C Blackwell Munksgaard 2004 Copyright
American Journal of Transplantation 2004; 4: 1826–1835 Blackwell Munksgaard
doi: 10.1111/j.1600-6143.2004.00579.x
Performance of Different Prediction Equations for Estimating Renal Function in Kidney Transplantation Flavio Gasparia , Silvia Ferraria , Nadia Stucchia , Emmanuel Centemeria , Fabiola Carraraa , Marisa Pellegrinoa , Giulia Gherardia , Eliana Gottia , Giuseppe Segolonib , Maurizio Salvadoric , Paolo Rigottid , Umberto Valentee , Donato Donatif , Silvio Sandrinig , Vito Sparacinoh , Giuseppe Remuzzia, ∗ and Norberto Perico2 on the behalf of the MY.S.S. study investigators a Department of Medicine and Transplantation, Azienda Ospedaliera, Ospedali Riuniti di Bergamo, Mario Negri Institute for Pharmacological Research, Bergamo, Italy, b Division of Nephrology, Dialysis, Transplantation, Azienda Ospedaliera S.G. Battista, Turin, Italy, c Unit of Nephrology and Dialysis, Azienda Ospedaliera Careggi Monna Tessa, Florence, Italy, d II Institute of General Surgery, Ospedale Giustinianeo, Padua, Italy, e Division of General Surgery and Organ Transplantation, Azienda Ospedaliera, San Martino, Genoa, Italy, f Renal Transplant Unit, Ospedale Regionale di Circolo Macchi, Varese, Italy, g Division of Nephrology, Dialysis and Transplantation, Azienda Ospedaliera Spedali Civili, Brescia, Italy and h Transplant Unit, Ospedale Civico, Palermo, Italy ∗Corresponding author: Giuseppe Remuzzi,
[email protected]
Numerous formulas have been developed to estimate renal function from biochemical, demographic and anthropometric data. Here we compared renal function derived from 12 published prediction equations with glomerular filtration rate (GFR) measurement by plasma iohexol clearance as reference method in a group of 81 renal transplant recipients enrolled in the Mycophenolate Mofetil Steroid Sparing (MY.S.S.) trial. Iohexol clearances and prediction equations were carried out in all patients at months 6, 9 and 21 after surgery. All equations showed a tendency toward GFR over-estimation: Walser and MDRD equations gave the best performance, however not more than 45% of estimated values were within ±10% error. These formulas showed also the lowest bias and the highest precision: 0.5 and 9.2 mL/min/1.73 m2 (Walser), 2.7 and 10.4 mL/min/1.73 m2 (MDRD) in predicting GFR. A significantly higher rate of GFR decline ranging from −5.0 mL/min/1.73 m2 /year (Walser) to −7.4 mL/min/1.73 m2 /year (Davis–Chandler) was estimated by all the equations as compared with iohexol clearance (−3.0 mL/min/1.73 m2 /year). The 12 predic-
1826
tion equations do not allow a rigorous assessment of renal function in kidney transplant recipients. In clinical trials of kidney transplantation, graft function should be preferably monitored using a reference method of GFR measurement, such as iohexol plasma clearance. Key words: GFR calculations, iohexol clearance, serum creatinine Received 16 February 2004, revised and accepted for publication 8 June 2004
Introduction Glomerular filtration rate (GFR) provides the measure of the filtering capacity of the kidneys and is considered the best overall index of renal function currently used to determine the effectiveness of therapies designed to slow the progression of chronic renal diseases. Glomerular filtration rate, however, cannot be measured directly. Thus, the assessment of this parameter is based on the evaluation of the renal clearance of an exogenous marker that fulfills the criteria for an ideal filtration substance (1,2). In this respect, the renal clearance of inulin has become the standard for measuring GFR because it is freely filtered at glomerular level but not acted on by the tubules, is not synthesized or degraded by the body and is physiologically inert. Renal clearance of radiolabeled exogenous compounds (51 Cr-EDTA, 99m Tc-DTPA or 125 I-Iothalamate) has also been used for GFR determination (3–5). More recently, the plasma clearance of nonradioactive contrast agents, such as iohexol and iothalamate, has been proposed as a reliable and more simple alternative to the renal clearance techniques (6–9) showing high correlation with the gold standard inulin clearance or the radioisotopic 51 Cr-EDTA clearance (6,7). Nevertheless, limitations related to the cost and time consumption of those procedures make them not easily implemented in clinical daily practice. For this reason, the most commonly used tool to measure GFR is still the renal clearance of the endogenous marker creatinine. Since 24-h urine sampling is required, that beside being inconvenient for the patients may be associated with improper collection, over- or under-estimation of the true GFR may occur. Moreover, creatinine metabolism and tubular secretion vary in each subject and between individuals with tendency to increase with the decline of renal function (1),
GFR Prediction in Kidney Transplantation
which ultimately make GFR measurement by creatinine clearance, at best, a gross estimate. To circumvent these drawbacks, a number of equations have been described to provide an estimate of creatinine clearance and a valid prediction of GFR by incorporating biometrical variables—age, height, weight, gender and race— in addition to serum creatinine concentration and other biochemical parameters. Most of these formulas have been just developed and tested to routinely assess renal function in patients with chronic renal failure, but their performance and suitability to monitor graft function in the kidney transplant settings have been tested so far in very few instances (10–12). Most of them, however, have been performed with a small number of patients, and none has explored their value in assessing the progressive decline in GFR post-transplant. Here we addressed the issue in a cohort of de novo kidney transplant recipients participating in the Mycophenolate Mofetil Steroid Sparing (MY.S.S.) study. This prospective, randomized, multi-center clinical trial sought to compare the steroid-sparing potential of a mycophenolate mofetil (MMF)- or azathioprine-based regimen, in patients entering the steroid tapering/withdrawal phase at 6 months posttransplantation. In particular, the present study aimed in kidney transplant recipients: (i) to identify the best reliable formula to estimate renal function among 12 different prediction equations by comparing them with the standard method of iohexol plasma clearance; and (ii) to assess their relative performance in detecting progressive decline in renal function at different time points post-transplantation by sequential measurements in individual patients over 15 months follow-up.
Methods Study population Adult recipients of a first kidney transplant from cadaver donors referred to nine Italian Renal Transplant Centers of the MY.S.S study organization since October 1997 to May 2001, and followed prospectively for 21 months post-surgery, were included in the present study. The study was approved by the Ethics Committees of all participating centers. All patients were on triple-drug immunosuppressive regimen including CsA Neoral, corticosteroids and a nucleoside synthesis inhibitor (azathioprine or MMF). Allocation to azathioprine or MMF was in a randomized fashion according to the MY.S.S., and was performed on the day of kidney transplant. At 6 months post-transplant, patients were allowed to enter the steroid-sparing phase if the following inclusion criteria were met: serum creatinine ≤ 2.0 mg/dL; stable renal function in the last 3 months; proteinuria 40 mL/min. Plasma was separated and stored at −20◦ C for HPLC determination of iohexol concentration (6). Demographic, biochemical and anthropometric data required for renal function estimation using prediction equations were recorded by each participating center on the day of plasma iohexol clearance determination and obtained from the electronic database managed at the study coordinating Clinical Research Center for Rare Diseases ‘Aldo e Cele Dacco’. ` Comparison between GFR measurements by plasma iohexol clearance and values estimated by 12 prediction equations was performed considering together the data obtained during the three time points at 6, 9 and 21 months post-transplant. Moreover, the predictive performance of each equation to reliably detect the rate of change in renal graft function over the 15 months study period was assessed. To this purpose, the slope of renal function decline was evaluated in the subgroup of 40 patients who underwent the three serial plasma iohexol clearances at 6, 9 and 21 months post-transplantation.
Analyses Determination of iohexol plasma concentrations was centralized at the Laboratory of Pharmacokinetics and Clinical Chemistry of the Clinical Research Center for Rare Diseases ‘Aldo e Cele Dacco’ ` of the Mario Negri Institute for Pharmacological Research. The plasma profiles were analyzed by a onecompartment open model system. All data were fitted by a nonlinear regression iterative program and the coefficients of correlation of the fittings were always >0.99. Iohexol clearance was estimated using the measurements from time period 120 min after the injection to the last sampling point, according to a one-compartment model (CL1 ) by the formula CL1 = Dose/AUC, where AUC is the area under the plasma concentration–time curve. According to Brochner-Mortensen ¨ (13), the plasma clearance was then estimated by the formula: CL = (0.990778 × CL1 ) − (0.001218 × CL21 ) Blood samples were analyzed for creatinine (modified rate Jaffe´ method, coefficient of variation 3.70%) and for urea (enzymatic rate method, coefficient of variation 2.03%) with an automatic device (Beckman Synchron CX5, Beckman Coulter S.p.A., Cassina De’ Pecchi, Italy).
Predictions of the graft renal function The 12 prediction equations were chosen among those reported for estimation of renal function in kidney disease (14) as well as in the general population (15). Three of them are considered predictors of GFR (Nankivell, Walser and MDRD), the remaining being prediction equations of creatinine clearance. Together, they are tools for assessing renal function (12,14,15), here generally referred as tests predicting GFR.
1827
Gaspari et al. The prediction equations we used are listed as follows: 1.
Cockcroft and Gault formula (16): = (140 − age) × weight/(72 × SCr) (×0.85 for woman)
2.
Nankivell formula (17):
Statistical analysis All data were expressed as means ± standard deviation of the mean (SD) or median. Correlations between iohexol plasma clearance and GFR estimates were studied first by regression statistics and the correlation of the linear regression. The Pearson correlation coefficient was calculated. The percent error in GFR prediction was assessed as:
= 6.7/(SCr × 0.0884) + 0.25 × weight − 0.5 × urea −100/height2 + 35(25 for woman) 3.
Bjornsson formula (18): = (((27 − (0.173 × age)) × weight × 0.07)/SCr for man = (((25 − (0.175 × age)) × weight × 0.07)/SCr for woman
4.
Davis–Chandler formula (19): = (140 − age)/SCr(×0.85 for woman)
5.
Edwards–Whyte (20): = (94.3/SCr) − 1.8 for man = (69.9/SCr) + 2.2 for woman
6.
Gates formula (21): = (89.4 × SCr−1.2 ) + ((55 − age) × (0.447 × SCr−1.1 )) for man = (60 × SCr−1.1 ) + ((56 − age) × (0.3 × SCr−1.1 )) for woman
7.
Hull formula (22): = (((145 − age)/SCr) − 3) × (weight/70) (×0.85 for woman)
8.
Jelliffe-1 formula (23): = (100/SCr) − 12 for man = (80/SCr) − 7 for woman
9.
Jelliffe-2 formula (24): = (98 − (16 × (age − 20))/20)/S.Cr. (×0.9 for woman
10.
Mawer formula (25): = (weight × (29.3 − (0.203 × age)) ×(1 − 0.03 × S.Cr.))/(14.4 × S.Cr.) for man = (weight × (25.3 − (0.175 × age)) ×(1 − 0.03 × S.Cr.))/(14.4 × S.Cr.) for woman
11.
Abbreviated MDRD (four variable) (27): = 1.86 × (S.Cr.)−1.154 × (age)−0.203 (×0.742 for woman, ×1.212 forAfrican − American)
In these formulas serum creatinine (SCr) is in mg/dL, urea in mmol/L, age in years, weight in kg and height in meter. The GFR-predicted values normalized for 1.73 m2 of body surface area (BSA) were used for all the analyses, except for the Walser’s equation. To obtain GFR values normalized for 1.73 m2 , results of Cockcroft and Gault, Bjornsson, Davis–Chandler, Edwards–Whyte, Hull, and Mawer equations have been multiplied by 1.73/BSA. The Walser model returns results normalized to 3 m2 /(height)2 .
1828
− measured value)/(measured value) × 100 In addition, a concordance study using the method of Bland and Altman (28) was performed. This method plots the differences of the results given by the two considered estimations of GFR for each individual (y-axis) and the mean of these same values (x-axis). Limits of agreement are calculated as mean difference ± 1.96 × SD of the differences. Bias in GFR estimation, a measure of systematic error, was defined by the mean prediction error (me) as: bias = me = 1/N(pei ) where pei = (predicted value − measured value) Precision of the GFR prediction was determined by the square root of the mean-squared prediction error (mse): precision = (mse)1/2 = (1/N(pe2i ))1/2 For both bias and precision, the smaller these quantities are, the greater is the performance of the GFR prediction model (29). The mean and median differences between estimated and measured GFR were calculated from: absolute difference = |predicted value − measured value| Accuracy in predicting GFR was also tested according to Minimi et al. (30). With this procedure, the different GFR estimates were ranked and compared according to the standard deviation of their respective true errors (TESD), i.e. their true dispersion. Difference between slopes of GFR decline with time post-transplant was assessed by Wilcoxon signed rank test. The statistical significance level was defined as p < 0.05.
Results
Walser formula (26): = 7.57 × (S.Cr × 0.0884)−1 − 0.103 × age + 0.096 × weight − 6.66 for man = 6.05 × (S.Cr × 0.0884)−1 − 0.080 × age + 0.080 × weight − 4.81 for woman
12.
% prediction error = (predicted value
Eighty-one patients (54 males, 27 females, mean age 42 ± 12 years), recipients of first kidney transplant from cadaver donors, who underwent GFR measurement by the plasma clearance of iohexol at 6 months post-Tx (Baseline), entered the study. At this time point post-Tx, all patients were on triple immunosuppression therapy with CsA, MMF or azathioprine and steroid. Mean body weight was 66.1 ± 11.5 kg and BSA averaged 1.75 m2 , and mean serum creatinine and urea concentrations were 1.37 mg/dL and 55 mg/dL, respectively. The mean measured GFR was 56.1 mL/min/1.73 m2 (range 21.8–86.1 mL/min/1.73 m2 ). In the follow-up, some patients experienced acute rejection during or after steroid discontinuation and were withdrawn from the study. Others did not repeat the plasma iohexol clearance. Thus, patients who underwent GFR determination at month 6, 9 and 21 post-Tx, were 81, 70 and American Journal of Transplantation 2004; 4: 1826–1835
GFR Prediction in Kidney Transplantation
Figure 1: Correlation between glomerular filtration rate (GFR) in mL/min/1.73 m2 by iohexol plasma clearance and the estimated GFR using 12 different GFR prediction equations.
American Journal of Transplantation 2004; 4: 1826–1835
1829
1830
39.3 39.3 44.4 32.1 45.4 54.1 39.3 58.1 61.2 44.4 64.8 60.2 Plasma iohexol clearance: 56.1 ± 14.5 mL/min/1.73 m2 (range 21.8–86.1 mL/min/1.73 m2 ).
31.1 29.6 31.6 19.9 34.2 37.2 29.6 39.8 40.3 31.6 45.9 44.4 9.3 11.1 10.1 12.7 9.0 7.6 10.6 7.2 6.8 10.0 6.2 6.5 64.1 ± 16.9 (28.8 to 116.2) 65.6 ± 14.1 (20.2 to 101.0) 66.1 ± 17.4 (30.8 to 123.6) 71.4 ± 22.1 (28.9 to 143.0) 64.1 ± 16.1 (31.6 to 107.6) 60.7 ± 15.8 (23.2 to 108.0) 66.6 ± 18.0 (28.8 to 122.3) 59.9 ± 15.2 (25.0 to 99.1) 59.5 ± 15.9 (26.0 to 104.8) 65.6 ± 17.7 (28.7 to 121.5) 56.1 ± 13.0 (23.7 to 90.7) 58.3 ± 15.2 (22.2 to 98.4) Cockroft-Gault Nankivell Bjornsson Davis–Chandler Edwards–Whyte Gates Hull Jelliffe-1 Jelliffe-2 Mawer Walser Abbreviated MDRD
17.30 ± 23.70 (−23.34 to 114.48) 21.70 ± 26.56 (−32.10 to 122.03) 21.01 ± 23.95 (−17.90 to 118.32) 30.19 ± 31.27 (−19.87 to 185.98) 18.55 ± 28.43 (−25.89 to 172.87) 11.44 ± 24.82 (−20.71 to 121.76) 21.64 ± 24.69 (−20.71 to 121.76) 9.78 ± 22.77 (−35.16 to 121.68) 8.79 ± 21.46 (−28.88 to 113.03) 19.76 ± 24.10 (−20.77 to 117.29) 3.43 ± 20.48 (−31.15 to 107.14) 6.68 ± 22.35 (−36.72 to 119.12)
−10 to 10% Formula
To overcome the limit of the arbitrary threshold in percent error estimate, a more conservative approach, that defines as acceptable (±1.96 standard deviation (SD) of the mean in the difference between measured and predicted GFR, was considered (28). Data again indicated the poor agreement of each GFR prediction model with plasma clearance of iohexol. Indeed for the Walser and the abbreviated MDRD models, which showed the best performance, the
Median absolute difference (mL/min/1.73 m2 )
Both Walser (mean error range: −31.15 to 107.14%) and abbreviated MDRD (mean error range: −36.73 to 119.12%) equations gave the best performance with 45% of estimated GFR within an error between ±10%. Considering error ranges between ±15% or ±20%, the Walser equation still performed the best with slightly superior results than abbreviated MDRD model. The Davis–Chandler equation had the worst performance with only 19.9, 32.1 and 43.8% of estimated GFR in the ±10%, ±15% and ±20% error ranges, respectively.
Mean GFR (mL/min/1.73 m2 ) (range)
Concordance studies Performance assessment of GFR equations is presented in Table 1 as the absolute differences and the percent error in GFR prediction as compared to the true GFR by iohexol clearance. All prediction equations showed a tendency toward GFR over-estimation, with the lowest rate (52% of GFR predictions) achieved by the Walser model. The abbreviated MDRD (57.7%), Cockroft-Gault (75.0%), Bjornsson (82.3%) and Nankivell (81.1%) equations showed an even higher GFR over-estimation, while the worst performance was with Davis–Chandler prediction model (87.2%). Mean estimated GFR ranged from 56.1 to 71.4 mL/min/1.73 m2 , median absolute difference from 6.2 to 12.7 mL/min/1.73 m2 and mean percent error in prediction from 3.43 to 30.19. The percentage of calculated values within ±10% of the measured GFR ranged from 19.9 to 45.9%, whereas the percentage within ±15% or ±20% of measured values ranged from 32.1 to 64.8% and from 43.8 to 79.6%, respectively.
Table 1: Mean calculated GFR and percent error in GFR prediction by different equations
Since 40 out of 81 patients had repeated measurements, to exclude the bias of the fact that data are not independent, we also performed additional analysis including only single measurement for 81 patients. Even with this approach, we found similar results (data not shown).
Mean % error (range)
Range of prediction error
Correlation analysis The overall relationships between measured GFR (iohexol clearance) and calculated GFR predictions are shown in Figure 1. Each prediction equation displayed a highly significant correlation (p < 0.01). Correlation coefficients (r) varied from 0.686 for the Edwards–Whyte to 0.778 for Jelliffe-2, while Cockcroft–Gault and abbreviated MDRD had intermediate r-values of 0.757 and 0.769, respectively.
−15 to 15%
−20 to 20%
45, respectively. Therefore, a total of 196 plasma iohexol clearances were available for the correlation analysis.
57.1 50.5 54.1 43.8 57.7 66.8 52.0 69.9 73.5 54.1 79.6 76.0
Gaspari et al.
American Journal of Transplantation 2004; 4: 1826–1835
GFR Prediction in Kidney Transplantation Table 2: Bias and precision (mL/min/1.73 m2 ) for GFR prediction equations
C&G Nankivell Bjornsson Davis–Chandler Edwards–Whyte Gates Hull Jelliffe-1 Jelliffe-2 Mawer Walser Abbreviated MDRD
All measurements (n = 196)
Month 6 (n = 40)
Month 9 (n = 40)
Month 21 (n = 40)
Bias
Precision
Bias
Precision
Bias
Precision
Bias
Precision
8.5 10.0 10.5 15.8 8.5 5.1 11.0 4.2 3.9 10.0 0.5 2.7
14.0 14.9 15.4 22.1 14.8 12.2 16.0 11.0 10.9 15.1 9.2 10.4
8.8 10.3 11.2 19.0 9.7 6.4 11.9 5.7 4.8 10.8 0.6 4.1
14.3 14.7 15.8 26.8 15.9 14.5 16.8 11.5 12.7 15.8 9.1 12.1
8.9 10.3 11.2 17.8 8.8 5.2 11.7 4.6 4.6 10.7 0.3 3.2
14.5 13.8 16.7 24.6 14.0 9.9 17.0 9.8 11.0 16.2 7.5 8.4
6.5 7.8 8.6 13.7 6.1 2.8 9.1 2.1 1.9 8.0 −1.6 0.6
12.2 13.1 13.8 20.3 12.8 9.4 14.0 9.5 10.1 13.3 8.5 8.4
limits of agreement were greater than 36 mL/min/1.73 m2 apart. To further examine the concordance between measured and predicted GFR, bias and precision of the 12 prediction equations were evaluated. As shown in Table 2, the best performance was found for Walser equation, with the lowest bias (0.5 mL/min/1.73 m2 ) and the highest precision (9.2 mL/min/1.73 m2 ) in predicting GFR in the whole group of the observations. Although second in the rank, the abbreviated MDRD equation showed markedly higher bias (2.7 mL/min/1.73 m2 ) and lower precision (10.4 mL/min/1.73 m2 ) than the previous model. Similar results were obtained with Jelliffe-2 equation, with 3.9 mL/min/1.73 m2 bias and 10.9 mL/min/1.73 m2 precision values. The 12 GFR equation models were also ranked from the highest to the lowest accuracy, according to the respective true error dispersion against the plasma iohexol clearance. The Walser model showed the highest accuracy (TESD = 8.29), followed by the abbreviated MDRD (TESD = 9.83) and Jelliffe-1 (TESD = 9.96); the Davis–Chandler was the least accurate model (TESD = 15.50). Comparisons between pairs of models for accuracy in GFR prediction are shown in Table 3. There was statistically significant difference between the Walser and all the other formulas, whereas the abbreviated MDRD was only statistically different with Bjornsson, Mawer, Hull, Edwards–Whyte, and Davis–Chandler. Conversely, the Davis–Chandler equation showed a significantly higher dispersion (p < 0.001) than all other models. Evaluation of renal graft function changes post-transplantation To examine the accuracy and precision of the 12 equations in predicting changes in graft function over time, we considered the subgroup of 40 patients who underwent the three measurements of plasma iohexol clearance at 6, 9 and 21 months post-transplantation. DemoAmerican Journal of Transplantation 2004; 4: 1826–1835
graphic, clinical and laboratory characteristics of these patients are summarized in Table 4. Mean serum creatinine and urea concentrations at baseline were 1.22 mg/dL and 52 mg/dL, respectively, and progressively increased during the study period. The mean body weight gain over time was mild but statistically significant as compared to baseline value. The mean measured GFR at 6 months (baseline) was 63.1 mL/min/1.73 m2 (range 35.1–86.1) and declined to 59.4 mL/min/1.73 m2 (range 26.4–79.8) and 58.6 mL/min/1.73 m2 (range 26.9–80.0) at 9 and 21 months post-transplantation, respectively. At the different time points the relationship between measured and predicted GFR showed a relatively low correlation coefficient (r). This varied from 0.61 (Gates) to 0.77 (Bjornsson) at month 6, from 0.74 (Nankivell) to 0.84 (abbreviated MDRD) at month 9 and from 0.65 (Davis–Chandler) to 0.79 (abbreviated MDRD) at month 21 post-transplant. Similarly, bias and precision analysis displayed a large scattering of the data enlightening the poor agreement of each GFR test with iohexol clearance at each of the time points considered (Table 2). Again, Walser and MDRD models were the best GFR predictors (Table 2). Both bias and precision of all the prediction equations improved over time despite a parallel mild increase in body weight, indicating that weight gain was not detrimental to the performance of renal function estimation. Figure 2 shows the mean least squares linear regression slopes of the rate of decline of GFR from 6 to 21 months post-transplant derived from GFR values predicted by each of the 12 equations. All models estimated a significantly higher rate of GFR decline than that obtained with the reference plasma iohexol clearance (−3.0 mL/min/ 1.73 m2 /year, p < 0.05). Results ranged from a mean slope of −5.0 mL/min/1.73 m2 /year (Walser) to −7.4 mL/min/1.73 m2 /year (Davis–Chandler). Thus, the percentage of error in the prediction of the rate of GFR decline varied from −66 to −146%. 1831
Gaspari et al.
1832
TESD: true error standard deviation of each model equation against iohexol plasma clearance. GFR formulas are ranked from the most (1) to the least (12) accurate and compared to each other according to their respective accuracy. NS: not significant.
NS NS NS NS NS NS p < 0.001 p < 0.001 p < 0.001 p < 0.001 NS NS NS NS p < 0.001 NS NS NS NS NS p < 0.001 NS NS NS NS NS p < 0.05 p < 0.001 1 2 3 4 5 6 7 8 9 10 11 12
Walser Abbreviated MDRD Jelliffe-1 Jelliffe-2 Nankivell Gates Cockroft-Gault Bjornsson Mawer Hull Edwards–Whyte Davis–Chandler
TESD = 8.29 TESD = 9.83 TESD = 9.96 TESD = 10.07 TESD = 10.20 TESD = 10.86 TESD = 11.10 TESD = 11.31 TESD = 11.48 TESD = 11.73 TESD = 11.77 TESD = 15.50
p < 0.025 p < 0.025 p < 0.01 p < 0.005 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.001
NS NS NS NS NS p < 0.05 p < 0.05 p < 0.025 p < 0.025 p < 0.001
NS NS NS NS NS p < 0.05 p < 0.025 p < 0.025 p < 0.001
NS NS NS NS NS p < 0.05 p < 0.05 p < 0.001
Gates Walser
Table 3: Paired comparison for GFR test accuracy
Abbreviated MDRD Jelliffe-1
Jelliffe-2
Nankivell
Cockroft -Gault
Bjornsson Mawer
Hull
Edwards– –Whyte
Discussion The present study shows that in kidney transplant recipients, predicted GFR values by all the 12 equations tested here—based on serum creatinine and urea concentration and on biometrical parameters—correlated well with the measured GFR by the standard reference method of the iohexol plasma clearance (6). When the error in GFR prediction was calculated, the Walser and abbreviated MDRD formulas performed similarly and gave better results than the other tested equations. Nevertheless, only 45% of the predicted GFR values with Walser or abbreviated MDRD models were estimated with an error between −10 and 10% as compared to the reference iohexol measurement. Taking into account that the reproducibility of the iohexol plasma clearance is 6.28% (31), the estimated GFR values lying in this ±10% error range should be considered as virtually identical to the measured GFR. This implies that less than 50% of the estimated GFR values were accurately predicted by the tested models. Even the two best prediction equations (Walser and abbreviated MDRD) in some cases largely under-estimated or over-estimated GFR as compared to iohexol plasma clearance. Both bias and precision analyses of the GFR prediction equation—more informative and powerful parameters to establish the performance of each formula—confirmed that Walser equation provides the least bias in predicting GFR in kidney transplant patients. When GFR equation models were ranked from the highest to the lowest accuracy according to the respective true error dispersion against the plasma iohexol clearance, the Walser formula performed even better than abbreviated MDRD or Nankivell equations. Our data are in keeping with previous findings in 127 transplant recipients, whose GFR values were below 50 mL/min, of a better performance of Walser model than Cockcroft–Gault, Jelliffe-2 and Mawer equations in predicting GFR (10). The Walser equation was originally developed in nontransplanted patients with severe chronic renal dysfunction (GFR < 37 mL/min), and usually is not intended to be employed for serum creatinine levels
