Modification of the Modification of Diet in Renal Disease (MDRD) Study Equation for Japan

Modification of the Modification of Diet in Renal Disease (MDRD) Study Equation for Japan Enyu Imai, MD, PhD,1 Masaru Horio, MD, PhD,2 Kosaku Nitta, MD, PhD,3 Kunihiro Yamagata, MD, PhD,4 Kunitoshi Iseki, MD, PhD,5 Yusuke Tsukamoto, MD, PhD,6 Sadayoshi Ito, MD, PhD,7 Hirofumi Makino, MD, PhD,8 Akira Hishida, MD, PhD,9 and Seiichi Matsuo, MD, PhD,10 on behalf of the Japan Chronic Kidney Disease Initiatives Background: Glomerular filtration rate (GFR)-estimating equations based on serum creatinine level may not be accurate across racial groups because of differences among races in creatinine generation. The Modification of Diet in Renal Disease (MDRD) Study equation was developed in whites and African Americans, but performance was not evaluated in Japanese. Study Design: Diagnostic test accuracy. Cross-sectional retrospective study of 3 patient groups. Equation development in 2 groups (n ⫽ 247 in 2002 to 2004; n ⫽ 214 in 2003 to 2004 with measured GFR ⬍90 mL/min/1.73 m2); external validation in a separate group (n ⫽ 153 from 1988 to 1994). Setting & Participants: Hospitalized Japanese patients with chronic kidney disease in 3 medical centers. Reference Test: Measured GFR (mGFR) computed from renal clearance of inulin. Index Test: Estimated GFR (eGFR) using the isotope dilution mass spectrometry (IDMS)-traceable 4-variable MDRD Study equation, a modified IDMS MDRD Study equation with a Japanese Society of Nephrology-Chronic Kidney Disease Initiatives (JSN-CKDI) coefficient derived in the development data set, and a new equation derived by refitting coefficients in the MDRD Study equation in the development data set. Measurements: Current creatinine assays were calibrated to standardized creatinine. Performance of equations was assessed as bias, accuracy, root-mean-squared error, and correlation coefficient of eGFR versus mGFR. Results: In the development data set, eGFR using the IDMS MDRD Study equation overestimated mGFR throughout the entire range. In the validation data set, the IDMS MDRD Study equation with the JSN-CKDI coefficient 0.741 and the new equation (JSN-CKDI) performed with significantly less bias and greater accuracy than the IDMS MDRD Study equation, but were similar to each other in accuracy and bias in patients with eGFR less than 60 mL/min/1.73 m2. In the combined development and validation data sets, the JSN-CKDI coefficient was 0.763 (95% confidence interval, 0.743 to 0.783). Limitations: Possible drift in creatinine assays over time, possible lower creatinine generation in hospitalized patients, exclusion of patients with higher GFR from the development data set. Conclusion: GFR estimates using the IDMS MDRD Study equation with the JSN-CKDI coefficient or the new JSN-CKDI equation are more accurate than the IDMS MDRD Study equation in hospitalized Japanese patients with eGFR less than 60 mL/min/1.73 m2. More studies are necessary to verify the accuracy of the JSN-CKDI coefficient and JSN-CKDI equation in other settings in Japan and elsewhere in Asia. Am J Kidney Dis 50:927-937. © 2007 by the National Kidney Foundation, Inc. INDEX WORDS: Creatinine; glomerular filtration rate; Modification of Diet in Renal Disease Study; inulin clearance.

From the Departments of 1Nephrology and 2Functional Diagnostic Science, Osaka University Graduate School of Medicine, Osaka; 3Fourth Department of Medicine, Tokyo Women’s University of Medicine, Tokyo; 4Department of Nephrology, Institute of Clinical Medicine, Graduate School of Comprehensive Human Sciences, University of Tsukuba, Ibaraki; 5Dialysis Unit, University Hospital of The Ryukyus, Okinawa; 6Syuwa General Hospital, Saitama; 7Division of Nephrology, Endocrinology and Vascular Medicine, Tohoku University Graduate School of Medicine, Miyagai; 8Department of Medicine and Clinical Science, Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences, Okayama; 9First Department of Medicine, Hamamatsu University School of Medicine, Shizuoka; and

10

Department of Nephrology, Nagoya University Graduate School of Medicine, Aichi, Japan. Received December 13, 2006. Accepted in revised form September 11, 2007. Address correspondence to Enyu Imai, MD, Department of Nephrology, Osaka University Graduate School of Medicine, Suita, Osaka 565-0871, Japan. E-mail: imai@medone. med.osaka-u.ac.jp © 2007 by the National Kidney Foundation, Inc. 0272-6386/07/5006-0005$32.00/0 doi:10.1053/j.ajkd.2007.09.004

American Journal of Kidney Diseases, Vol 50, No 6 (December), 2007: pp 927-937

927

928

G

lomerular filtration rate (GFR) is the best indicator of overall kidney function.1 Accurate estimation of GFR is important for making diagnostic decisions in clinical practice. Kidney function often is assessed by using serum creatinine (SCr) level alone. However, the assessment method is not recommended because SCr level is affected not only by GFR, but also by various factors other than GFR, such as muscle mass, sex, race, diet, age, certain drugs, and tubular function.2 It is ideal to measure GFR by means of clearance of such exogenous markers as inulin, but the method is time consuming and requires labor. To establish improved GFR estimation in clinical practice, Kidney Disease: Improving Global Outcome (KDIGO) introduced GFR-estimating equations developed from large cohorts of various racial and ethnic groups.3 The 4-variable (isotope dilution mass spectrometry [IDMS]) Modification of Diet in Renal Disease (MDRD) Study equation meets most criteria and is widely accepted by clinical practitioners.4 However, the MDRD Study equation has shortcomings in GFR estimation for Asian populations, including Japanese. We previously derived a modified MDRD Study equation to improve the performance of the MDRD Study equation in Japanese patients with chronic kidney disease (CKD).5 However, variability in SCr values in different clinical laboratories introduced errors in GFR estimation.2 To use the MDRD Study equation, SCr values must be calibrated to the value of the Cleveland Clinic (Cleveland, OH). This has been an obstacle against clinical use of the equation to date. Although a Chinese coefficient recently was reported for the original MDRD Study equation,6 use of the Chinese coefficient did not improve the performance of the MDRD Study equation in Japanese patients.5 Levey et al7 reexpressed the MDRD Study equation using standardized SCr values. The IDMS-traceable creatinine-based 4-variable MDRD Study equation (IDMS MDRD Study equation) was obtained from values of SCr samples reassayed using the Roche/Hitachi P module creatinine plus enzymatic assay (Roche Diagnostics, Basel, Switzerland) and traceable to IDMS in the National Institute of Standards and Technology. The reexpressed 4-variable MDRD

Imai et al

Study equation for estimated GFR (eGFR) is as follows7: eGFR ⫽ 175 ⫻ SCr⫺1.154 ⫻ Age⫺0.203 ⫻ 0.742 (if female) In the present study, we attempted to find an equation for accurate GFR estimation in Japanese patients with CKD by modifying the IDMS MDRD Study equation, for which data for simultaneously measured SCr and renal inulin clearance (Cin) in 3 patient groups were used. Performance of the equations was evaluated by assessing bias, accuracy, root-mean-squared error (RMSE), and the correlation coefficient between eGFR and measured GFR (mGFR). METHODS We followed 5 steps to obtain an equation that accurately estimates GFR for Japanese patients with CKD.

Step 1 The IDMS MDRD Study equation was evaluated by using 2 data sets, data 1 and data 2 combined (development data set, n ⫽ 247). SCr values were calibrated to IDMS traceable values. mGFR was obtained by means of renal Cin. eGFR was compared with mGFR by assessing bias, accuracy, RMSE, and the correlation coefficient between eGFR and mGFR.

Step 2 The IDMS MDRD Study equation was modified by adding a Japanese Society of Nephrology-Chronic Kidney Disease Initiatives (JSN-CKDI) coefficient derived by using the development data set (equation 1). The correction factor was determined from the data set of patients with mGFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2; n ⫽ 214) by minimizing the sum of squared errors of the estimate, the sum of (eGFR ⫺ mGFR).2

Step 3 A new equation, the JSN-CKDI equation (equation 2), was developed in 214 patients with CKD with mGFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2) from the development data set by using a multiple linear regression model. Age, sex, and SCr values were used as variables. Regression was performed on log-transformed data in reduced variability in differences between eGFR and mGFR. All variables in the model showed P less than 0.001.

Step 4 Equations 1 and 2 were evaluated by using a separate data set (validation data set; n ⫽ 153) for bias, accuracy, root square of errors, and the correlation coefficient between eGFR and mGFR.

Modification of the MDRD Study Equation for Japan Step 5 The JSN-CKDI coefficient in equations 1 and 2 was readjusted to improve performance in 349 patients with CKD with mGFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2) from the combined development and validation data sets. Equations 3 and 4 were derived by using the same method as for equations 1 and 2, respectively. Age, sex, and SCr values were used as variables. All variables in the model showed P less than 0.001. Finally, correlations among eGFRs by using the IDMS MDRD Study equation, equation 3, equation 4, and mGFR were examined in all patient populations combined (n ⫽ 400).

Data Set Characteristics Data set 1 included 116 patients who participated in the insulin safety clinical trial conducted by Fuji Pharmaceuticals Ltd, Saitama, Japan from 2000 to 2002.5 Patient inclusion criteria were age of 20 to 75 years and creatinine clearance of 30 to 80 mL/min/1.73 m2 (0.5 to 1.3 mL/s/1.73 m2). Data set 2 included 131 patients with CKD admitted to the Tokyo Women’s Medical University (TWMU) Hospital for renal biopsy, education about lifestyle modification in CKD, or treatment of kidney disease from 2003 to 2004. Patients included those hospitalized for 5 to 7 days to have a renal biopsy performed and patients with chronic renal failure also hospitalized for 2 weeks to receive education about dietary restriction, renal dialysis, and renal transplantation. Combined data sets 1 and 2 (n ⫽ 247) were used as the development data set. The JSN-CKDI coefficient and JSN-CKDI equation were derived in patients with mGFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2) from the development data set to compare with the IDMS MDRD Study equation for performance, for which we excluded 33 patients with GFR greater than 90 mL/min/1.73 m2 (⬎1.50 mL/s/1.73 m2) because the IDMS MDRD Study equation was derived from the majority of patients with GFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2). Data set 3 was used as the validation data set, including 153 patients with CKD hospitalized in the University of Tsukuba Hospital for renal biopsy, treatment for kidney disease, and education about lifestyle improvement from 1988 to 1994. The institutional review boards of each study institution approved anonymous use of data for this study. All patients gave written informed consent.

Inulin Renal Clearance Cin was calculated from serum inulin and urine inulin concentration and urine volume. Three sets of serum and urine samples were collected at 3 different times during the 2-hour continuous intravenous infusion of 1% inulin administered under fasting and hydrated conditions.5 Patients received 500 mL of water orally 30 minutes before the infusion. To maintain hydration, 60 mL of water was given at 30, 60, and 90 minutes after starting the inulin infusion. The infusion rate was 300 mL/h for the first 30 minutes and 100 mL/min for the following 90 minutes. Blood samples for serum inulin were collected at 0, 45, and 105 minutes after starting the inulin infusion. Urine samples for urinary

929 inulin concentration were collected between 30 and 60 minutes, 60 and 90 minutes, and 90 and 120 minutes after the patient completely emptied his or her bladder at 30 minutes. Inulin samples were assayed by means of enzymatic methods8 or the anthron method.9 Mean Cin was used as mGFR for the patient. Stored samples were not available to compare inulin assays for the 3 study populations.

SCr Measurement SCr from all samples was assayed by means of an enzymatic method using Hitachi creatinine auto-analyzer model 7170 (Hitachi, Tokyo, Japan) in the central laboratory of the inulin safety trial, model 7700 in TWMU, and model 7250 in the University of Tsukuba Hospital. For all auto-analyzers, enzyme solutions (Pureauto-s CRE-N; Daiichi Pure Chemicals Co, Tokyo, Japan) were used. Values obtained in the 3 different participating laboratories were calibrated by using IDMS-traceable reference materials. Interlaboratory differences were evaluated by using a calibration panel10 that consisted of 40 serum samples with values assigned by Roche-Hitachi P-Module Creatininase Plus enzyme assay (Global Medical Instrumentation Inc, Ramsey, MN) in the Cleveland Clinic (provided by Dr Frederic Van Lente, Cleveland Clinic). Correlation of measured and assigned values was evaluated in each laboratory by means of Bland-Altman analysis.11 Stored samples were not available to assess drift from the time of original study to the time of calibration.

Statistical Analyses Data are expressed as mean ⫾ SD. New coefficients and equations were derived by using GFR expressed on the log scale. Equation 1 was derived by fixing all slopes and varying intercepts to have the lowest mean squared error. Bias, RMSE, correlation coefficient, and overall accuracy were used to describe the relationship between GFR estimated by using each equation and mGFR. Bias of the equations was expressed as the mean difference between eGFR and mGFR (eGFR ⫺ mGFR). The difference in absolute values of biases of eGFR by equations was tested by means of paired t-test. RMSE and correlation coefficients were computed on the raw scale. RMSE was calculated as the square root of (sum of squared errors of the estimate/ [N]). Accuracies of GFR estimated by using each equation were expressed as percentage of points deviating less than 15%, 30%, and 50% from mGFR. The difference in accuracy of eGFR by using the equations was tested by means of the sign test. P less than 0.05 is considered statistically significant. Statview, version 4.02 (SAS Institute, Cary, NC) and JMP 6.02 (SAS Institute) were used for statistical analysis and calculation of correction factors and confidence intervals. We used JMP 6.02 to plot smoothed functions in the figures.

RESULTS

Creatinine Calibration

Results of creatinine calibration in the 3 laboratories are shown in Table 1 and Fig 1. For 40 reference samples of the calibration panel, val-

930

Imai et al Table 1. Creatinine Measurement in the 3 Laboratories

No. of samples assayed Mean of assigned values Mean of measured values Mean of the difference Regression plot Intercept Confidence interval P Slope Confidence interval P Bland-Altman plot Intercept Confidence interval P Slope Confidence interval P

Central Laboratory

Tokyo Women’s Medical University Hospital

University of Tsukuba

40 1.44 ⫾ 0.63 1.41 ⫾ 0.62 ⫺0.03 ⫾ 0.05

40 1.44 ⫾ 0.63 1.40 ⫾ 0.63 ⫺0.04 ⫾ 0.06

40 1.44 ⫾ 0.63 1.41 ⫾ 0.60 ⫺0.03 ⫾ 0.05

0.008 ⫺0.030-0.046 0.7 1.018 0.993-1.043 0.2 ⫺0.004 ⫺0.042 to ⫺0.033 0.8 ⫺0.020 ⫺0.045-0.004 0.09

0.045 ⫺0.000-0.089 0.05 0.999 0.969-1.028 0.9 ⫺0.039 ⫺0.084-0.006 0.09 ⫺0.003 ⫺0.032-0.027 0.9

⫺0.031 ⫺0.065-0.004 0.08 1.047 1.024-1.069 0.0002 ⫺0.033 ⫺0.000 to ⫺0.066 0.05 ⫺0.048 ⫺0.069-0.260 ⬍0.0001

Note: Values expressed as mean ⫾ SD unless noted otherwise. The regression plot shows the intercept and slope for the regression relating measured (dependent variable) to assigned creatinine (independent variable). P for slopes show whether they are significantly different from 1.0. The Bland-Altman plot shows the intercept and slope for the regression relating the difference between measured and assigned creatinine (dependent variable) to the average of measured and assigned creatinine (independent variable). Confidence interval and P for each are presented.

ues obtained by means of enzymatic methods in the central and TWMU laboratories highly correlated with values assigned by the Cleveland Clinic based on nonsignificant laboratory-specific linear regression intercepts and slopes close to 1. Creatinine values for the central laboratory and TWMU laboratory were not adjusted, although the regression slope for the central laboratory (0.993 to 1.043) and intercept from the TWMU laboratory (⫺0.000 to 0.089) were of borderline significance. Creatinine values for the University of Tsukuba Hospital were adjusted because the regression slope was significant (1.024 to 1.069). Thus, creatinine values were standardized using the equation y ⫽ ⫺0.0306 ⫹ 1.0465x, where y is the corrected value and x is the creatinine value obtained by using the enzyme method at Tsukuba University Hospital laboratory. Patient Characteristics

Patient characteristics are listed in Table 2. According to mGFR, both data sets included patients with CKD stages 1 to 5. However, most patients had a GFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2). The primary cause of CKD was chronic glomerulonephritis.

Step 1: Evaluation for Performance of the IDMS MDRD Study Equation in the Development Data Set Performance of the original IDMS MDRD equation (Caucasian coefficient, 1.0) was evaluated in the development data set (data sets 1 and 2 combined; n ⫽ 247), as listed in Tables 3 and 4. When eGFR were plotted against mGFR, the correlation coefficient between eGFR and mGFR was 0.850. However, eGFR was overestimated at the entire range of GFRs. Step 2: Modification of the IDMS MDRD Study Equation by the Addition of the JSN-CKDI Coefficient in the Development Data Set To improve performance, the IDMS MDRD Study equation was modified by using a correction coefficient. The coefficient was 0.741 for the IDMS MDRD Study equation (Table 3). The GFR-estimating equation 0.741 ⫻ IDMS MDRD Study equation was designated equation 1. When performance of equation 1 was compared with that of the IDMS MDRD Study equation, eGFR using equation 1 had significantly less bias than that using the IDMS MDRD Study equation in

Modification of the MDRD Study Equation for Japan

931

Figure 1. Bland-Altman plot for differences in serum creatinine values assayed in study laboratories and the Cleveland Clinic. Plotted creatinine values include values obtained from the Cleveland Clinic and the (A) central laboratory of the inulin trial, (B) Tokyo Women’s Medical University Hospital (TWMU), and (C) University of Tsukuba Hospital (Tsukuba).

the test of overall GFR, as listed in Table 4. The accuracy of eGFR using equation 1 also was significantly greater than that of eGFR using the IDMS MDRD Study equation in the overall GFR. Equation 1 underestimated GFR, especially in patients in the development data set with GFR greater than 60 mL/min/1.73 m2. Step 3: Development of a New Equation (JSN-CKDI Equation) by Refitting Coefficients for the IDMS MDRD Study Equation in the Development Data Set We derived a new GFR-estimating equation, equation 2 (the JSN-CKDI equation), 171 ⫻

SCr⫺1.004 ⫻ Age⫺0.287 ⫻ 0.782 (if female), in the development population with mGFR less than 90 mL/min/1.73 m2, as listed in Table 3. When evaluated in the total development population for overall GFR, eGFR using equation 2 had significantly less bias than eGFR using the IDMS MDRD Study equation, as listed in Table 4. Equation 2 provided eGFR with significantly greater accuracy than the IDMS MDRD Study equation for overall GFR. When the bias and accuracy of eGFR using equations 1 and 2 were also compared in the development data set, no significant differences

932

Imai et al Table 2. Characteristics of the Study Populations Development Data Set (n ⫽ 247) Data Set 1

Period No. of subjects (men/women) Age (y) SCr (mg/dL) Calibrated SCr (mg/dL) GFR (mL/min/1.73 m2) GFR 0-29 mL/min/1.73 m2 GFR 30-59 mL/min/1.73 m2 GFR 60-89 mL/min/1.73 m2 GFR ⬎90 mL/min/1.73 m2 Height (cm) Weight (kg) Body surface area (m2) Diagnosis Chronic glomerulonephritis Diabetes mellitus Polycystic kidney disease Nephrosclerosis Lupus Miscellaneous

Validation Data Set (n ⫽ 153) Data Set 2

Data Set 3

2000-2002 116 (76/40) 59.2 ⫾ 14.6 (23-85) 1.61 ⫾ 0.57 (0.56-3.28)

2003-2004 131 (59/72) 41.2 ⫾ 17.2 (18-77) 1.11 ⫾ 0.54 (0.46-4.26)

35.0 ⫾ 14.4 (13.0-179.2) 48 (41%) 61 (53%) 7 (6%) 0 (0%) 160.5 ⫾ 9.2 59.8 ⫾ 11.0 1.62 ⫾ 0.17

68.4 ⫾ 32.9 (8.2-169.8) 15 (11%) 39 (30%) 44 (34%) 33 (25%) 162.4 ⫾ 8.8 59.3 ⫾ 11.9 1.63 ⫾ 0.18

1988-1994 153 (54/99) 38.4 ⫾ 14.7 (18-75) 1.10 ⫾ 0.47 (0.6-3.5) 1.13 ⫾ 0.49 59.0 ⫾ 22.5 (10.2-111.3) 17 (11%) 55 (36%) 63 (41%) 18 (12%) 158.6 ⫾ 8.2 53.7 ⫾ 9.6 1.53 ⫾ 0.16

81 33 0 0 0 2

106 2 0 7 3 13

101 4 1 7 12 28

Note: Values expressed as mean ⫾ SD (range) or number (percent). Data set 1 indicates patients with CKD who participated in the inulin clinical trial; data set 2, patients at Tokyo Women’s Medical University; and data set 3, patients at the University of Tsukuba Hospital. To convert GFR in mL/min/1.73 m2 to mL/s/1.73 m2, multiply by 0.01667; serum creatinine in mg/dL to ␮mol/L, multiply by 88.4. Abbreviations: SCr, serum creatinine; GFR, glomerular filtration rate.

were found. Equation 2 underestimated GFR, especially in the GFR range greater than 60 mL/min/1.73 m2 (⬎1.0 mL/s/1.73 m2), similar to equation 1. Step 4: Validation of the Modified and New Equations in the Validation Data Set Correlation between mGFR and eGFR using equation 1 was evaluated using the validation data set. Comparisons in bias and accuracy between eGFRs using the IDMS MDRD Study equation and equation 1 are listed in Table 5. The bias of eGFR using equation 1 was significantly less than that of eGFR using the IDMS MDRD Study equation for overall GFR. The accuracy of eGFR using equation 1 was significantly greater than that of eGFR using the IDMS MDRD Study equation. Equation 1 underestimated, especially in the GFR range greater than 30 mL/min/1.73 m2 (⬎0.5 mL/s/1.73 m2), in the validation data set. Correlation between mGFR and eGFR using equation 2 was examined in the validation data set. Comparisons of bias and accuracy between

eGFRs using equation 1 and the IDMS MDRD Study equation are listed in Table 5. Bias was significantly less in equation 2 than in the IDMS MDRD Study equation for overall GFR. Equation 2 provided eGFRs with significantly greater accuracy than the IDMS MDRD Study equation. Equation 2 underestimated GFR, especially in the range greater than 30 mL/min/1.73 m2 (⬎0.5 mL/s/1.73 m2), in the validation data set. There were no significant differences between bias and accuracy of eGFRs estimated using equations 1 and 2. Step 5: Refitting Coefficients in the Development and Validation Combined Data Sets The correction coefficient for the IDMS MDRD Study equation was 0.763 when derived from data from the total 349 patients with CKD with mGFR less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2) for all study patients, including the development and validation data sets together. The final IDMS MDRD Study equation

Modification of the MDRD Study Equation for Japan

933

Table 3. Intercepts and Coefficients for GFR-Estimating Equations Used in This Study Exponent Transformed Intercept (95% CI)

Coefficient of Continuous Parameters (95% CI)

Exponent Transformed Coefficient of Dichotomous (95% CI)

Creatinine

Age

Female

Race 1.0 (if white) 1.212 (if black) 0.741 (if Japanese) (0.714-0.767)

IDMS MDRD Study Equation Equation 1

175

⫺1.154

⫺0.203

0.742

175

⫺1.154

⫺0.203

0.742

Equation 2 Equation 3

171 (111-264) 175

⫺1.004 (⫺1.108 to ⫺0.899) ⫺1.154

⫺0.287 (⫺0.397 to ⫺0.176) ⫺0.203

0.782 (0.719-0.851) 0.742

Equation 4

168 (126-225)

⫺1.044 (⫺1.127 to ⫺0.980)

⫺0.274 (⫺0.351 to ⫺0.197)

0.775 (0.727-0.826)

0.763 (if Japanese) (0.743-0.783)

Equation 1: IDMS MDRD Study equation with JSN-CKDI coefficient from the development data set GFR ⫽ 0.741 ⫻ SCr⫺1.154 ⫻ 175 ⫻ Age⫺0.203 ⫻ 0.742 (if female) Equation 2: JSN-CKDI equation from the development data set GFR ⫽ 171 ⫻ SCr⫺1.004 ⫻ Age⫺0.287 ⫻ 0.782 (if female) Equation 3: IDMS MDRD Study equation with JSN-CKDI coefficient from the combined data set GFR ⫽ 0.763 ⫻ SCr⫺1.154 ⫻ 175 ⫻ Age⫺0.203 ⫻ 0.742 (if female ) Equation 4: JSN-CKDI equation from the combined data set GFR ⫽ 168 ⫻ SCr⫺1.044 ⫻ Age⫺0.274 ⫻ 0.775 (if female) For all equations, the intercept is multiplied by the race coefficient. For example, the product of the intercept and the Caucasian coefficient for the MDRD Study equation (175) can be compared with the product of the intercept and the Japanese coefficients in equations 1 and 3 (130 and 133, respectively) and the product of the intercept for equations 2 and 4 (171 and 168, respectively). Abbreviations: GFR, glomerular filtration rate; CI, confidence interval; IDMS, isotope dilution mass spectrometry; MDRD, Modification of Diet in Renal Disease; SCr, serum creatinine; JSN-CKDI, Japanese Society of Nephrology-Chronic Kidney Disease Initiatives.

remodified in the combined study population was defined to be 0.763 ⫻ the IDMS MDRD equation (equation 3). The final new equation derived from patients with CKD with Cin less than 90 mL/min/1.73 m2 (⬍1.50 mL/s/1.73 m2) for all study data was 162 ⫻ SCr⫺1.050 ⫻ Age⫺0.266 ⫻ 0.774 (if female), as listed in Table 3. Correlations between mGFR and eGFR using the IDMS MDRD Study equation, equation 3, or equation 4 in the total 400 study patients with CKD are shown in Fig 2. Both equations appear to underestimate mGFR when eGFR is less than 40 mL/min/1.73 m2 (⬍0.7 mL/s/1.73 m2). Correlation coefficients between mGFR and eGFR were similar between equations 3 and 4 at 0.839 and 0.838, respectively. DISCUSSION

The goal of the study is to establish an equation of accurate GFR estimation for Japanese patients with CKD. Applying the JSN-CKDI coefficient, the modified IDMS MDRD Study equation improved GFR estimation in patients with CKD. The JSN-CKDI equation also provided significantly more accurate GFRs than the

original 4-variable IDMS MDRD Study equation. At present, we propose that the modified IDMS MDRD Study equation using the JSNCKDI coefficient (equation 1) and the JSNCKDI equation (equation 2) would be useful in Japanese patients if GFR is estimated from SCr values determined using the IDMS-traceable enzymatic method. Use of equations 3 and 4 should be considered after validation with additional data. Accurate estimation of GFR using the IDMS MDRD Study equation requires standardization of the SCr measurement. SCr values in most clinical laboratories in Japan are determined by means of the enzymatic method. The method provides greater precision and accuracy than the Jaffé method. The Jaffé method overestimates SCr because of noncreatinine chromogen interference. Even if inference is compensated, a nonspecific error remains in the Jaffé method that affects measurement results. According to the external quality survey program conducted by the Japan Medical Association, the interlaboratory coefficient of variation in creatinine measured using the enzymatic method was very

934

Imai et al Table 4. Performance of GFR-Estimating Equations in the Development Population Accuracy Within

eGFR

IDMS MDRD GFR (0-29 mL/min/1.73 m2) GFR (30-59 mL/min/1.73 m2) GFR (60-89 mL/min/1.73 m2) GFR (⬎90 mL/min/1.73 m2) Overall Equation 1 GFR (0-29 mL/min/1.73 m2) GFR (30-59 mL/min/1.73 m2) GFR (60-89 mL/min/1.73 m2) GFR (⬎90 mL/min/1.73 m2) Overall Equation 2 GFR (0-29 mL/min/1.73 m2) GFR (30-59 mL/min/1.73 m2) GFR (60-89 mL/min/1.73 m2) GFR (⬎90 mL/min/1.73 m2) Overall

No.

Bias (mL/min/1.73 m2) (confidence interval)

31 92 70 54 247

4.9 (3.4-6.4) 8.1 (5.9-10.3) 12.8 (8.8-16.8) 21.3 (14.2-28.3) 11.9 (9.8-14.1)

RMSE (mL/min/1.73 m2) (NRMSE)

15%

30%

50%

(confidence interval)

6.3 13.3 21.0 33.3 21.0 (0.68)

32 (19-50) 28 (20-38) 27 (18-39) 28 (18-41) 28 (23-34)

55 (38-71) 54 (44-64) 53 (41-64) 56 (42-68) 54 (48-60)

77 (60-89) 76 (66-84) 74 (63-83) 74 (61-84) 75 (69-80)

66 ⫺2.8 (⫺4.4 to ⫺1.1) 115 ⫺3.8 (⫺6.6 to ⫺1.1) 51 ⫺7.5 (⫺13.4 to ⫺1.7) 15 ⫺11.8 (⫺29.7-6.0) 247 ⫺4.8 (⫺6.9 to ⫺2.8)*

7.2 15.4 22.0 33.3 17.1 (0.58)

39 (29-51) 51 (42-60) 43 (31-57) 33 (15-58) 45 (39-51)†

80 (69-88) 97 (90-99) 76 (67-83) 93 (87-96) 86 (74-93) 94 (84-98) 67 (42-85) 100 (82-100) 79 (73-84)* 95 (92-97)*

63 ⫺1.1 (⫺2.8-0.6) 121 ⫺4.1 (⫺6.7 to ⫺1.5) 51 ⫺12.8 (⫺19.1 to ⫺6.5) 12 ⫺7.9 (⫺25.6-9.8) 247 ⫺5.3 (⫺7.4 to ⫺3.3)*

6.8 14.9 25.7 27.8 17.2 (0.56)

43 (31-55) 49 (40-58) 39 (27-53) 33 (14-61) 45 (39-51)†

81 (70-89) 95 (87-98) 76 (68-83) 93 (87-97) 82 (70-90) 96 (87-99) 75 (47-91) 100 (78-100) 79 (73-84)† 94 (90-96)*

Note: Bias (estimated GFR ⫺ measured GFR) expressed as mean (95% confidence interval). RMSE calculated as the square root of the mean of squared error. NRMSE values are shown for each row labeled “Overall”. Differences in bias of eGFR between 2 equations were tested by using paired t-test. Differences in accuracy of eGFR between 2 equations were tested by using sign test. To convert GFR in mL/min/1.73 m2 to mL/s/1.73 m2, multiply by 0.01667. Abbreviations: RMSE, root-mean-squared error; GFR, glomerular filtration rate; eGFR, estimated GFR; NRSME, normalized root of mean square error; IDMS, isotope dilution mass spectrometry; MDRD, Modification of Diet in Renal Disease. *P ⬍ 0.0001 versus IDMS MDRD Study equation. †P ⬍ 0.001 versus IDMS MDRD Study equation.

small at about 3%. Creatinine values for the calibration panel assayed in all 3 study laboratories differed slightly from values assigned by the Cleveland Clinic laboratory, requiring calibration in 1 laboratory. These differences suggest that residual differences in assays may contribute to error in the new equations. Although the interlaboratory coefficient of variation was 0.88%, there is some uncertainty about whether use of the JSN-CKDI coefficient–modified IDMS MDRD Study equation will be as accurate when used with creatinine values in other laboratories, even when all laboratories have been calibrated to a reference material traceable to IDMS. To standardize creatinine measurement for use of the MDRD Study equation was difficult among countries. However, Vickery et al12 carried out validation in the use of the IDMS MDRD Study equation and concluded that the IDMS MDRD Study equation may estimate GFR reliably using creatinine assayed by means of the IDMStraceable assay method without direct standard-

ization to the value of the Cleveland Clinic laboratory. Effort was made to find the method for accurate GFR estimation in the Asian population. Some studies evaluated the original MDRD Study equation in the Asian population with or without CKD.13-16 Zuo et al13 reported that GFRs estimated using both the 6- and 4-variable MDRD Study equations were not accurate in Chinese patients with CKD and were significantly less than mGFR values in patients with CKD stage 1 and significantly greater in patients with CKD stages 3, 4, and 5, although the 4-variable and 6-variable MDRD Study equations showed better performance than the Cockcroft-Gault (CG) equation based on the bias of their data. Jafer et al14 observed that GFR estimated by means of the 4-variable MDRD Study equation and creatinine clearance estimated by means of the CG equation were overestimated in comparison to measured 24-hour creatinine clearance in patients of a south Asian population with creatinine

Modification of the MDRD Study Equation for Japan

935

Table 5. Performance of GFR-Estimating Equations in Validation Population Accuracy Within

eGFR

No.

Bias (mL/min/1.73 m2) (confidence interval)

IDMS MDRD GFR (0-29 mL/min/1.73 m2) 8 4.3 (⫺0.8-9.5) GFR (30-59 mL/min/1.73 m2) 40 6.0 (2.3-9.8) GFR (60-89 mL/min/1.73 m2) 74 9.6 (6.3-12.9) GFR (⬎90 mL/min/1.73 m2) 31 22.9 (16.3-29.4) Overall 153 11.1 (8.6-13.5) Equation 1 GFR (0-29 mL/min/1.73 m2) 19 ⫺0.6 (⫺4.2-3.0) GFR (30-59 mL/min/1.73 m2) 83 ⫺9.8 (⫺12.9 to ⫺16.8) GFR (60-89 mL/min/1.73 m2) 50 ⫺4.8 (⫺9.3 to ⫺0.3) GFR (⬎90 mL/min/1.73 m2) 1 ⫺15.8 (ND) Overall 153 ⫺7.1 (⫺9.4 to ⫺4.8)* Equation 2 GFR (0-29 mL/min/1.73 m2) 17 1.1 (⫺3.0-5.3) GFR (30-59 mL/min/1.73 m2) 88 ⫺8.8 (⫺11.7 to ⫺5.9) GFR (60-89 mL/min/1.73 m2) 47 ⫺4.8 (⫺9.6-0.0) GFR (⬎90 mL/min/1.73 m2) 1 ⫺20.7 (ND) Overall 153 ⫺6.5 (⫺8.9 to ⫺4.2)*

RMSE (mL/min/1.73 m2) (NRMSE)

7.2 13.1 17.1 28.8 18.9 (0.84)

15%

30%

50%

(confidence interval)

13 (2-47) 45 (31-60) 34 (24-45) 26 (14-43) 34 (27-42)

50 (22-78) 65 (50-78) 69 (58-78) 58 (41-74) 65 (57-72)

63 (31-86) 78 (62-88) 86 (77-92) 71 (53-84) 80 (73-85)

7.3 17.1 16.3 15.8 15.9 (0.71)

32 (15-54) 68 (46-85) 89 (69-97) 33 (23-43) 76 (66-84) 98 (92-99) 54 (40-67) 80 (67-89) 96 (87-99) 100 (ND) 100 (ND) 100 (ND) 40 (32-48) 76 (68-82)* 96 (92-98)†

8.0 16.3 16.9 20.7 15.8 (0.70)

18 (6-41) 59 (36-78) 82 (59-94) 38 (28-48) 80 (70-87) 98 (92-99) 51 (37-65) 81 (67-90) 98 (86-99) 0 (ND) 100 (ND) 100 (ND) 39 (32-47) 78 (71-84)‡ 96 (92-98)†

Bias (estimated GFR ⫺ measured GFR) expressed as mean (95% confidence interval). RMSE calculated as the square root of the mean of squared error. NRMSE values are shown for each row labeled “Overall”. Differences in bias of eGFR between 2 equations were tested by using paired t-test. Differences in accuracy of eGFR between 2 equations were tested by using sign test. To convert GFR in mL/min/1.73 m2 to mL/s/1.73 m2, multiply by 0.01667. Abbreviations: RMSE, root-mean-squared error; GFR, glomerular filtration rate; eGFR, estimated GFR; NRSME, normalized root of mean square error; IDMS, isotope dilution mass spectrometry; MDRD, Modification of Diet in Renal Disease; ND, not determined. *P ⬍ 0.05 versus IDMS MDRD Study equation. †P ⬍ 0.01 versus IDMS MDRD Study equation. ‡P ⬍ 0.0001 versus IDMS MDRD Study equation.

clearance less than 60 mL/min/1.73 m2 (⬍1.0 mL/s/1.73 m2). Mahajan et al15 reported that both the 6-variable and 4-variable MDRD Study equations overestimated GFR measured by means of technetium-99m-diethylene triamine pentaacetic acid plasma clearance in an Indian population. However, SCr in the study was not calibrated. In a healthy Korean population, Kang et al16 conducted a validation study with comparison between GFR estimated by means of the CG equation and the 4-variable MDRD Study equation from SCr and GFR measured using technetium-99m-diethylene triamine pentaacetic acid renal clearance. GFR estimated using the CG and 4-variable MDRD Study equations were more accurate than 100/SCr and 24-hour urinary creatinine clearance, but both equations substantially underestimated GFR. Overall, GFR estimated by means of the MDRD Study equation did not fit with measured GFR in the Asian population,

suggesting the need of a new GFR-estimating equation or an ethnic correction coefficient for the MDRD Study equation. Recently, Ma et al6 improved GFR estimation for the Chinese population by applying a Chinese coefficient of 1.233 to the original 4-variable MDRD Study equation. However, the Chinese-coefficient–modified MDRD Study equation overestimated GFR in our study population when creatinine values were assayed using either the enzymatic method or noncompensated Jaffé method. The cause of overestimation was extensively discussed elsewhere.5 In brief, there were substantial differences in methods of measurements of GFR and SCr, conditions causing CKD, and prandial conditions for creatinine measurement (fasting and after meal) between the 2 studies. Limitations of the present study are as follows. (1) Inulin assays were performed in 3 different

936

Figure 2. Correlation between estimated and measured glomerular filtration rate (GFR) in the combined population. Estimated GFR was obtained by using the (A) isotope dilution mass spectrometry (IDMS) Modification of Diet in Renal Disease (MDRD) Study equation, (B) IDMS MDRD Study equation with the Japanese Society of Nephrology-Chronic Kidney Disease Initiatives (JSN-CKDI) coefficient (equation 3), and (C) the JSN-CKDI equation (equation 4). Smooth curves of the best fit to the data are shown.

Imai et al

Modification of the MDRD Study Equation for Japan

937

laboratories in 3 different periods and could not be compared directly. Hence, there may be some error in the gold standard. (2) SCr assays were performed in 3 different laboratories and in different periods. Although creatinine samples were assayed using the same method, a drift in creatinine measurement may exist because of the time difference. SCr assays could not be directly compared or corrected for drift over time; hence, there likely is some residual calibration error. (3) Development and validation data sets were collected in different periods with different criteria, although both the IDMS MDRD Study equation modified by the JSN-CKDI coefficient and equation 3 performed adequately with the validation data set. (4) Most study patients had CKD with GFR less than 90 mL/min/1.73 m2 (1.50 mL/s/ 1.73 m2). Therefore, these equations may not apply to individuals with normal GFR. (5) The entire study population was hospitalized. Creatinine generation may be different from that of a healthy population because of differences in muscle mass and dietary limitation. In conclusion, use of either the IDMS MDRD Study equation modified with the JSN-CKDI coefficient 0.741 or the newly derived JSN-CKDI equation significantly improved GFR estimation in patients with CKD. Either equation may be useful for the diagnosis of CKD for clinical practice in Japan. The coefficient of 0.763 for the IDMS MDRD Study equation needs to be evaluated in a large population in a future study. However, both equations appear to underestimate mGFR when eGFR is less than 40 mL/min/1.73 m2. Thus, more study is necessary to verify the accuracy of the new equation and understand the discrepancy between the Chinese equation and our equation.

2. Mayers GL, Miller WG, Coresh J, et al: Recommendations for improving serum creatinine measurement: A report from the Laboratory Working Group of the National Kidney Disease Education Program. Clin Chem 52:5-18, 2006 3. Levey AS, Eckardt KU, Tsukamoto Y, et al: Definition and classification of chronic kidney disease: A position statement from Kidney Disease: Improving Global Outcomes (KDIGO). Kidney Int 67:2089-2100, 2005 4. Stevens LA, Coresh J, Greene T, Levey AS: Assessing kidney function—Measured and estimated glomerular filtration rate. N Engl J Med 354:2473-2483, 2006 5. Imai E, Horio M, Nitta K, et al: Estimation of glomerular filtration rate by the MDRD Study equation modified for Japanese patients with chronic kidney disease. Clin Exp Nephrol 11:41-50, 2007 6. Ma YC, Zuo L, Chen JH, et al: Modified glomerular filtration rate estimating equation for Chinese patients with chronic kidney disease. J Am Soc Nephrol 17:2937-2944, 2006 7. Levey AS, Coresh J, Greene T, et al: Using standardized serum creatinine values in the Modification of Diet in Renal Disease Study equation for estimating glomerular filtration rate. Ann Intern Med 145:247-254, 2006 8. Summerfield AL, Hortin GL, Smith CH, Wilhite TR, Landt M: Automated enzymatic analysis of inulin. Clin Chem 39:2333-2337, 1993 9. Young MK, Raisz LG: An anthrone procedure for determination of inulin in biological fluids. Proc Soc Exp Biol Med 80:771-774, 1952 10. Levey AS, Coresh J, Greene T, et al: Expressing the Modification of Diet in Renal Disease Study equation for estimating glomerular filtration rate with standardized serum creatinine values. Clin Chem 53:1-7, 2007 11. Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307-310, 1986 12. Vickery S, Stevens PE, Dalton RN, van Lante F, Lamb EJ: Does the IDMS traceable MDRD equation work and is it suitable for use with compensated Jaffe and enzymatic creatinine assays? Nephrol Dial Transplant 21:24392445, 2006 13. Zuo L, Ma YC, Zhou YH, Wang M, Xu GB, Wang HY: Application of GFR-estimating equations in Chinese patients with chronic kidney disease. Am J Kidney Dis 45:463-472, 2005 14. Jafer TH, Schmid CH, Levey AS: Serum creatinine as marker of kidney function in South Asians: A study of reduced GFR in adults in Pakistan. J Am Soc Nephrol 16:1413-1419, 2005 15. Mahajan S, Mukhiya GK, Singh R, et al: Assessing glomerular filtration rate in healthy Indian adults: A comparison of various prediction equations. J Nephrol 18:257-261, 2005 16. Kang YS, Han KH, Han SY, Kim HK, Cha DR: Characteristics of population with normal serum creatinine impaired renal function and: The validation of an MDRD formula in a healthy general population. Clin Nephrol 63:258266, 2005

ACKNOWLEDGEMENTS We thank Dr Lesley A. Stevens for critical reading of the manuscript and kind efforts in coordinating the exchange of samples with Dr Van Lente’s laboratory and Drs Shigeko Hara, Toshiki Moriyama, Yasuhiro Ando, Hideki Hirakata, Tsuyoshi Watanabe, Kenji Wakai, Ichiei Narita, Yutaka Kiyohara, and Yoshinari Yasuda for helpful discussions. Fuji Yakuhin Co Ltd provided data from the Cin clinical trial. Support: None. Financial Disclosure: None.

REFERENCES 1. Wesson L: Renal hemodynamics in physiologic state, in Physiology of the Human Kidney. New York, Grune & Stratton, 1969, pp 96-108