (equation (2)). In other words, when both cystatin C and serum creatinine are available, averaging the GFR estimates by each analyte improved correlation with measured GFR. As equation models predicted logarithmic GFR, the geometric mean was used:Composite estimated GFR(eGFR)=(cystatinCeGFR)×(serum creatinineeGFR)(4) This study characterized the relationship between cystatin C and measured GFR (iothalamate clearance) in a variety of clinical presentations. At the same cystatin C level, transplant recipients had a 19% higher GFR than native kidney disease patients. As expected, there was a much stronger association between cystatin C and GFR among native kidney disease patients (r2=0.853) than among healthy persons (r2=0.382). Because clinical presentation was an important predictor of GFR, a cystatin C equation was derived with native kidney disease patients only. Averaging the estimated GFR between a cystatin C equation and a serum creatinine equation improved the prediction of GFR over a serum creatinine equation alone in CKD patients. These findings help clarify the relationship between cystatin C and GFR as compared to serum creatinine and GFR. The relationship between either serum analyte with GFR differed among clinical presentations. This, in part, may explain the discrepancy among prior studies that compared cystatin C with serum creatinine.16.Laterza O.F. Price C.P. Scott M.G. Cystatin C: an improved estimator of glomerular filtration rate?.Clin Chem. 2002; 48: 699-707PubMed Google Scholar For this study, equations were only developed for patients with a clinical diagnosis of CKD. A diagnosis of CKD was based on an elevated serum creatinine level, other evidence of kidney damage (e.g., proteinuria) or a clinical presentation where most patients have reduction in GFR (transplant recipient). GFR can be estimated with a cystatin C equation for native kidney disease patients (equation (1)) or with a cystatin C equation for transplant recipients (equation (3)) as well as with the MDRD equation. Improved prediction may be obtained by averaging (geometric mean) the estimated GFR from the appropriate cystatin C equation with the MDRD equation. When clinical presentation narrows the GFR distribution, this will decrease modeling error in the GFR estimate. This was demonstrated by the lower root mean square error for the healthy group compared to that for the native kidney disease groups (Table 3). Thus, it is important to have a diagnosis of CKD before applying these equations to estimate GFR. Any GFR estimated by a serum analyte should still be interpreted with caution. Confounding factors associated with cystatin C (possibly inflammation or immunosuppression therapy) or serum creatinine (muscle mass or protein intake) may lead to inaccurate GFR estimates. This inherent limitation from factors that influence serum analyte levels independent of GFR is illustrated in Figure 2. With prediction equations, several analyte factors (production, tubular secretion, tubular reabsorption and extra-renal clearance) are not directly measured. Instead they are assumed to be constant or are modeled based on easily measured characteristics (e.g. age and gender) of the population used to derive the equation. Thus, an equation will not be accurate if the modeled effect of these analyte factors based on the equation population is incorrect for a particular patient (Figure 2). For example, applying an equation derived in native kidney disease patients to a liver graft recipient will overestimate GFR by serum creatinine but underestimate GFR by cystatin C (Table 2). For serum creatinine, liver disease among liver graft recipients will decrease creatinine production leading to an overestimation of GFR.17.Lam N.P. Sperelakis R. Kuk J. et al.Rapid estimation of creatinine clearances in patients with liver dysfunction.Dig Dis Sci. 1999; 44: 1222-1227Crossref PubMed Scopus (10) Google Scholar For cystatin C, the inflammation or immunosuppression therapy associated with having a liver graft may increase cystatin C production leading to an underestimation of GFR. The methods used in this study were chosen to minimize bias in the comparison of two different serum analytes. Sampling or measurement bias can occur when applying previously derived statistical models (e.g., MDRD equation and Cockcroft–Gault equation) to new data. Thus, comparisons were made between statistical models specifically derived for this study that varied either the serum analyte or the sampled population. This improved internal validity for the comparison between cystatin C and serum creatinine (as measured in this study) for predicting GFR (as measured in this study) among different populations (as sampled in this study). Finally, samples were not identified by measured GFR, because measured GFR was the dependent variable in the regression models. The results of this study are consistent with previously published studies. The 19% higher GFR at the same cystatin C level among transplant recipients compared to native kidney disease patients is consistent with reports by other investigators.18.Bokenkamp A. Domanetzki M. Zinck R. et al.Cystatin C serum concentrations underestimate glomerular filtration rate in renal transplant recipients.Clin Chem. 1999; 45: 1866-1868PubMed Google Scholar, 19.Hermida J. Romero R. Tutor J.C. Relationship between serum cystatin C and creatinine in kidney and liver transplant patients.Clin Chim Acta. 2002; 316: 165-170Crossref PubMed Scopus (43) Google Scholar The mechanism for this finding is not understood, but plausible hypotheses include increased cystatin C production from systemic inflammation12.Knight E.L. Verhave J.C. Spiegelman D. et al.Factors influencing serum cystatin C levels other than renal function and the impact on renal function measurement.Kidney Int. 2004; 65: 1416-1421Abstract Full Text Full Text PDF PubMed Scopus (720) Google Scholar, 13.Koenig W. Twardella D. Brenner H. Rothenbacher D. Plasma concentrations of cystatin C in patients with coronary heart disease and risk for secondary cardiovascular events: more than simply a marker of glomerular filtration rate.Clin Chem. 2005; 51: 321-327Crossref PubMed Scopus (241) Google Scholar, 20.Curhan G. Cystatin C: a marker of renal function or something more?.Clin Chem. 2005; 51: 293-294Crossref PubMed Scopus (57) Google Scholar or use of immunosuppresion therapy21.Bjarnadottir M. Grubb A. Olafsson I. Promoter-mediated, dexamethasone-induced increase in cystatin C production by HeLa cells.Scand J Clin Lab Invest. 1995; 55: 617-623Crossref PubMed Scopus (127) Google Scholar, 22.Risch L. Herklotz R. Blumberg A. Huber A.R. Effects of glucocorticoid immunosuppression on serum cystatin C concentrations in renal transplant patients.Clin Chem. 2001; 47: 2055-2059PubMed Google Scholar among transplant recipients. In this study, the regression fit between ln GFR and ln cystatin C was stronger than between ln GFR and ln serum creatinine for each clinical presentation. This was consistent with a meta-analysis that found cystatin C to be either superior or equivalent to serum creatinine in the correlation with GFR.16.Laterza O.F. Price C.P. Scott M.G. Cystatin C: an improved estimator of glomerular filtration rate?.Clin Chem. 2002; 48: 699-707PubMed Google Scholar Another study also found that a cystatin C (Dade–Behring assay) equation (r2=0.91) had a better model fit than a serum creatinine equation (r2=0.84), but effects of clinical presentation, body size, age and sex were not considered in the comparison. The equation for that study, GFR (ml/min)=77.2 × cystatin C-1.26, gives GFR estimates 15–20% higher than the native kidney disease equation (equation (1)) of this study.23.Larsson A. Malm J. Grubb A. Hansson L.O. Calculation of glomerular filtration rate expressed in mL/min from plasma cystatin C values in mg/L.Scand J Clin Lab Invest. 2004; 64: 25-30Crossref PubMed Scopus (330) Google Scholar There were several potential limitations to this study. First, patients were classified into clinical presentations based on a medical chart review. To decrease misclassification bias, patients were only grouped by levels of easily identifiable clinical characteristics. Second, among native kidney disease patients, only those with an increased severity of illness, such that a nephrologist would measure the patient's GFR, were represented. Thus, patients with microalbuminuria and a normal serum creatinine level were inadequately represented in the native kidney disease group. Third, the generalizability of the cystatin C equation needs to be tested in other centers with more diverse racial groups and different mixtures of CKD etiologies. Finally, any calibration differences between the Dade-Behring BN II Nephelometer used in this study and other cystatin C assays can lead to inaccurate GFR estimates, a well-described problem with serum creatinine equations.24.Coresh J. Astor B.C. McQuillan G. et al.Calibration and random variation of the serum creatinine assay as critical elements of using equations to estimate glomerular filtration rate.Am J Kidney Dis. 2002; 39: 920-929Abstract Full Text Full Text PDF PubMed Scopus (600) Google Scholar In conclusion, the relationship between cystatin C and GFR can depend on the clinical presentation. In patients with CKD, a cystatin C equation is complementary to the MDRD equation or other serum creatinine equations for improving GFR estimates. Depending on the clinical setting, estimated GFR with a cystatin C equation can be averaged with a serum creatinine equation or used in place of the serum creatinine equation. For example, among patients with very high or very low muscle mass, a cystatin C equation alone may be preferential. A better understanding of the factors that effect cystatin C levels independent of GFR could potentially improve Cystatin C equations and determine the best settings for their application. A previously reported series of consecutive patients (n=502) had an outpatient GFR measurement by iothalamate clearance, between 27 October 1999 and 3 March 2000 and agreed to participate in the study.3.Rule A.D. Larson T.S. Bergstralh E.J. et al.Using serum creatinine to estimate glomerular filtration rate: accuracy in good health and in chronic kidney disease.Ann Intern Med. 2004; 141: 929-937Crossref PubMed Scopus (876) Google Scholar Medical records were abstracted for demographic and clinical characteristics present at the time of the iothalamate clearance. Clinical presentation was grouped into three mutually exclusive categories: native CKD, solid-organ transplant recipient (with or without known CKD) and healthy (potential kidney donor). Among the native kidney disease patients, the suspected etiology was hypertension or unknown (36%), diabetes mellitus (13%), other glomerulopathy (26%) and other non-glomerulopathy (25%). Transplant recipients were further divided into those with a kidney graft alone and those with other organ grafts (liver, heart, pancreas or lung) with or without a kidney graft. Transplant recipients typically have GFR measurements during routine outpatient follow-up when serum creatinine levels are stable. Patients who did not fit into the above categories were excluded due to inadequate sample size and heterogeneity (n=32). This included protocol evaluation for potential non-kidney transplant recipients (pancreas, n=5; heart, n=6; or lung, n=9), spinal cord injury patients with neurogenic bladders being screened for kidney disease (n=8), a cancer patient who needed chemotherapy dosing (n=1), and reason for GFR measurement indeterminate (n=3). Children (ages <17 years) were also excluded (n=10). All patients had a non-radiolabeled iothalamate clearance using a previously described standard laboratory method.25.Wilson D.M. Bergert J.H. Larson T.S. Liedtke R.R. GFR determined by nonradiolabeled iothalamate using capillary electrophoresis.Am J Kidney Dis. 1997; 30: 646-652Abstract Full Text PDF PubMed Scopus (113) Google Scholar Briefly, after oral hydration with 4–6 glasses of water, patients received a subcutaneous injection of non-radiolabeled iothalamate (Conray, Mallinckrodt Medical, St Louis, MO). Following a 1-h equilibrium period, the patient voided, the first serum sample was drawn and a timed urine collection was begun. A sonographic scanner assessed bladder emptying and a bladder catheter was placed in patients with urinary retention. Following the timed urine collection (approximately 45–60 min), a second serum sample was obtained. GFR was calculated by the clearance equation (UIoV/PIo where UIo and PIo are the urine and plasma concentrations of iothalamate, and V is the urine flow) using the mean of two serum samples and one urine sample assayed for iothalamate via capillary electrophoresis. The between-day assay coefficient of variation was 4.3%. All GFR measurements were standardized for body surface area (per 1.73 m2) by multiplying by 1.73 and dividing by body surface area.26.Rule A.D. Gussak H.M. Pond G.R. et al.Measured and estimated GFR in healthy potential kidney donors.Am J Kidney Dis. 2004; 43: 112-119Abstract Full Text Full Text PDF PubMed Scopus (301) Google Scholar, 27.DuBois D. DuBois E. A formula to estimate the approximate surface area if height and weight be known.Arch Intern Med. 1916; 17: 863-871Crossref Scopus (3867) Google Scholar Cystatin C and creatinine levels were assayed from the first serum sample obtained during the iothalamate clearance test. Cystatin C was assayed by particle-enhanced immuno-nephelometry (Dade-Behring BN II Nephelometer). The between-day assay coefficient of variation was 3.5%. The assay showed linearity based on serial dilutions (measured=0.996 × expected+0.015; n=16) over a range of 0.16 to 2.61 mg/l. Samples showed stability at room temperature (up to 7 days), when frozen at -20°C (up to 2 years) and through three freeze–thaw cycles. Serum creatinine was measured by the uncompensated rate-Jaffe reaction (Hitachi 911 auto-analyzer). The between-day assay coefficient of variation was 4.7% at 1.1 mg/dl and 1.8% at 5.6 mg/dl. GFR was estimated using the abbreviated MDRD equation14.Levey A.S. Bosch J.P. Lewis J.B. et al.A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group.Ann Intern Med. 1999; 130: 461-470Crossref PubMed Scopus (12277) Google Scholar, 28.Levey A.S. Greene T. Kusek J.W. Beck G.J. A simplified equation to predict glomerular filtration rate from serum creatinine.J Am Soc Nephrol. 2000; 11: A0828PubMed Google Scholar and the Cockcroft–Gault equation (standardized for body surface area and adjusted to predict GFR in the same native kidney disease population used to derive the MDRD equation).2.Cockcroft D.W. Gault M.H. Prediction of creatinine clearance from serum creatinine.Nephron. 1976; 16: 31-41Crossref PubMed Scopus (12499) Google Scholar, 14.Levey A.S. Bosch J.P. Lewis J.B. et al.A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group.Ann Intern Med. 1999; 130: 461-470Crossref PubMed Scopus (12277) Google Scholar The original Cockcroft–Gault equation predicts creatinine clearance and because of tubular creatinine secretion, it overestimates GFR in CKD patients. Univariate statistics (frequency, mean, standard deviation) were assessed overall and stratified by clinical presentation. For each clinical presentation group, ln GFR was regressed on ln analyte (cystatin C or serum creatinine). GFR was ln transformed for constant variability. Cystatin C and serum creatinine levels were ln transformed for linearity within groups. Multiple linear regression was used to predict ln GFR with the covariates: ln analyte, ln age, female (0–1 indicator variable) and clinical presentation (0–1 indicator variables with native kidney disease as the reference group). Age was ln transformed to allow easier comparison with the MDRD equation, but findings were similar when age was not ln transformed. Height and weight were not included as predictors in models, because GFR had been standardized for body surface area using a formula based on height and weight.27.DuBois D. DuBois E. A formula to estimate the approximate surface area if height and weight be known.Arch Intern Med. 1916; 17: 863-871Crossref Scopus (3867) Google Scholar Interaction terms between the analyte and clinical presentation were included in additional models. As the relationship between cystatin C and GFR differed among clinical presentations, a cystatin C equation was derived using only patients with native kidney disease. The model fit (r2) of this equation was compared to a serum creatinine equation derived with the same data, the MDRD equation and the Cockcroft–Gault equation. The root mean square error between different models was compared. Statistical significance for the difference in r2 between equations was determined using the method of Meng et al.15.Meng X. Rosenthal R. Rubin D. Comparing correlated correlation coefficients.Psychol Bull. 1992; 111: 172-175Crossref Scopus (1809) Google Scholar An additional cystatin C equation was developed for transplant recipients. To assess performance on independent data sets, new equations were internally validated with bootstrapping.29.Harrell F.E. Regression Modelling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analyses. Springer-Verlag, New York2001Crossref Google Scholar Each equation was evaluated using 500 random bootstrap samples from the full set of data used to derive the equation. Stepwise selection was used to add the serum analyte, age and gender to each model. Model optimism (including stepwise selection) was determined from the mean difference between the r2 of each bootstrapped sample and the r2 when applying that bootstrapped sample's parameters to the original data set. r2 adjusted for optimism was determined from the original data r2 minus model optimism. Statistics were performed with JMP 5.1 and with SAS 8.2 (SAS Institute, Cary, NC). We thank Steven J Jacobsen MD, PhD and Stephen T Turner, MD for advice on the data analysis and Mary F Burritt, PhD for advice on the laboratory methods. This work was supported by a National Research Service Award (F32 DK068996) and grants from the Division of Nephrology, Mayo Clinic.