Surveillance of glomerular filtration rate (GFR) is crucial in the management of kidney transplant recipients. With especial emphasis on serum creatinine (SCr) calibration assay, we assessed the performance of estimation equations as compared to iothalamate GFR (iGFR) in 209 patients using the modification of diet in renal disease (MDRD), Nankivell and Cockcroft-Gault methods. Fifty-five percent of patients were treated with a calcineurin inhibitor (CNI) and all were taken trimethroprim-sulfametoxazole at the time of SCr measurement. The mean iGFR was 44 ± 26 mL/min/1.73 m^{2}. The MDRD equation showed a median difference of 0.9 mL/min/1.73 m^{2} with 53% of estimated GFR within 20% of iGFR. Median differences were 7.5 and 7.0 mL/min/1.73 m^{2} for Nankivell and Cockcroft-Gault formulas, respectively. The accuracy of the Nankivell and Cockcroft-Gault formulas was such that only 38% and 37% of estimations, respectively, fell within 20% of iGFR. The performance of all equations was not uniform throughout the whole range of GFR, with some deterioration at the extremes of GFR levels. In addition, good performance of the MDRD equation was seen in subjects taking CNI. In conclusion, the overall performance of the MDRD equation was superior to the Nankivell and Cockcroft-Gault formulas in renal transplant recipients including subjects treated with CNI.

Management of kidney transplant recipients requires a simple, reliable and accurate method for the estimation of glomerular filtration rate (GFR). Using solely serum creatinine (SCr) to estimate GFR is the simplest and most commonly used approach; although when compared to creatinine clearances and isotope measurements of GFR, its performance is quite variable particularly in renal transplant patients (1–6). Creatinine clearance is also an inaccurate estimate of GFR especially at the extremes of age and GFR (7). In contrast, the clearance of some radioisotopes like ^{125}I-iothalamate (iGFR) demonstrated accurate measurement of GFR in a non-transplant population when compared to the ‘gold standard’ inulin clearance (8). However, the expense and complexity of this tool limit its wide application.

In order to accurately and simply estimate GFR, different creatinine-based estimation equations have been developed mainly in the non-transplant population. The most widely used methods are Cockcroft-Gault (CG) (9), 4-variable modification of diet in renal disease (MDRD) (10,11) and, in kidney transplant recipients, the Nankivell formula (12). Special emphasis needs to be placed on SCr calibration bias because it is a determining factor in these equations. It has been clearly demonstrated that the assessment of the performance of these equations without careful SCr calibration could greatly impair the clinician's ability to interpret the results (13,14). A few European studies tested the applicability of these formulas to a kidney transplant population; however, these analyses were done in the absence of rigorous assessment of SCr measurement calibration bias (5,15–19).

The purpose of this study was to evaluate the performance of the four-variable MDRD, Nankivell and Cockcroft-Gault equations as compared to measurements of GFR by ^{125}I-iothalamate renal clearance in a US cohort of renal transplant recipients that includes African-American subjects with stable graft function and to provide insight into the mechanisms involved in GFR estimation. Special emphasis was placed on calibration of the SCr assay. We showed that the MDRD equation overall performs better with respect to bias and accuracy than the Nankivell and Cockcroft-Gault formulas in this study population. Moreover, the performance of these equations is not affected when applied to individuals treated with a calcineurin inhibitor (CNI).

Materials and Methods

The Renal Function Laboratory at the Cleveland Clinic Foundation (CCF) performed approximately 9000 measurements of GFR by ^{125}I-iothalamate clearance from 1982 to 2002 and maintained a database with demographic as well as laboratory variables. A retrospective data base review was performed and data were retrieved after approval by the Institutional Review Board at CCF. All stable outpatients older than 18 years, with a functioning kidney transplant were considered for analysis. Only subjects who had an iGFR test between January 1996 and December 2002 were analyzed because SCr calibration data were available for this time period only. A computerized database was reviewed and 10 patients were removed from the analysis because of acute renal failure at the time of the iGFR. Stability of graft function was defined as <15% change in SCr over at least a month period. For patients who had more than one iGFR test, the first measurement was used for analysis. A total of 209 individuals were included in this analysis. The primary diagnoses for renal failure were as follows: diabetes mellitus (n= 30), hypertension (n= 15), glomerular disease (n= 84), autosomal dominant polycystic kidney disease (n= 24), interstitial disease/pyelonephritis (n= 17) and other/unknown (n= 39). The immunosuppressive regimen at the time of the iGFR consisted of: cyclosporine (n= 106), tacrolimus (n= 9), mycophenolic acid (n= 59), azathioprine (n= 91) and sirolimus (n= 18). Calcineurin inhibitor-based regimen (CNI-treated subgroup) was used in 115 patients (55%) while the rest were treated with a CNI-free regimen (CNI-free subgroup). All patients were also receiving prednisone and trimethroprim-sulfamethoxazole (80 mg/400 mg orally per day), while none of the patients was taking cimetidine at the time of the SCr measurement.

The GFR was measured using a modified renal clearance of ^{125}I iothalamate as described by Israelit et al. (20). These measurements were done by the same laboratory that performed the measurements for the MDRD study using a very similar technique. A detailed description of the procedure has been previously reported (14). In brief, after hydration and blockage of the thyroid uptake of ^{125}I-iothalamate, 25 μCuries (Glofil, Questor Pharmaceuticals, Inc., Union City, CA) of radioisotope were injected subcutaneously. Baseline urine and blood samples were obtained followed by two time clearance urine collections, which were used for analysis. Blood samples were also obtained at the same time. Isotope activity was determined by gamma counting of 0.5 mL samples on a Packard Minaxi 5000 series counter (Perkin Elmer Life Sciences, Downers Grove, IL). The serum counts were taken as the average of the bracketed blood samples for each clearance period. The mean GFR was calculated from two consecutive values. Mean GFR results were then corrected to standard body surface area (1.73 m^{2}).

Blood samples obtained simultaneously with the iGFR were used to measure SCr by the modified kinetic Jaffe reaction, using a Hitachi 747–200 Chemistry Analyzer (1996–2001) or a Hitachi D 2400 Modular Chemistry Analyzer thereafter (Nakakojo, Japan). SCr measurements at CCF were calibrated using College of American Pathology (CAP) samples to ensure that large calibration bias was absent. Paired SCr measurements by the MDRD and CCF laboratories were compared for 89 CAP samples obtained at 14 different time points between 1996 and 2002 (Figure 1). Values falling within ± 0.3 mg/dL or 15% of the mean SCr are considered by the CAP to be within an acceptable range (13). Overall, the mean (±SE) SCr for the 89 paired CAP specimens was similar between the MDRD and the CCF laboratories (MDRD–CCF mean difference ±SE = 0.04 ± 0.02 mg/dL, p = 0.12), suggesting that large calibration bias was unlikely to affect the interpretation of the eGFR results in this study. However, among 28 CAP specimens in which the average SCr for the two laboratories was <2 mg/dL, the mean SCr was significantly higher for the MDRD laboratory (0.09 ± 0.03 mg/dL, p = 0.006), suggesting the possibility of a limited calibration bias at lower SCr levels. Nevertheless, because of the relatively low GFR values of this cohort of patients, the effects of limited calibration bias would not be expected to have a big impact on the results.

eGFRs were calculated using the following equations, and results were then adjusted for BSA for the Nankivell and Cockcroft-Gault formulas (mL/min/1.73 m^{2}):

a. For males: eGFR_{CG}=[(140 – age) × weight (kg)]/SCr × 72

b. For females: eGFR_{CG}= ([(140 – age) × weight (kg)]/SCr × 72) ×0.85

Statistical analysis

Population characteristics and subgroup comparisons were studied by using two-sample t-test or Wilcoxon Rank Sum test as appropriate. The agreement of eGFRs (eGFR_{MDRD}, eGFR_{NK}, eGFR_{CG}) with iGFR was evaluated graphically by plotting each eGFR against iGFR with the 45° line indicating perfect agreement, and by the use of residual plots. Precision was evaluated by using the Pearson correlation (Pearson R) after log-transformation of the data. Bias, a measure of systematic error, was assessed by both median difference and median percent difference (eGFR-iGFR). Overall agreement was evaluated by median absolute difference, median percent absolute difference and percent of eGFR values falling within 20% and 50% of iGFR. We compared the agreement of eGFR_{NK} and eGFR_{CG} with iGFR to that of eGFR_{MDRD} with iGFR by using the McNemar test for percentage of eGFR values within 20% or 50% of iGFR and the bootstrap method (using 400 independent replications) for other indices.

The relationship of iGFR with the terms included as predictor variables in the three equations was evaluated by performing multiple regression to relate log transformed iGFR to log SCr, log age, log weight, 0–1 indicator variables for African-American race and female gender. To clarify the interpretation, the estimated effects of SCr, age and weight in these regression models were expressed as the percentage change in iGFR associated with a 10% increase in each of these factors. Similar regression analyses were performed to relate the eGFR_{MDRD}, eGFR_{NK} and eGFR_{CG} to the same set of predictor variables. The estimated effects of each predictor variable on iGFR were compared to the corresponding estimates of the effects on eGFR_{MDRD}, eGFR_{NK} and eGFR_{CG} to determine whether the relationships of iGFR with SCr, race, gender, age and weight agreed with relationships predicted by the MDRD, Cockcroft-Gault and Nankivell equations. Multiple regression analyses were used to relate the mean bias of eGFR to the use of CNI with and without adjustment for the inverse of SCr (1/SCr) in addition to other terms included as predictor variables in the estimation equations (age, gender, race and weight and time from transplantation to measurement of GFR).

Results

Patient characteristics

Table 1 summarizes the population characteristics of the study group. The mean SCr for the entire population was 2.4 ± 1.8 mg/dL with a mean iGFR of 44 ± 26 mL/min/1.73 m^{2}. African-American patients represented 11.5% of the entire cohort. Table 1 also shows the characteristics of the CNI-treated and CNI-free subgroups. As expected, the CNI-treated subgroup had a higher SCr (2.8 ± 1.8 vs. 1.8 ± 1.8 mg/dL, p < 0.001) than the subgroup not exposed to CNI. This higher SCr was reflected in a lower iGFR for the CNI-treated patients (34 ± 21 vs. 56 ± 26 mg/dL, p < 0.001). Other indices of estimated GFR are shown in Table 1.

Table 1. Characteristics of study population

All transplant patients, n= 209

CNI-treated subgroup, n= 115

CNI-free subgroup, n= 94

p-Value*

All data are expressed as n (%) or mean ±SD (10th percentile, median, 90th percentile).

*The p-values correspond to comparisons between CNI-based versus CNI-free subgroups.

Age (years)

48 ± 12 (30, 49, 65)

47 ± 14 (27, 48, 65)

49 ± 11 (37,49, 62)

0.548

Female gender

91 (43.5)

60 (52.2)

31 (33)

0.005

Non-African-American race

185 (88.5)

97 (84.3)

88 (93.6)

0.040

Weight (kg)

80 ± 21 (41, 77, 107)

79 ± 21 (51,75,104)

82 ± 21 (56, 79, 108)

0.443

Height (cm)

168 ± 11(124, 168, 182)

168 ± 10 (154, 167, 181)

168 ± 11 (154, 169, 182)

0.742

Body surface area (m^{2})

1.89 ± 0.27 (1.55, 1.87, 2.26)

1.88 ± 0.27 (1.51, 1.86, 2.21)

1.91 ± 0.28 (1.57, 188, 2.26)

0.453

Time from transplant to iGFR (years)

12 ± 9 (2, 9, 26)

8 ± 7 (2, 5, 17)

17 ± 10 (2, 20, 27)

<0.001

Serum creatinine (mg/dL)

2.4 ± 1.8 (0.9, 1.6, 4.8)

2.8 ±1.8 (1.2, 2.1, 5.3)

1.8 ±1.8 (0.8, 1.3, 3.3)

<0.001

Iothalamate GFR (mL/min/1.73 m^{2})

44 ± 26 (12, 42, 80)

34 ± 21 (11, 31, 65)

56 ± 26 (20, 54, 94)

<0.001

eGFR_{MDRD} (mL/min/1.73 m^{2})

47 ± 29 (13, 43, 81)

34 ± 19 (11, 31, 61)

62 ± 32 (22, 61, 96)

<0.001

eGFR_{NK} (mL/min/1.73 m^{2})

52 ± 30 (15, 53, 87)

39 ± 24 (9, 41, 66)

68 ± 30 (32, 66, 101)

<0.001

eGFR_{CG} (mL/min/1.73 m^{2})

54 ± 31 (17, 51, 94)

41 ± 22 (15, 41, 70)

71 ± 33 (29, 66, 110)

<0.001

Association between estimated GFR and iothalamate GFR

Figure 2 depicts the association of estimated GFRs and measured iGFRs. The overall precision, bias and agreement of estimation equations with respect to iGFR are shown in Table 2. All three formulas demonstrated a high level of association with iGFR. Bias, as calculated by either median difference or median percent difference was significantly smaller (p < 0.001) when comparing the MDRD equation to the Nankivell and Cockcroft-Gault formulas, but was not significantly different between the two latter equations. There was also a better agreement for the MDRD equation (53% of eGFR fell within 20% of the iGFR and 90% were within 50%), as compared to both the Nankivell and the Cockcroft-Gault formulas (p < 0.05 for both formulas against MDRD). Table 2 also shows the performance of these equations in CNI-treated and CNI-free subgroups. Again, the MDRD equation outperformed the other two methods in both subgroups (p < 0.05 for both formulas against MDRD). The mean bias of each of the three estimation equations appeared to be significantly smaller in patients receiving CNI as part of their immunosuppression regimen after controlling for age, gender, race, weight and time from transplantation to iGFR measurement. However, this difference was no longer present after adjusting for the degree of renal function by entering 1/SCr in the multiple regression model (data not shown). The MDRD equation was consistently better than Nankivell and Cockcroft-Gault formulas at all GFR levels (Table 3). Residual plots relating the difference between eGFR and iGFR to the patient estimated level of renal function based on eGFR by different methods are shown in Figure 3 (21). It is notable that the MDRD equation was almost unbiased at lower GFR levels but appeared to slightly overestimate GFR at progressively higher GFR levels (Figure 3A). On the other hand, the Nankivell and Cockcroft-Gault formulas overestimated the iGFR even at the low levels of GFR (Figures 3B and 3C).

Table 2. Precision, bias and agreement of MDRD, Nankivell and Cockcroft-Gault equations versus measured iGFR

Pearson R (in ln scale)^{†}

Mean difference*

Median difference*

Median % difference

Median absolute difference^{a}

Median % absolute difference

Agreement within

20%

50%

*Data expressed as mL/min/1.73 m^{2}.

^{†}Five patients with a negative estimated GFR obtained by the Nankivell formula were excluded from the Pearson correlation analyses as negative values cannot be log-transformed.

^{1}p < 0.05 comparing Cockcroft-Gault or Nankivell with MDRD equation.

^{2}p < 0.001 comparing Cockcroft-Gault or Nankivell with MDRD equations.

All Patients (n= 209)

eGFR_{MDRD}

0.91

2.6

0.9

3%

6.5

19%

53%

90%

eGFR_{NK}

0.88^{1}

8.1^{2}

7.5^{2}

21%^{2}

9.4^{2}

28%^{2}

38%^{1}

75%^{2}

eGFR_{CG}

0.89^{1}

10.2^{2}

7.0^{2}

25%^{2}

9.4^{2}

27%^{2}

37%^{1}

75%^{2}

CNI-treated regimen (n= 115)

eGFR_{MDRD}

0.91

−0.6

−1.2

−5%

4.8

17%

57%

93%

eGFR_{NK}

0.84^{1}

4.6^{2}

4.6^{2}

14%^{2}

7.6^{1}

28%^{1}

39%^{1}

72%^{2}

eGFR_{CG}

0.90

6.5^{2}

6.1^{2}

25%^{2}

6.6^{1}

26%^{1}

37%^{1}

77%^{2}

CNI-free regimen (n= 94)

eGFR_{MDRD}

0.87

6.5

4.5

9%

9.5

21%

50%

86%

eGFR_{NK}

0.89

12.3^{2}

11.8^{2}

26%^{2}

13.8^{1}

28%^{1}

37%^{1}

79%

eGFR_{CG}

0.84

14.8^{2}

11.3^{2}

26%^{2}

12.8^{2}

29%^{2}

37%^{1}

72%^{1}

Table 3. Performance of the MDRD, Nankivell and Cockcroft-Gault equations at different levels of GFR and by CI-based or CI-free regimen subgroups

Pearson R (in log scale)^{†}

Median Δ*

Median % Δ^{1}

Median absolute Δ*

Median absolute % Δ^{1}

Accuracy within

20%^{1}

*Expressed as mL/min/1.73 m^{2}.

^{†}Five patients with a negative estimated GFR obtained by the Nankivell formula were excluded from the Pearson correlation analyses as negative values cannot be log-transformed.

^{1}Expressed as %.^{2}The GFR ranges were defined using the MDRD equation.

eGFR_{MDRD} (mL/min/1.73 m^{2})^{2}

<30 (n = 70)

0.74

−0.7

−4%

3.3

24%

43%

90%

30–60 (n = 77)

0.62

−2.1

−5%

7.0

14%

65%

92%

>60 (n = 62)

0.42

9.5

12%

14.4

20%

50%

87%

eGFR_{Nankivell} (mL/min/1.73 m^{2})^{2}

<30 (n = 70)

0.64

3.4

18%

7.4

39%

29%

61%

30–60 (n = 77)

0.53

7.5

18%

10.2

23%

48%

83%

>60 (n = 62)

0.39

16.5

25%

21.4

28%

37%

81%

eGFR_{CG} (mL/min/1.73 m^{2})^{2}

<30 (n = 70)

0.71

5.9

35%

6.4

37%

23%

64%

30–60 (n = 77)

0.50

6.5

15%

10.1

21%

49%

83%

>60 (n = 62)

0.40

16.7

25%

25.5

26%

39%

77%

Effect of changes on predictor variables on mean iothalamate GFR and estimated GFR

The relationships of iGFR and eGFR (MDRD, Nankivell and Cockcroft-Gault) to SCr, race, gender, age and weight based on multiple regression analysis are shown in Table 4. Because the regression models use the same logarithmic forms as the original MDRD equation, the estimated effects of each factor on the eGFR_{MDRD} are exact and correspond directly to the coefficients of MDRD equation regardless of study population (14). For a 10% decrease in SCr, all three equations appropriately adjusted for the same change in iGFR. In contrast, while the MDRD equation corrected similarly to the iGFR for a change in race (24% change in iGFR and 21% change in eGFR_{MDRD}), the Nankivell and Cockcroft-Gault formulas failed to do so. The observed 19% mean reduction in iGFR for females agreed more closely with the MDRD equation and the Nankivell formula than Cockcroft-Gault formula. The effects of female gender and African-American race on the iGFR were similarly predicted by the MDRD equation in the CNI-treated subgroup and the subgroup not exposed to CNI. Age was comparably adjusted by the MDRD and the Nankivell methods. The age effect was stronger for the Cockcroft-Gault formula. The effects of weight is of minimal impact. In general terms, the MDRD equation appears to correlate better with the CCF data than the Cockcroft-Gault and Nankivell formulas did in those patients taking CNI, effects that may be dependent on the different level of GFR of the subgroups.

Table 4. Effects of predictor variables on mean iGFR and estimation equations in multiple regression analysis

Predictor variables

% change in iGFR (95% CI)*

% change in eGFR_{MDRD}^{1, 2}

% change in eGFR_{NK}^{1}

% change in eGFR_{CG}^{1}

*Estimated changes based on CCF data.

^{1}Estimated changes based on estimation equation.

^{2}Because the regression models have the same logarithmic form as the MDRD equation, the estimated effects of each factor on the eGFR_{MDRD} are exact and correspond directly to the coefficients of that formula regardless of the study population.

All subjects (n= 209)

SCr (10% decrease)

+12 (10.8, 12.4)

+13

+11

+11

African-American race

+24 (8.3, 40.9)

+21

+4

−2

Female gender

−19 (−26.2, −10.7)

−26

−25

−10

Age (10% increase)

−2 (−3.8, −1.0)

−2

−1

−4

Weight (10% increase)

+2 (−0.2, 3.5)

0

+2

+5

CNI-treated subgroup (n= 115)

SCr (10% decrease)

+13 (11.3, 13.6)

+13

+13

+11

African-American race

+22 (4.7, 42.2)

+21

+10

−1

Female gender

−22 (−31.3, −12.0)

−26

−34

−10

Age (10% increase)

−3 (−4.8, −1.5)

−2

−1

−4

Weight (10% increase)

+1 (−1.1, 3.7)

0

+2

+5

CNI-free subgroup (n= 94)

SCr (10% decrease)

+11 (9.4, 12.1)

+13

+10

+11

African-American race

+31 (1.8, 69.3))

+21

−4

−2

Female gender

−14 (−26.7, 1.1)

−26

−9

−9

Age (10% increase)

−1 (−3.1,2.1)

−2

+1

−5

Weight (10% increase)

+2 (−0.4, 5.1)

0

+2

+5

Discussion

Rapid estimation of GFR is often assessed by the application of different creatinine-based equations that were derived mainly from populations that differ in many respects to renal transplant recipients. The applicability of these methods to estimate GFR by the growing transplant community is limited by the uncertainty of their generalization. In this study, we evaluated 209 kidney transplant recipients with stable graft function who underwent an iGFR and compared this result to the eGFR by the 4-variable MDRD, Nankivell and Cockcroft-Gault equations in order to determine which of these formulas demonstrated superior performance. In contrast with other similar studies, we purposely used the first measured iothalamate GFR and not repeated measures for the same patient to assure that the data points were independent of factors such as age, gender and race (17,18). More importantly, we also performed a careful assessment of SCr measurement calibration between the CCF and MDRD laboratories in order to minimize potential misleading result interpretations that might derive from the use of non-calibrated SCr values (13,14). In this study, the 4-variable MDRD equation demonstrated superior performance to the Nankivell and Cockcroft-Gault formulas in kidney transplant recipients from a single US center.

Each of the estimation methods has different origins that may help to explain findings in the current study population. The Cockcroft-Gault equation was derived from predominantly non-transplanted males with reasonably preserved renal function based on 24-h creatinine clearances (CrCl) (9). Although designed to calculate CrCl, Cockcroft-Gault estimated GFR with acceptable performance (22). However, CrCl is known to be inaccurate in kidney transplant recipients (1,3,6), at the lower ends of GFR, extremes of patient age, weight and BSA (7). Thus, it is unlikely that this formula would perform well when compared to non-creatinine clearances, such as iothalamate GFR. On the other hand, the Nankivell formula (derived from a white kidney transplant population who underwent ^{99m}Tc DTPA clearances) has gained popularity in the transplant community in the past few years (12). Despite using the largest number of variables (age, gender, weight, height, SCr and urea), its performance is inferior when compared with inulin clearances, iohexol or even ^{99m}Tc DTPA clearances (16–18). Lastly, the MDRD equation was developed from data obtained from a non-transplant population with chronic kidney disease, and performs very well in patients with moderate to advanced kidney failure (14,23). Several European studies have reported the performance of these equations compared to GFR measurements (inulin and iohexol clearances) in renal transplantation; however, these analyses were performed in the absence of either direct or indirect calibration of SCr values with the MDRD laboratory, a limiting factor of extreme importance that deserves focused consideration (13,14).

As demonstrated by Coresh et al. and by us (13,14), small systematic error in SCr measurement can greatly affect the results of eGFR in those situations where SCr values are within the normal range, whereas large systematic errors would affect the results of equations in those individuals with more advanced kidney dysfunction. Lack of evidence for SCr assay calibration does not imply that the results derived from a particular data analysis are incorrect, but it does preclude certainty of the validity of the reported data. In this report, large calibration bias was unlikely to have adversely influenced our results because of the overall good agreement between the CCF and the MDRD laboratory on the CAP samples. The effects of potential limited calibration bias in those individuals with SCr less than 2 mg/dL would not be expected to greatly affect the overall interpretation of the data due to the relatively low GFR of the entire population in this study. In contrast to previous studies in kidney transplant recipients, the present study applied meticulous SCr calibration assessment to avoid systematic inaccuracies of the analyzed data and the potential misinterpretation of the results.

The three equations demonstrated high and similar correlation coefficients when compared to iGFR. The correlation coefficient, an index of precision, represents the strength of the linear association between estimated and measured GFR, but it does not provide clinically applicable information on the overall agreement of these equations with iGFR. On the other hand, bias, a measure of systematic error, and agreement, as measured by accuracy, provide the clinician with better tools to assess the performance of estimation equations. The MDRD equation showed the lowest bias, while the Nankivell and Cockcroft-Gault formulas tended to significantly overestimate the measured GFR (Tables 2 and 3). Similarly, the agreement of eGFR with measured GFR was better with the MDRD equation when compared to the other two formulas. The superior performance of the MDRD equation reported in this study may be explained by the closer agreement of the iGFR data of this population with the MDRD equation. Moreover, in this study the MDRD equation is being applied to a population of individuals with a mean iGFR between 30 and 60 mL/min/1.73 m^{2}, similar to the iGFR of the original MDRD study (10), along with the fact that this method was derived from an American population (including African-American subjects), and using the same laboratory for the measurements of iothalamate GFR as the original MDRD study. Interestingly, the bias of each of the three equations were superior in those patients treated with CNI after controlling for age, gender, race, weight and time from transplantation to the measurement of iGFR. However, this effect disappeared after adjusting for serum creatinine level, and therefore, we hypothesized that the better performance seen in this subgroup is mainly due to the lower GFR seen in the CNI-treated subgroup, as the performance of patients with estimated GFR between 30 and 60 mL/min/1.73 m^{2} yielded very similar results. We would like to point out that while we can be quite confident that SCr calibration bias is not adversely influencing the interpretation of the results in the CNI-treated subgroup, we cannot completely eliminate this possibility in the CNI-free subgroup (mean SCr for this subgroup was <2.0 mg/dL, a range in which the CCF lab is not fully calibrated against the MDRD lab). More importantly, a significant limitation of this study is the simultaneous use of trimethoprim and its potential effects on serum creatinine levels, which could enter bias when these values are used for estimation of GFR using creatinine-based methods. Nevertheless, subject to the caveats presented above, the MDRD equation demonstrated the best overall performance from all tested methods in this study.

The mechanisms involved in the performance of these equations have been investigated by multivariable regression analyses. We assessed the influence of predictor factors such as SCr, age, gender, race and weight on iGFR of the CCF data versus their effects on each formula (Table 4). A 10% change in SCr in the measured GFR was appropriately adjusted for by all three formulas. African-American race was adjusted for only by the MDRD equation in contrast to Nankivell and Cockcroft-Gault formulas, which do not include a race factor, suggesting that in African-American patients the performance of the MDRD equation would be superior to the other methods. The ability of the MDRD equation to track all assessed variables explains in part why its overall performance was superior in this particular cohort of patients.

In recent European studies that used different methods to measure GFR (inulin and iohexol clearance), none of these equations were deemed suitable for use in clinical trials in which rigorous GFR assessment was needed (17–19). The population of our study had lower measured GFR, included African-American individuals, and uses iothalamate GFR as the reference method. Nevertheless, despite the different statistical methods used and demographic characteristics, we report results very similar with respect to bias and accuracy to those reported by Gaspari et al. (17), confirming the better performance of the MDRD equation over the Nankivell and Cockcroft-Gault formulas. We agree with previous reports stating that when rigorous GFR assessment is needed, isotope GFR measurement needs to be considered. However, the application of the MDRD equation should suffice for routine clinical practice, especially when applied to a population with similar characteristics to the one studied here.

In conclusion, the overall performance of the MDRD equation in this study is superior to Nankivell and Cockcroft-Gault formulas in a cohort of renal transplant recipients with a mean measured GFR of between 30 and 60 mL/min/1.73 m^{2}, as is seen in the non-transplant population (14,23). These data also suggest that the MDRD equation is applicable in patients treated with a CNI-based regimen. Until a more universally applicable equation is developed, isotope GFR measurement remains the ideal method to determine GFR in renal transplant recipients with preserved GFR, especially in clinical trials that use graft function as an ultimate surrogate marker.

Acknowledgments

The authors would like to thank Diane Pexa and Henry Rolin for their technical expertise in the performance of the iothalamate GFRs, and Sandra Bronoff for manuscript preparation. Especial thanks to Dr. Tom Greene for his statistical expertise and critical review.