A New Modified CKD-EPI Equation for Chinese Patients with Type 2 Diabetes

Objective To improve the performance of glomerular filtration rate (GFR) estimating equation in Chinese type 2 diabetic patients by modification of the CKD-EPI equation. Design and patients A total of 1196 subjects were enrolled. Measured GFR was calibrated to the dual plasma sample 99mTc-DTPA-GFR. GFRs estimated by the re-expressed 4-variable MDRD equation, the CKD-EPI equation and the Asian modified CKD-EPI equation were compared in 351 diabetic/non-diabetic pairs. And a new modified CKD-EPI equation was reconstructed in a total of 589 type 2 diabetic patients. Results In terms of both precision and accuracy, GFR estimating equations all achieved better results in the non-diabetic cohort comparing with those in the type 2 diabetic cohort (30% accuracy, P≤0.01 for all comparisons). In the validation data set, the new modified equation showed less bias (median difference, 2.3 ml/min/1.73 m2 for the new modified equation vs. ranged from −3.8 to −7.9 ml/min/1.73 m2 for the other 3 equations [P<0.001 for all comparisons]), as was precision (IQR of the difference, 24.5 ml/min/1.73 m2 vs. ranged from 27.3 to 30.7 ml/min/1.73 m2), leading to a greater accuracy (30% accuracy, 71.4% vs. 55.2% for the re-expressed 4 variable MDRD equation and 61.0% for the Asian modified CKD-EPI equation [P = 0.001 and P = 0.02]). Conclusion A new modified CKD-EPI equation for type 2 diabetic patients was developed and validated. The new modified equation improves the performance of GFR estimation.


Introduction
Diabetic nephropathy is the leading cause of end stage renal disease and is associated with significantly high risk of cardiovascular events [1]. Glomerular filtration rate (GFR) is the best index of kidney function. [2]. American Diabetes Association standards highlight GFR screening for nephropathy in diabetic patients [3]. The Modification of Diet in Renal Disease (MDRD) equation and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation are most frequently used and favored in North America, Europe and Australia [4]. However, when either equation was applied to type 2 diabetic patients, both have imperfections [5][6][7][8], because of the intrinsic factors such as serum glucose status [9] and body mass index [10] in diabetic subjects that could affect the accuracy of GFR estimates. Recently, a four-level race variable (Black, Asian, Native American and Hispanic, and White and other) CKD-EPI equation [11] was developed. In order to know whether the most frequently used equations really performed worse in diabetic subjects, a well-designed paired cohort was set up in this study to exclude other impact factors. And if this hypothesis was confirmed, a new equation was reconstructed later by modification of the original GFR estimating equation in a cohort of type 2 diabetic patients.

Participant selection
This study enrolled participants consequently from Jan 2010 to Dec 2012 in the Third Affiliated Hospital of Sun Yat-sen University, China. Participants were excluded if they had any of the following: 1) younger than 18 years, or 2) type 1 diabetes, or 3) type 2 diabetes with known non-diabetic renal disease. The other exclusion criteria were described elsewhere [12]. GFR category was classified according to the National Kidney Foundation Disease Outcomes Quality Initiative clinical practice guidelines [13]. A total of 1196 subjects were enrolled, including 589 type 2 diabetic patients and 607 non-diabetic participants. The study protocol was approved by the institutional review board at the Third Affiliated Hospital of Sun Yat-sen University. Written informed consent was obtained from each participant.

Laboratory methods
GFR was measured by the 99mTc-diethylene triamine pentaacetic acid ( 99 m Tc-DTPA) renal dynamic imaging method [14][15], as described previously [16]. The minimum sample size was determined to be as 36 based in the findings in a previous study [17]. The calibration equation form DTPA renal dynamic imaging GFR to dual plasma sample DTPA-GFR in this study was as the following: dual plasma sample DTPA-GFR (ml/min/1.73 m 2 ) = 0.167+1.057* DTPA renal dynamic imaging-GFR (ml/ min/1.73 m 2 ) (R 2 = 0.767, P,0.001). Serum creatinine (SC) level was measured by the enzymatic method on a Hitachi 7180 autoanalyzer (Hitachi, Tokyo, Japan; reagents from Roche Diagnostics, Mannheim, Germany), and recalibrated to isotope dilution mass spectrometry.

Statistical analysis
We used a stratified random sampling method based on age, body mass index (BMI) and GFR categories to obtain paired samples of participants represented either the type 2 diabetes or the non-diabetic cohorts. The bias between mGFR and estimated GFR (eGFR) was defined as mGFR minus eGFR. Precision was measured as the interquartile range (IQR) for difference. Accuracy was determined as the percentage of eGFR not deviating more the 30% from the mGFR. Confidence intervals for all metrics were calculated by means of bootstrap methods [18]. A Wilcoxon Mann-Whitney test was used for bias. In comparison between two data sets, independent samples t test was used for quantitative variables, and two independent samples test for accuracy. In comparison within a data set, McNemar test was used for accuracy. GFR was estimated by using the following equations: reexpressed 4-variable MDRD equation (GFR~175|SC {1:154 | Age {0:203 |½0:742 if patient isfemale|½1:212if patient isblack) [19], CKD-EPI equation (Table 1) [20] and Asian modified CKD-EPI equation (Table 1) [11]. All analyses were performed using SPSS software (version 11.0 SPSS, Chicago IL, USA) and Matlab software (version 2011b The Mathworks, Boston MA, USA).

Performance of the equations between diabetic and nondiabetic cohorts
Study population in this part of study. Three hundred and fifty-one pair of participants were selected from the total population of this study. In the type 2 diabetic cohort, the mean (6SD) mGFR was 62.8628.1 ml/min/1.73 m 2 . The mean mGFR was similar in the non-diabetic cohort (60.7627.9 ml/ min/1.73 m 2 ), as were the mean age, BMI, body-surface area, SC and gender ( Table 2).
Comparison of the performances between diabetic and non-diabetic cohorts. Bias of both the CKD-EPI equation and the Asian modified CKD-EPI equation in the non-diabetic cohort were less than those in the type 2 diabetic cohort (median difference, 2.9 and 0.3 ml/min/1.73 m 2 vs. 23.7 and 27.3 ml/ min/1.73 m 2 [P,0.001 for both comparisons]). In terms of both precision and accuracy, GFR estimating equations all achieved better results in the non-diabetic cohort comparing with those in the type 2 diabetic cohort (IQR for difference, ranged from 20.5 to 22.2 ml/min/1.73 m 2 for all 3 equations vs. ranged from 28.0 to 31.5 ml/min/1.73 m 2 ; 30% accuracy, ranged from 64.4% to 66.7% vs. ranged from 53.0% to 57.3% [P#0.01 for all comparisons]). However, Bias of re-expressed 4 variable MDRD

Development of a new modified equation to estimate GFR for type 2 diabetic patients
Patient's characteristics in this part of study. The general characteristics of type 2 diabetic patients are presented in Table 4. The diabetic cohort here enrolled the participants in the analyses for the first result in this paper. A total of 589 patients were enrolled, including 327 men and 260 women, and the mean age was 61.0612.7 yr, with body mass index 24.964.1 kg/m 2 , fasting plasma glucose 160.4677.0 mg/dL, glycated hemoglobin 9.3611.9%, SC 1.461.5 mg/dL, and mGFR 74.4631.0 ml/ min/1.73 m 2 . We randomly selected 379 subjects (the development data set) from the entire study population, and the remaining 210 patients were included in the validation data set.
Development of the new modified equation. We reconstructed a new modified equation using data from the development data set of this part of study by the generalized additive model. The new modified equation used the same three variables (age, gender and SC), the same knot points for SC and the same forms of smooth functions as the CKD-EPI equation. mGFR and SC were transformed to natural logarithms. The development data set was divided in to four categories according to gender and the knot points for SC. And four linear regression models were developed. The coefficients were estimated by the least-square error method (Table 1).

Overall performance of the predicting models
In the validation data set, the new modified equation showed less bias (median difference, 2.3 ml/min/1.73 m 2 for the new

Discussions
In the present study, we first confirmed the hypothesis that GFR estimating equations all achieved better results in the non-diabetic cohort comparing with those in the type 2 diabetic cohort by paired samples study. Then, we developed a new equation in a cohort of type 2 diabetic patients by modification of the CKD-EPI equation which performed the best in diabetic subjects. In the validation data set, the new modified equation achieved less bias, higher precision and greater accuracy compared with all three original equations. These results were consistent with the previous findings [15,[21][22][23][24][25][26] that modification of the original equation in a local cohort which was quiet different to the original one may improve the performance of GFR estimation in the same population. And our results will help clinicians to make suitable clinical decision for diabetic patients and avoid unnecessary examination and treatment.
Why the modified CKD-EPI equation outperformed all three original equations? There are several reasons. First, the development data set in this study was mainly type 2 diabetic patients, which was differed to the other original equations. Obesity is common in diabetic patients [27], leading to a relative low body muscle mass [28], influenced the generation of creatinine in the body. And hyperglycemia status in diabetic patients influences the measurement of GFR [29]. Second, GFR measurement method in this study was calibrated to the dual sample DTPA clearance. However, the development data sets of the other original  Table 5. Performance of bias, precision and accuracy between measured GFR and estimated GFR in the validation data set. equations used urinary clearance of iothalamate instead. Third, the validation cohort in this study had similar characters as those in the development cohort. Systematic differences generally lead to bias, whereas variation in populations' characteristics leads to imprecision.
There are limitations in our study. First, the new modified equation needs further external validations. Second, the study population in this study was restricted to Chinese patients with type 2 diabetes. Third, difference in the method to measure GFR between different equations would lead to systemic error in comparison with each other. Forth, the sample size of the validation data set in this study was relatively small. Fifth, there is not suitable statistic method for the comparison of IQR of difference between different GFR predicting models up till now.
In conclusion, we confirmed that the performances of GFR estimating equations in type 2 diabetic patients were worse than those in non-diabetic participants. And a new modified CKD-EPI equation for type 2 diabetic patients was developed and validated. The new modified equation improves the performance of GFR estimation, which may help physician to evaluate the kidney function in diabetic patients. Extensive external validations will be the next step before broadly applications.