Validation of the modification of diet in renal disease formula for estimating GFR with special emphasis on calibration of the serum creatinine assay

Validation of the Modification of Diet in Renal Disease Formula for Estimating GFR With Special Emphasis on Calibration of the Serum Creatinine Assay Stein Hallan, MD, PhD, Arne Åsberg, MD, PhD, Morten Lindberg, MD, and Harald Johnsen, MD ● Background: The Modification of Diet in Renal Disease (MDRD) formula is recommended by European and American guidelines for estimating glomerular filtration rate (GFR). However, the accuracy of the formula has been questioned in several studies. Our objective is to evaluate the performance of the MDRD formula with special emphasis on the possibility that interlaboratory calibration differences for serum creatinine reduce the accuracy of the formula. Methods: The MDRD and 7 other commonly used formulae were evaluated regarding bias, precision, and accuracy. The 215 adults included were patients with chronic kidney disease, potential kidney donors, and patients referred before nephrotoxic chemotherapy. Serum creatinine was measured by means of a kinetic Jaffe´ method (Hitachi 917, Hitachi, Tokyo, Japan; reagents from Roche Diagnostics, Mannheim, Germany). GFR, measured as plasma clearance of chromium 51–labeled EDTA (Cr-EDTA), ranged from 3 to 162 mL/min/1.73 m2. Results: The MDRD formula was heavily biased, but it still had significantly better accuracy than the other formulae tested. After recalibrating our serum creatinine values (serum creatinine [mg/dL] ⴝ ⴚ0.215 ⴙ 1.08 * serum creatinine), systematic bias was greatly reduced and better accuracy was achieved: 45.6% of results differed less than 15% from Cr-EDTA, 64.2% differed less than 30%, and 81.4% differed less than 50%. The equation for recalibrating creatinine values was based on data with traceability to reference methods and on sensitivity analysis. Conclusion: The MDRD formula seems to be the best formula available for GFR estimating, but it is based on a serum creatinine method calibrated to give much lower values than most laboratories, leading to underestimation of GFR in mild renal insufficiency. Am J Kidney Dis 44:84-93. © 2004 by the National Kidney Foundation, Inc. INDEX WORDS: Glomerular filtration rate (GFR); prediction equations; serum creatinine; calibration; validation; Modification of Diet in Renal Disease (MDRD) equation.

G

LOMERULAR filtration rate (GFR) is probably the best variable for diagnosing and monitoring kidney disease, deciding when to start dialysis therapy or perform renal transplantation, and dosing drugs, as well as many other aspects of nephrology practice. Although endstage renal disease (ESRD) is very uncommon, a mild to moderate reduction in renal function seems to affect 5% to 10% of the adult population.1 These people have a greatly increased risk for cardiovascular death,2 as well as a risk for progression to ESRD. There now are several treatment options available that have been shown to prevent or slow the progression to both cardio-

From the Departments of Medicine, Clinical Biochemistry, and Nuclear Medicine, St Olavs Hospital/Norwegian University of Science and Technology, Trondheim, Norway. Received December 22, 2003; accepted in revised form March 11, 2004. Address reprint requests to Stein Hallan, MD, PhD, St Olavs Hospital/NTNU, Department of Medicine, Division of Nephrology, Olav Kyrres gt. 17, N-7006 Trondheim, Norway. E-mail: [email protected] © 2004 by the National Kidney Foundation, Inc. 0272-6386/04/4401-0007$30.00/0 doi:10.1053/j.ajkd.2004.03.027

84

vascular disease and ESRD. Screening programs therefore are now starting up,3 and there will be a need for determining GFR in a large number of people in the years to come. Several equations for estimating GFR based on the measurement of serum creatinine and such variables as age and sex have been developed during the last 40 years. The new National Kidney Foundation–Kidney Disease Outcomes Quality Initiative guidelines4 and the European Best Practice Guidelines5 state that such estimation formulae are superior to relying on serum creatinine level alone, and they recommend using the equation developed from the Modification of Diet in Renal Disease (MDRD) Study6 or the Cockcroft-Gault (CG) equation.7 The MDRD formula has been evaluated with differing results.8-18 The reasons for inferior performance in several of these reports probably are different study populations and the multiple sources of measurement variation, for which lack of a common calibration of the serum creatinine assay across different laboratories probably is one of the most important.19 Of 11 evaluated common biochemical tests, serum creatinine had the largest variation in calibration across US laborato-

American Journal of Kidney Diseases, Vol 44, No 1 (July), 2004: pp 84-93

ESTIMATING GFR WITH THE MDRD FORMULA

ries, whereas intralaboratory reproducibility for serum creatinine was better than for most tests.20 Similar results were found among the 102 laboratories participating in the Nordic Reference Interval Project (NORIP).21 We evaluate the performance of the MDRD formula, as well as other formulae, for estimating GFR. Estimated GFR was compared with measured GFR by using plasma clearance of chromium 51–labeled EDTA (Cr-EDTA) as reference standard. We also explore whether recalibration of our serum creatinine measurements improves the performance of the MDRD formula. Hopefully, this will reduce the confusion associated with the performance of the MDRD formula. METHODS Between May 1996 and May 2003, a total of 219 adult patients underwent a successful Cr-EDTA plasma clearance study for measuring GFR at St Olav’s Hospital, Trondheim, Norway. The 3 main reasons for referral were evaluation of patients with chronic kidney disease, potential donors for renal transplantation, and patients scheduled for nephrotoxic chemotherapy. On a retrospective chart review, all except 4 patients had sufficient data for formula GFR estimation, including sex, weight, height, and a nonfasting serum creatinine measurement within the last 3 days. Cr-EDTA analysis was performed as described by Broechner-Mortensen and Rodbro.22 A dose of 0.056 MBq of Cr-EDTA (Amersham Pharmacia Biotech, Buckinghamshire, UK) per kilogram of body weight (a half dose for serum creatinine level ⬎ 2.3 mg/dL [⬎200 ␮mol/L]) was injected intravenously into the upper arm. Blood was drawn from the opposite arm at 3, 4, and 5 hours, and for patients with a serum creatinine level greater than 2.3 mg/dL (⬎200 ␮mol/L), also at 24 hours. Samples were counted in a 1282 Compugamma (LKB Wallac, Stockholm, Sweden). GFR was calculated by using a 1-compartment model with correction according to Broechner-Mortensen.23 All patients had their serum creatinine measured at the Department of Medical Biochemistry, St Olav’s Hospital. During the entire period, samples were analyzed by means of a kinetic Jaffe´ method with sample blank on a Hitachi 917 analyzer (Hitachi, Tokyo, Japan) with reagents from Roche (Roche Diagnostics, Mannheim, Germany). The method was calibrated using Cfas, identification number 759350, for which creatinine concentration (usually ⬃3.6 mg/dL [⬃320 ␮mol/L]) is determined by using gas chromatography-mass spectrometry (GC-MS) as a reference method. Because 0.9% sodium chloride was used as the zero standard and the Jaffe´ method is subject to the problem of codetermination of noncreatinine chromogens, our results are expected to be too high at values less than the calibrator level, whereas results will be too low at higher levels.24–26 Internal quality controls showed a coefficient of variation of 2% during the period. External quality controls (Lab Quality, Helsinki, Finland) showed that our values were

85

constantly 0.045 to 0.090 mg/dL (4 to 8 ␮mol/L) greater than the mean for laboratories using photometric Jaffe´ methods throughout the period. Ensuring that our creatinine values are equally calibrated with the MDRD study is an important task for which we used 3 different approaches. 1. A direct comparison using frozen samples from the MDRD study would be the optimal approach. Coresh et al19 found their results to be 0.23 mg/dL (20 ␮mol/L) greater than those in the MDRD laboratory throughout the range of values. Such a study probably will never be repeated. However, we used the same method, the same analyzer, and the same reagents as Coresh et al19; thus, we also can expect our serum creatinine values to be significantly greater than in the MDRD laboratory. 2. NORIP data for serum creatinine can provide another link to the MDRD laboratory through true creatinine values.21 The target value for serum creatinine in NORIP was 0.798 mg/dL (70.6 ␮mol/L), measured by means of isotope dilution with GC-MS at 5 different reference institutes around the word, coordinated by the National Institute of Standards and Measurements (United States). Our laboratory found a mean value of 0.962 mg/dL (85.1 ␮mol/L; SD, 0.020 mg/dL [1.8 ␮mol/L]; n ⫽ 10 measurements). The mean value for 10 different laboratories using a method similar to the one used in the MDRD laboratory (Beckman method) was 0.882 mg/dL (78.0 ␮mol/L; SD, 0.079 mg/dL [7.0 ␮mol/L]). This imprecise estimate indicates that MDRD was calibrated 0.085 mg/dL (7.5 ␮mol/L) greater than true creatinine values. A better estimate is found in an earlier evaluation of the Beckman CX3 analyzer (Beckman Coulter Inc, Fullerton, CA) in the MDRD laboratory, which showed that the mean difference between their CX3 and a high-pressure liquid chromatography method was constant and only 0.07 mg/dL (6 ␮mol/ L).27 We therefore assume that our laboratory and the MDRD laboratory will report values of 0.962 mg/dL (85 ␮mol/L) and 0.868 mg/dL (76 ␮mol/L; point A) for a true creatinine value of 0.798 mg/dL and values of 3.62 mg/dL (320 ␮mol/L) and 3.69 mg/dL (326 ␮mol/L; point B) for a true creatinine value of 3.62 mg/dL. These points give the line y ⫽ ⫺0.16 ⫹ 1.06 x. Our baseline recalibration equation, which incorporates the best available external information on calibration differences between our laboratory and the MDRD laboratory, therefore is: Recalibrated serum creatinine ⫽ ⫺0.16 mg/dL (⫺14 ␮mol/L) ⫹ 1.06 * measured serum creatinine 3. A 2-way sensitivity analysis will finally be used to find the recalibration equation giving optimal performance of the MDRD formula in our hospital. We used our baseline recalibration equation as described, the intercept then was varied between ⫺0.08 and ⫺0.28 mg/dL (⫺7 and ⫺25 ␮mol/L), and the slope was varied between 1.00 and 1.11 in the sensitivity analysis.

86

HALLAN ET AL Table 1.

MDRD28 CG7 Cockcroft GFR Jelliffe and Jelliffe45 Salazar and Corcoran46 Mawer et al47

Bjornsson48 Gates49

Formulae Used for Estimating GFR

F: 186.3 * serum creatinine ** ⫺1.154 * age ** ⫺0.203 * 0.742 M: 186.3 * serum creatinine ** ⫺1.154 * age ** ⫺0.203 F: (140 ⫺ age)/serum creatinine * (weight/72) * 0.85 * BSA/1.73 M: (140 ⫺ age)/serum creatinine * (weight/72) * BSA/1.73 Cockcroft * 0.84 F: (98 ⫺ (0.8 * [age ⫺ 20]))/serum creatinine * 0.90 * BSA/1.73 M: (98 ⫺ (0.8 * [age ⫺20]))/serum creatinine * BSA/1.73 F: (146 ⫺ age) * ([0.287 * weight] ⫹ [9.74 * height2])/(60 * serum creatinine) M: (137 ⫺ age) * ([0.285 * weight] ⫹ [12.1 * height2])/(51 * serum creatinine) F: weight * (25.3 ⫺ [0.175 * age]) * (1 ⫺ [0.03 * serum creatinine])/([14.4 * serum creatinine] * [70/weight]) M: weight * (29.3 ⫺ [0.203 * age]) * (1 ⫺ [0.03 * serum creatinine])/([14.4 * serum creatinine] * [70/weight]) F: (25 ⫺ [0.175 * age]) * weight * 0.07/serum creatinine M: (27 ⫺ [0.173 * age]) * weight * 0.07/serum creatinine F: (60 * serum creatinine ** ⫺1.1) ⫹ (56 ⫺ age) * (0.3 * serum creatinine ** ⫺1.1) M: (89.4 * serum creatinine ** ⫺1.2) ⫹ (55 ⫺ age) * (0.447 * serum creatinine ** ⫺1.1)

NOTE. Creatinine is measured as milligrams per deciliter in all formulae; to convert mg/dL to ␮mol/L, multiply by 88.4. When correction to a standard body surface area (BSA) was necessary, we estimated BSA according to Dubois and Dubois50: BSA ⫽0.20247 * height (m)**0.725 * weight (kg)**0.425.

We used the simplified form of the MDRD formula because this has been shown to perform as well as the full formula.28 We also evaluated other well-known formulae, listed in Table 1. Comparing these formulae with the reference method is a problem of measuring agreement between 2 methods. Bland and Altman29-31 studied this problem extensively, and we followed their recommendations for such evaluations.32 We used bias plots of the difference against the average of 2 methods to see whether bias and variability were uniform throughout the range of measurements. This was formally checked with Spearman’s rank correlation test between the absolute differences and the average. If they were uniform, mean difference was used for estimating bias, and the SD of the differences was used to calculate 95% limits of agreement, which reflects the precision of the method. If they were not, we first tried logarithmic transformation of the measurements. If this did not solve the problem, we regressed the difference between the methods on the average of the 2 methods. Ninety-five percent limits of agreement then were constructed around this linear regression line. To compare bias for the 8 different GFR formulae evaluated, we calculated the area between the regression line and a common distance along the zero Table 2.

Age (y) Weight (kg) Height (m) Serum creatinine (mg/dL) Cr-EDTA clearance (mL/min/1.73 m2)

difference line. To compare precision, we measured the width between the 95% limits of agreement. Accuracy was measured as the percentage of results not deviating more than 15%, 30%, and 50% from the measured GFR result. When appropriate, we used the sum of these 3 accuracies as an overall accuracy score. The accuracy of the formulae was compared by using chi-square tests. Although plasma clearance of Cr-EDTA is a reference method for measuring GFR, it has a significant biological and analytical variation. We therefore used Passing and Bablok regression instead of the traditional leastsquared method when linear regression was needed to describe the relationship of a formula to the Cr-EDTA method. Medcalc for Windows, version 4.3 (Medcalc Software, Mariekerke, Belgium) and SPSSm version 11.51 (SPSS Inc, Chicago, IL) were used for data analysis.

RESULTS

Physical and biochemical characteristics of the 215 patients, who represent a broad range of age, sex, kidney function, and indications for referral, are listed in Table 2. There were 112

Patient Characteristics Mean (SD)

Median

Observed Range

56.3 (17.2) 71.0 (15.4) 1.699 (0.104) 2.596 (1.899) 49.8 (42.4)

58 69 1.70 1.595 35

16–88 29–111 1.02–2.02 0.690–10.18 3–162

NOTE. To convert creatinine in mg/dL to ␮mol/L, multiply by 88.4.

ESTIMATING GFR WITH THE MDRD FORMULA

Fig 1. Relation of GFR estimated with the MDRD formula using original serum creatinine values to the reference method Cr-EDTA. Solid line represents regression line, and dotted lines represent 95% confidence intervals for the regression line. Line of identity also is indicated.

women and 103 men, and all except 1 patient were Caucasian. Forty-five patients were referred before nephrotoxic chemotherapy, 63 patients were potential kidney donors, and 107 patients were referred because of various grades of kidney failure. GFRs ranged from 3 to 162 mL/min/1.73 m2, measured by plasma clearance of Cr-EDTA, and serum creatinine levels ranged from 0.69 to 10.18 mg/dL (61 to 900 ␮mol/L). Figure 1 shows the relation of GFR estimated using the MDRD formula to the reference method Cr-EDTA. The visual impression of the scatter plot is that the MDRD formula performs only moderately. The regression line (MDRD ⫽ 5.1 ⫹ 0.76 * GFR Cr-EDTA) has an intersection with the y-axis significantly different from 0 (95% confidence interval, 3.8 to 6.5) and a slope significantly different from 1 (95% confidence interval, 0.71 to 0.81). Figure 2 illustrates the disagreement between the methods. There is a significant correlation between the difference and the average of the methods (r ⫽ ⫺0.55; P ⬍ 0.001), and logarithmic transformation of the data did not make the bias uniform throughout the range of data. We therefore calculated the regression line that shows that the MDRD formula overestimates GFR at low levels and underestimates GFR at near-normal levels. The overall magnitude of the bias is large. It is calculated to 2,275 arbitrary units, which is the absolute area

87

between the regression line and the zero difference line (from 0 to 140). Variability of results also is large, and we see that differences of ⫾50% are not unusual. The width between the 95% limits of agreement is 58 mL/min/1.73 m2. Accuracy is only moderate, with 37.7% of results differing less than 15% from the reference standard, 62.3% differing less than 30%, and 81.4% differing less than 50%. Bias, precision, and accuracy for the other formulae were calculated according to the same principles as for the MDRD formula. Data on the performance of all formulae are listed in Table 3. The MDRD formula had the largest overall bias, whereas the CG formula was least biased. However, the CG formula was very imprecise, whereas the MDRD formula had much tighter 95% limits of agreement. The other formulae were in between. The greatest accuracy was obtained with the MDRD and the Gates formulae. These 2 formulae had nearly identical accuracy and were clearly superior to the others. They had significantly more patients differing less than 15% from the measured GFR than any other formula. They performed significantly better than 5 other formulae when the cutoff value was set to 30% and better than 4 other formulae when the cutoff value was set to 50%. The CG formula had the lowest accuracy of all formulae tested. We first tried a constant recalibration of our

Fig 2. Bland-Altman plot showing the disagreement between the MDRD formula using original serum creatinine values and the Cr-EDTA method. Solid line represents the estimated difference between methods at various levels, dashed lines represent 95% limits of agreement, shaded area represents the bias of the MDRD formula.

88

HALLAN ET AL Table 3.

Bias, Precision, and Accuracy of the Serum Creatinine–Based GFR Formulae Compared With Measured Plasma Clearance of Cr-EDTA Bland-Altman*

MDRD28 Gates49 Jelliffe and Jelliffe45 Mawer et al47 Cockcroft GFR Bjornsson48 Salazar and Corcoran46 CG7

Accuracy Within

Bias

Precision

15%

2,275 2,075 2,112 1,060 1,327 700 775 630

58 65 58 65 85 72 70 95

37.7 34.9 24.7§ 27.9‡ 25.6§ 26.0§ 24.2§ 26.5‡

30%

50%

62.3 62.8 54.9† 56.7 51.6‡ 52.1‡ 50.2§ 48.8§

81.4 83.3 81.4 74.9† 76.3 68.8§ 70.7§ 68.8§

NOTE. Formulae are ordered according to decreasing accuracy score (sum of accuracies for the 3 different cutoff values). *Based on Bland-Altman difference plots: bias is measured as the absolute area between the regression line describing the estimated difference between the methods and the zero line of difference from 0 to 140. A large area means a large bias. Precision is measured as the width (mL/min/1.73 m2) between the 95% limits of agreement. A large width means a low precision. Also see Fig 2. †P ⬍0.10 comparing accuracy of MDRD to other formulae. ‡P ⬍0.05 comparing accuracy of MDRD to other formulae. §P ⬍0.01 comparing accuracy of MDRD to other formulae.

serum creatinine values similar to that of Coresh et al19 (recalibration I):

We then performed a sensitivity analysis using this equation as baseline to find the optimal recalibration in our hospital. Results are shown in Fig 3. The highest point of the “accuracy mountain” corresponds to the equation giving optimal accuracy performance (recalibration II):

Recalibrated serum creatinine ⫽ ⫺0.23 mg/dL (⫺20 ␮mol/L) ⫹ 1.0 * measured serum creatinine

Recalibrated serum creatinine

This improved the overall bias, but 95% limits of agreement increased, as listed in Table 4. The overall result was decreased accuracy. We then, as described in the Methods, used all available external calibration data to find a nonconstant recalibration, which we used as baseline in a sensitivity analysis to find the optimal recalibration in our hospital. Table 4.

⫽ ⫺0.215 mg/dL (⫺19 ␮mol/L) ⫹ 1.08 * measured serum creatinine This is nearly identical to the equation we found based on external calibration data. Compared with uncalibrated serum creatinine values, accuracy seemed to increase when using less

Performance of the MDRD Formula After Recalibration of Serum Creatinine Values Bland Altman*

MDRD, original values MDRD, recalibration I MDRD, recalibration II

Accuracy Within

Bias

Precision

15%

30%

50%

2,275 630 320

58 77 73

37.7 34.9 45.6†

62.3 60.9 64.2

81.4 73.0‡ 81.4

NOTE. Creatinine is measured as milligrams per deciliter; to convert mg/dL to ␮mol/L, multiply by 88.4. *See Table 3 for explanation. †P ⬍0.10 comparing accuracy of uncalibrated to calibrated values. ‡P ⬍0.05 comparing accuracy of uncalibrated to calibrated values. §P ⬍0.01 comparing accuracy of MDRD to other formulae.

ESTIMATING GFR WITH THE MDRD FORMULA

89

Fig 3. Two-way sensitivity analysis on the accuracy of the MDRD formula for estimating GFR evaluates the effect on accuracy score when varying the intercept and slope of a nonconstant equation for recalibrating serum creatinine values (recalibrated creatinine ⴝ a ⴙ b * measured creatinine). Accuracy was measured as the percentage of results not deviating more than 15%, 30%, and 50% from the measured GFR result, and the overall accuracy score is the sum of these 3 results. The greatest accuracy was achieved with the recalibration equation: creatinine ⴝ ⴚ0.215 mg/dL ⴙ 1.08 * creatinine. Creatinine is measured in milligrams per deciliter; to convert mg/dL to ␮mol/L, multiply by 88.4.

than 15% as the cutoff value (P ⫽ 0.09) and was unchanged for the other 2 cutoff levels (Table 4). After recalibration II, the MDRD formula performed significantly better than the Gates formula (P ⫽ 0.024). Recalibration II improved the overall bias and precision compared with recalibration I, and it had significantly better accuracy (⬍15%, P ⫽ 0.024; ⬍50%, P ⫽ 0.039). Revising point A in the Methods section using the recalibration II equation gives the x,y point (0.962, 0.824; corresponding points for creatinine expressed as ␮mol/L are [85,72.8]). This

indicates that the MDRD laboratory (0.824 mg/dL [72.8 ␮mol/L]) was calibrated even closer to true creatinine level (0.798 mg/dL [70.6 ␮mol/L]) than assumed in the Methods section. DISCUSSION

We evaluated the MDRD and other formulae for estimating GFR in 215 subjects with a GFR ranging from 3 to 162 mL/min/1.73 m2, measured as plasma clearance of EDTA. MDRD results were highly biased because they overestimated GFR at low levels and underestimated

90

GFR at near-normal values. A nonconstant recalibration of our serum creatinine values greatly reduced the bias and indicates that interlaboratory calibration differences influence performance of the MDRD formula. We found that the MDRD formula had significantly greater accuracy than all other formulae tested, and the CG formula was among the worst. The validity of a test depends on a sufficiently large and relevant test sample, testing of all subjects with the best reference standard available, and that test and reference standard are measured and read independently of each other.33 The MDRD formula undoubtedly is the formula that best fulfils these criteria. It therefore would be expected to perform well, which it also did in a large separate validation set in the original report of Levey et al.6 We found 10 other reports describing the performance of the MDRD formula. Lamb et al9 and Harmoinen et al15 found a very high diagnostic accuracy, but no superiority to the CG formula. Kingdon et al8 also found that the MDRD formula performed well, but results are hampered by a very low number of subjects (n ⫽ 28). The only study that confirmed the superiority and very high performance of the MDRD formula is the African-American Study of Hypertension and Kidney Disease.17 This is a very large and well-performed study to which we should attach great importance. That they had a similar study population and used the same laboratory as the MDRD study for measuring serum creatinine (CX3 Beckman) and GFR (renal clearance of iothalamate) were important contributions to obtaining these results. Six validation reports showed much more inferior test characteristics than in the original report,10-14,18 and most of these studies used different reference standards, other variations of the Jaffe´ method, different analyzers, or tested different populations: Lin et al14 studied potential kidney donors, analyzed serum creatinine at a wide variety of laboratories, and measured GFR using 2 different methods (iodine 125–labeled iothalamate or technetium 99m–labeled diethylenetriaminepentaacetate). Bostom et al18 evaluated 8 different formulae in 109 patients with kidney disease and serum creatinine levels less than 1.5 mg/dL (⬍132 ␮mol/L). GFR was measured as plasma clear-

HALLAN ET AL

ance of iohexol, and serum creatinine was measured by means of the Jaffe´ method in 8 different participating laboratories. Vervoort et al13 evaluated 46 patients with diabetes mellitus and 46 healthy subjects by means of renal inulin clearance. Our study also found more inferior results for the MDRD formula than in the original report, and the 3 previously mentioned possible reasons for this are discussed. First, we used a different reference standard than in the MDRD study (plasma clearance of EDTA versus renal clearance of iothalamate). Renal clearance of EDTA has been shown to be identical to renal clearance of inulin.34,35 However, plasma clearance of EDTA overestimates GFR with 3 to 6 mL/min because of a small constant extrarenal clearance, which is particularly relevant at low levels of GFR.34,36-39 Renal clearance of iothalamate also overestimates renal clearance of inulin with 3 to 5 mL/min at low levels of GFR, but the overestimation increases to 15 to 25 mL/min for healthy subjects.40-43 The overestimation, caused at least partly by renal tubular secretion of iothalamate,41 seems to be nonconstant and has considerable interindividual variation. Our use of a different reference standard therefore could introduce a systematic error at greater GFR levels. The second possible reason for our more inferior results is the serum creatinine assay. We found that calibration of the serum creatinine assay had a great influence on results. Other studies also found that performance of the MDRD formula is very dependent on the method used to measure serum creatinine.19,44 According to our findings, the MDRD laboratory probably was calibrated only 0.026 mg/dL (2.3 ␮mol/L) greater than true creatinine values, whereas the mean for Nordic laboratories was 0.133 mg/dL (11.8 ␮mol/L) greater than true values.21 For 60-yearold white men with serum creatinine measured to 0.826 mg/dL (73 ␮mol/L) in the MDRD laboratory, these calibration differences will introduce a systematic GFR bias of ⫺14 mL/min/1.73 m2 or more in 44% of Nordic laboratories, and 20% of laboratories will have a bias of ⫺20 mL/min/ 1.73 m2 or more. Bias decreases as serum creatinine level increases and becomes negligible with a serum creatinine level greater than 2.5 mg/dL (⬎220 ␮mol/L).

ESTIMATING GFR WITH THE MDRD FORMULA

The third possible reason for our more inferior results is a different study population. Only half our subjects had kidney disease, opposed to all subjects in the MDRD study. This might be a reason for lower accuracy, but it is difficult to quantify its specific contribution. Our overall conclusion is that a major part of the large systematic bias found when evaluating the MDRD formula in our study, as well as in other studies,13,14,18 seems to occur because the formula was developed in a laboratory with an unusually low calibrated serum creatinine assay. The MDRD formula therefore often shows an unjust low performance when tested elsewhere. Is it then possible to improve the MDRD formula? In our study, recalibrating serum creatinine values clearly improved accuracy. This was important for local use of the formula in our hospital, but our approach to the problem has limitations and is not an ideal long-term worldwide solution of interlaboratory calibration differences. What we need is a new version of the MDRD formula based on true serum creatinine values and clinical laboratories also using assays giving true serum creatinine values. This can be accomplished by using more properly calibrated and specific methods for analyzing serum creatinine, ie, enzymatic methods.24,25,44 Even Jaffe´ methods will reduce their bias using multipoint calibration against authenticated serum-based standards. The international society should adopt GC-MS values as a commonly accepted reference platform, and manufactors and users of creatinine assays should calibrate against such values. One example of such work is NORIP, which published reference method adjusted results based on a representative sample of 3,036 healthy subjects,21 and these new reference intervals will be implemented in Nordic countries during 2004. Such efforts hopefully will increase the MDRD formula’s performance toward that found in the original report. For what purposes should we use the MDRD formula? First, we must be aware that although the MDRD formula seems to be the best GFRestimating formula available, disagreement between the formula and reference standard can be large. This can have serious implications if using the formula for important individual decisions. The formula should be used only as an additional

91

test, together with available epidemiological and clinical data. Although there is no general agreement of screening for mild kidney failure, most criteria for a successful screening program are present, and the MDRD formula will be an interesting screening tool. Patients with moderate kidney failure also will have some benefit from the MDRD formula. Drug dosing often is dependent on a more accurate GFR estimate than that given by serum creatinine level alone, and the formula will be well suited for this purpose. For patients approaching ESRD, the MDRD formula will be useful for better timing of such important predialytic preparations as construction of an arteriovenous fistula, placement of a peritoneal dialysis catheter, or transplantation workup. However, although GFR is a useful adjunct to clinical data for deciding when to start renal replacement therapy, GFR probably should be measured using a plasma clearance technique as the predictive range of the MDRD estimates, at least in our study, seem to be too wide to be useful. We conclude that the MDRD formula had significantly better accuracy than the 7 other GFR-estimating formulae tested. However, the MDRD formula was heavily biased because of interlaboratory serum creatinine calibration differences. After a nonconstant recalibration of creatinine values, the systematic bias was greatly reduced and better accuracy was achieved. Results indicate that the MDRD formula is based on a serum creatinine method calibrated to give values close to that measured using reference methods, leading to underestimation of GFR in patients with mild renal insufficiency in most other laboratories. Epidemiological studies of patients with mildly reduced kidney function not taking this problem into account can be seriously flawed. Physicians also should be aware that even the MDRD formula is hampered by significant random error. When high accuracy is needed for individual decision making, reference techniques for measuring GFR should be used. REFERENCES 1. Coresh J, Astor BC, Greene T, Eknoyan G, Levey AS: Prevalence of chronic kidney disease and decreased kidney function in the adult US population: Third National Health and Nutrition Examination Survey. Am J Kidney Dis 41:112, 2003

92

2. Henry RM, Kostense PJ, Bos G, et al: Mild renal insufficiency is associated with increased cardiovascular mortality: The Hoorn Study. Kidney Int 62:1402-1407, 2002 3. Brown WW, Peters RM, Ohmit SE, et al: Early detection of kidney disease in community settings: The Kidney Early Evaluation Program (KEEP). Am J Kidney Dis 42:2235, 2003 4. National Kidney Foundation: K/DOQI Clinical Practice Guidelines for Chronic Kidney Disease: Evaluation, classification, and stratification. Am J Kidney Dis 39:S1S246, 2002 (suppl 1) 5. European Best Practice Guidelines for Haemodialysis (Part 1): Section I. Measurement of renal function, when to refer and when to start dialysis. Nephrol Dial Transplant 17:S7-S15, 2002 (suppl 7) 6. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D: 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 130:461-470, 1999 7. Cockcroft DW, Gault MH: Prediction of creatinine clearance from serum creatinine. Nephron 16:31-41, 1976 8. Kingdon EJ, Knight CJ, Dustan K, et al: Calculated glomerular filtration rate is a useful screening tool to identify scleroderma patients with renal impairment. Rheumatology (Oxford) 42:26-33, 2003 9. Lamb EJ, Webb MC, Simpson DE, Coakley AJ, Newman DJ, O’Riordan SE: Estimation of glomerular filtration rate in older patients with chronic renal insufficiency: Is the Modification of Diet in Renal Disease formula an improvement? J Am Geriatr Soc 51:1012-1017, 2003 10. Pierrat A, Gravier E, Saunders C, et al: Predicting GFR in children and adults: A comparison of the CockcroftGault, Schwartz, and Modification of Diet in Renal Disease formulas. Kidney Int 64:1425-1436, 2003 11. Rodrigo E, Fernandez-Fresnedo G, Ruiz JC, et al: Assessment of glomerular filtration rate in transplant recipients with severe renal insufficiency by Nankivell, Modification of Diet in Renal Disease (MDRD), and Cockroft-Gault equations. Transplant Proc 35:1671-1672, 2003 12. Van Den Noortgate NJ, Janssens WH, Delanghe JR, Afschrift MB, Lameire NH: Serum cystatin C concentration compared with other markers of glomerular filtration rate in the old old. J Am Geriatr Soc 50:1278-1282, 2002 13. Vervoort G, Willems HL, Wetzels JF: Assessment of glomerular filtration rate in healthy subjects and normoalbuminuric diabetic patients: Validity of a new (MDRD) prediction equation. Nephrol Dial Transplant 17:1909-1913, 2002 14. Lin J, Knight EL, Hogan ML, Singh AK: A comparison of prediction equations for estimating glomerular filtration rate in adults without kidney disease. J Am Soc Nephrol 14:2573-2580, 2003 15. Harmoinen A, Lehtimaki T, Korpela M, Turjanmaa V, Saha H: Diagnostic accuracies of plasma creatinine, cystatin C, and glomerular filtration rate calculated by the CockcroftGault and Levey (MDRD) formulas. Clin Chem 49:12231225, 2003 16. Bertolatus JA, Goddard L: Evaluation of renal func-

HALLAN ET AL

tion in potential living kidney donors. Transplantation 71:256260, 2001 17. Lewis J, Agodoa L, Cheek D, et al: Comparison of cross-sectional renal function measurements in African Americans with hypertensive nephrosclerosis and of primary formulas to estimate glomerular filtration rate. Am J Kidney Dis 38:744-753, 2001 18. Bostom AG, Kronenberg F, Ritz E: Predictive performance of renal function equations for patients with chronic kidney disease and normal serum creatinine levels. J Am Soc Nephrol 13:2140-2144, 2002 19. Coresh J, Astor BC, 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 39:920-929, 2002 20. Ross JW, Miller WG, Myers GL, Praestgaard J: The accuracy of laboratory measurements in clinical chemistry: A study of 11 routine chemistry analytes in the College of American Pathologists Chemistry Survey with fresh frozen serum, definitive methods, and reference methods. Arch Pathol Lab Med 122:587-608, 1998 21. Rustad P: Reference intervals for 25 of the most frequently used properties in clinical chemistry; Proposal by Nordic Reference Interval Project (NORIP). Klinisk Biokemi Norden 15:10-17, 2003 22. Brochner-Mortensen J, Rodbro P: Selection of routine method for determination of glomerular filtration rate in adult patients. Scand J Clin Lab Invest 36:35-43, 1976 23. Brochner-Mortensen J: A simple method for the determination of glomerular filtration rate. Scand J Clin Lab Invest 30:271-274, 1972 24. Thienpont LM, Van Landuyt KG, Stockl D, De Leenheer AP: Candidate reference method for determining serum creatinine by isocratic HPLC: Validation with isotope dilution gas chromatography-mass spectrometry and application for accuracy assessment of routine test kits. Clin Chem 41:995-1003, 1995 25. Blijenberg BG, Brouwer RJ, Baadenhuijsen H, Boerma GJ: Creatinine and surveys: An assessment. Eur J Clin Chem Clin Biochem 33:855-858, 1995 26. Blijenberg BG, Brouwer HJ: The accuracy of creatinine methods based on the Jaffe reaction: A questionable matter. Eur J Clin Chem Clin Biochem 32:909-913, 1994 27. Coresh J, Toto RD, Kirk KA, et al: Creatinine clearance as a measure of GFR in screenees for the AfricanAmerican Study of Kidney Disease and Hypertension pilot study. Am J Kidney Dis 32:32-42, 1998 28. Levey AS, Greene T, Kusek J, Beck GJ, Group MS: A simplified equation to predict glomerular filtration rate from serum creatinine. J Am Soc Nephrol 11:0828A, 2000 (abstr) 29. Bland JM, Altman DG: Comparing methods of measurement: Why plotting difference against standard method is misleading. Lancet 346:1085-1087, 1995 30. Bland JM, Altman DG: Comparing two methods of clinical measurement: A personal history. Int J Epidemiol 24:S7-S14, 1995 (suppl 1) 31. Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307-310, 1986

ESTIMATING GFR WITH THE MDRD FORMULA

32. Bland JM, Altman DG: Measuring agreement in method comparison studies. Stat Methods Med Res 8:135160, 1999 33. Irwig L, Macaskill P, Glasziou P, Fahey M: Metaanalytic methods for diagnostic test accuracy. J Clin Epidemiol 48:119-130 discussion, 13, 1995 34. Jagenburg R, Attman PO, Aurell M, Bucht H: Determination of glomerular filtration rate in advanced renal insufficiency. Scand J Urol Nephrol 12:133-137, 1978 35. Sambataro M, Thomaseth K, Pacini G, et al: Plasma clearance rate of 51Cr-EDTA provides a precise and convenient technique for measurement of glomerular filtration rate in diabetic humans. J Am Soc Nephrol 7:118-127, 1996 36. Rehling M, Nielsen BV, Pedersen EB, Nielsen LE, Hansen HE, Bacher T: Renal and extrarenal clearance of 99mTc-MAG3: A comparison with 125I-OIH and 51Cr-EDTA in patients representing all levels of glomerular filtration rate. Eur J Nucl Med 22:1379-1384, 1995 37. Mortensen JB, Rodbro P: Comparison between total and renal plasma clearance of [51Cr]EDTA. Scand J Clin Lab Invest 36:247-249, 1976 38. Norden G, Bjorck S, Granerus G, Nyberg G: Estimation of renal function in diabetic nephropathy. Comparison of five methods. Nephron 47:36-42, 1987 39. Rehling M, Moller ML, Thamdrup B, Lund JO, Trap-Jensen J: Simultaneous measurement of renal clearance and plasma clearance of 99mTc-labelled diethylenetriaminepenta-acetate, 51Cr-labelled ethylenediaminetetraacetate and inulin in man. Clin Sci (Lond) 66:613-619, 1984 40. Perrone RD, Steinman TI, Beck GJ, et al: Utility of radioisotopic filtration markers in chronic renal insufficiency: Simultaneous comparison of 125I-iothalamate, 169YbDTPA, 99mTc-DTPA, and inulin. The Modification of Diet in Renal Disease Study. Am J Kidney Dis 16:224-235, 1990

93

41. Odlind B, Hallgren R, Sohtell M, Lindstrom B: Is 125I iothalamate an ideal marker for glomerular filtration? Kidney Int 27:9-16, 1985 42. Back SE, Krutzen E, Nilsson-Ehle P: Contrast media and glomerular filtration: Dose dependence of clearance for three agents. J Pharm Sci 48:765-767, 1988 43. Petri M, Bockenstedt L, Colman J, et al: Serial assessment of glomerular filtration rate in lupus nephropathy. Kidney Int 34:832-839, 1988 44. Wuyts B, Bernard D, Van den Noortgate N, et al: Reevaluation of formulas for predicting creatinine clearance in adults and children, using compensated creatinine methods. Clin Chem 49:1011-1014, 2003 45. Jelliffe RW, Jelliffe SM: A computer program for estimation of creatinine clearance from unstable serum creatinine levels, age, sex and weight. Math Biosci 14:17-24, 1972 46. Salazar DE, Corcoran GB: Predicting creatinine clearance and renal drug clearance in obese patients from estimated fat-free body mass. Am J Med 84:1053-1060, 1988 47. Mawer GE, Lucas SB, Knowles BR, Stirland RM: Computer-assisted prescribing of kanamycin for patients with renal insufficiency. Lancet 1:12-15, 1972 48. Bjornsson TD: Use of serum creatinine concentrations to determine renal function. Clin Pharmacokinet 4:200222, 1979 49. Gates GF: Creatinine clearance estimation from serum creatinine values: An analysis of three mathematical models of glomerular function. Am J Kidney Dis 5:199-205, 1985 50. Dubois D, Dubois EF: A formula to estimate the approximate surface area if height and weight be known. Arch Intern Med 17:863-871, 1916