Should the CKD EPI Equation Be Used for Estimation of the Glomerular Filtration Rate in Obese Subjects?

Chronic kidney disease (CKD) is one of the major global health burdens. It is associated with the risk for end-stage kidney disease, and additionally, it is an independent risk factor for cardiovascular disease and premature deaths [1-3]. According to the systematic analysis from the Global Burden of Disease Study, the global prevalence of the CKD in 2017 was 9.1%, with a record of 697.5 million cases of all-stage CKD [4]. Besides diabetes and hypertension, obesity is the most important factor associated with the increase of CKD prevalence [2, 5, 6]. Obesity, causing impairment of renal function and structure, is associated with development and progression of CKD and in the initial stages, leads to glomerular hyperfiltration (GHF). GHF particularly when associated with albuminuria is an independent risk of CKD progression, as well as risk for cardiovascular incidents and death [7-12]. Debate is going on about the proper definitions of CKD and GHF, but all are based on the glomerular filtration rate (GFR). Many equations have been proposed for estimation of GFR (eGFR). Currently, the CKD epidemiology collaboration (CKD EPI) equation is recommended by the international societies [13-19]. However, there are several important limitations of the CKD EPI equation: it is not validated in obese subjects, and it uses a unique body surface area (BSA) of 1.73 m² that represents the average BSA of 25-year-old Americans a hundred years ago, which is significantly smaller than the average BSA of today’s population [20-26]. By indexing GFR for the unique BSA of 1.73, renal function is underestimated in approximately one-third of the population, leading to a falsely higher CKD prevalence. On the other hand, it reduces true prevalence of GHF, still a neglected subgroup of individuals with increased kidney and cardiovascular risk [27-29]. Directly measured GFR is the gold standard, but cost and (relative) impracticality disable its implementation in everyday work [30]. Therefore, we aimed to compare eGFR values obtained using the CKD EPI equation with GFR values calculated with equations where individual BSA was included, so-called de-indexed equations, and to analyze the differences in the prevalence of eGFR categories (surrogate CKD stages) and GHF.

The study included adults from the Endemic Nephropathy in Croatia – epidemiology, diagnosis and etiopathogenesis (ENAH) study cohort. The study included respondents (door-to-door, 85% participation rate) from rural part of Croatia and originally included 3,305 adults. As this was an endemic area, all subjects with Balkan endemic nephropathy (diseased and suspect) were excluded. At the end, in total, 2,058 subjects were included. All participants signed a written informed consent, completed an extensive questionnaire, and were clinically examined. Measurements and data collection were done by physicians, trained medical students, and nurses. The Ethical Committee of School of Medicine, University of Zagreb, and the Croatian Public Health Institute approved the research.

Subjects were divided into three body mass index (BMI) categories: ≥30 kg/m², obese; 25–30 kg/m², overweight; and <25 kg/m², normal body weight. Abdominal obesity was defined as waist circumference >102 cm and >88 cm for men and women, respectively. BSA was calculated with DuBois and DuBois and Moesteller equations (online suppl. Table S1; for all online suppl. material, see www.karger.com/doi/10.1159/000526115). Blood pressure (BP) was measured (Omron M6) according to the ESH/ESC guidelines. The average BP value was obtained from the second and third measurements. Hypertension was defined as BP ≥140/90 mm Hg and/or taking antihypertensive therapy. Diabetes was defined as fasting glucose >7 mmol/L and/or taking antidiabetic therapy. Fasting blood sample was drawn, and the first morning urine sample was taken. Serum creatinine was measured on the Olympus AU2700 analyzer using the Jaffé kinetic uncompensated method with continuous measurement (Beckman Coulter, Brea, CA, USA). Calibration was performed using calibrators from the same company that can be monitored by the IDMS method and standard reference material. The eGFR was calculated with the CKD EPI equation, de-indexed equations using each individual’s BSA values (CKD EPI equation adjusted for individual BSA values calculated using the DuBois equation [CKD DBi] and CKD EPI equation adapted for individual BSA values calculated using the Moesteller equation [CKD Mi]), and with the Salazar-Corcoran equation (online suppl. Table S1). According to the KDIGO guidelines, subjects were classified into CKD stages regarding the eGFR values. GHF was defined as the 95th percentile of eGFR for age decade depending on gender. Other laboratory parameters were determined using routine laboratory methods.

Categorical variables were showed as percentages and continuous variables as an arithmetic mean with standard deviation or as medians with interquartile range (25th and 75th percentile). Data distribution was tested by the D’Agostino Pearson test and graphically. In the analysis of categorical variables, the χ² test was used and in the case of small samples, Fisher’s exact test. To compare the two continuous variables, a two-way T test for independent or paired samples for parametric or the Mann-Whitney U test or Wilcoxon test for nonparametric analysis was used. Variance analysis (ANOVA) and the Tukey-Kramer test for post hoc analysis were used to compare more than two groups of correctly distributed variables, the and Kruskal-Wallis test and post hoc Mann-Whitney U test were used to compare more than two groups of incorrectly distributed variables. The correlation was shown using the Pearson correlation coefficient. Statistical significance was set at 0.05. The agreement between the eGFR equations was presented using statistical agreement indices for continuous data: concordance correlation coefficient (CCC), total deviation index (TDI), and coverage probability (CP). The CCC is weak if CCC <0.90; moderate if CCC = 0.90–0.95; significant if CCC = 0.95–0.99; and almost perfect if CCC >0.99. The ideal situation is when the TDI is <10% which means that 90% of the eGFR values are between ±10% of the reference values. The CP shows the ratio of the matched data that are within the tolerance based on the absolute difference or proportional change. The CP displays values 0–1 and estimates whether the TDI is less than a predetermined fixed percentage. To compare the mean values of different eGFR equations, the Bland-Altman method was used to estimate the deviation of individual equations in different BMI categories. Limits of agreement (LoA) are defined as the mean difference ± 1.96 SD difference. The narrower the distance between the boundaries, the better the two methods match. If these limits do not exceed the maximum allowable difference between the Δ methods (differences within the mean ± 1.96 SD are not clinically relevant), the two methods are considered consensual and can be used equally. Bias is defined as the mean difference between two equations. Precision is defined as the standard deviation of mean difference where wide ranges mean low precision. Accuracy is defined as the percentage of eGFR values within 10% (P10) or 30% (P30) of the eGFR reference value limits. In studies that describe GFR equations, the most commonly used indicator is P30, which is defined as the percentage of GFR values that are ±30% of the measured GFR, in our case, the reference eGFR. P10 is the indicator defined as the percentage of GFR values that are ±10% of the reference value. The statistical program IBM SPSS version 23 was used in the data analysis.

The characteristics of 2,058 subjects are shown in Table 1, and the characteristics of the subjects classified according to BMI categories are listed in online supplementary Table S2. BSA was the lowest in BMI <25 kg/m² subgroup and gradually but significantly increased in the higher BMI categories. In the BMI subgroup, 25–30 kg/m², and in the BMI subgroup, ≥30 kg/m², the CKD EPI eGFR values were significantly lower than the values calculated using other equations (p < 0.001). In the BMI ≥30 kg/m² subgroup, the prevalence eGFR ≥stage 3A was significantly higher when eGFR was calculated using the CKD EPI equation than values obtained either with CKD DBi (χ² = 5.422; p = 0.019) or CKD Mi (χ² = 8.586; p = 0.003) (Table 2; Fig. 1). On contrary, in the BMI <25 kg/m² subgroup, the prevalence of CKD was lower when we used the CKD EPI equation compared to CKD DBi (χ² = 3.277; p = 0.07) and to CKD Mi equations (χ² = 3.277; p = 0.07) (online suppl. Table S3). In the BMI <25 kg/m² subgroup, we found more GHF using the CKD EPI equation compared to CKD DBi (χ² = 47.79; p < 0.001) and CKD Mi (χ² = 37.95; p < 0.001) equations. In contrast, the BMI ≥30 kg/m² subgroup had significantly less GHF with the CKD EPI equation compared to CKD DBi (χ² = 30.02; p < 0.001) and CKD Mi (χ² = 32.6; p < 0.001) equations. Subjects with a BMI of 25–30 kg/m² did not differ in prevalence of GHF using any of the equation. When using the CKD EPI equation in the BMI <25 kg/m² subgroup, we observed significantly more GHF than in the BMI 25–30 kg/m² subgroup (χ² = 5.79; p = 0.016) and in the BMI ≥30 kg/m² subgroup (χ² = 6.32; p = 0.011). With the CKD DBi equation, we found completely opposite results. In subjects with BMI <25 kg/m², there was a significantly lower proportion of GHF than in the BMI 25–30 kg/m² subgroup (χ² = 32.37; p = 0.001) and in the BMI ≥30 kg/m² subgroup (χ² = 88.95; p < 0.001). Using the CKD Mi equation, results were similar. In the whole group, there were 5.9% and 13.9% less GHF when we compared the CKE PI equation to the CKD Mi and CKD DBi, respectively. When using CKD EPI, subjects with BMI ≥30 kg/m² had 69.3% and 70.5% less GHF compared to CKD DBi and CKD Mi, respectively. The opposite results were obtained in the group of subjects with BMI <25 kg/m², where 85.7% and 91% more GHFs were found using CKD EPI compared to CKD DBi and CKD Mi equations, respectively.

Proper estimation of GFR is particularly important in high-risk obese subjects who have higher risk for developing CKD and in the early phase are more prone to have GHF. One of the main drawbacks of the CKD EPI equation is indexing to BSA and using the unique number of 1.73 m² [20-26]. The average BSA values of our study population of are in line with the average BSA of the general adult Croatian population (1.94 m²) and current data from the USA (2.06 m² men, 1.83 m² women) and many other countries primarily reflecting increase of obesity prevalence [23, 25, 31-37]. In our group, only 34.4% of subjects have normal BMI, and only in this subgroup, the average BSA was 1.73 m². Being aware that equations for estimation of GFR are unreliable, particularly when they are indexed to the unique number of 1.73, the National Kidney Disease Education Program recommended using de-indexed equations in subjects with very small or very large body proportions where individual BSA is included into the equation [38]. In our group, CKD EPI eGFR values were significantly lower than values obtained with de-indexed equations. Only in the BMI <25 kg/m² subgroup, we failed to find differences in GFR values between CKD EPI and de-indexed equations because the average BSA in this subgroup was 1.73 m², exactly the same as it is in the CKD EPI equation. On contrary, in the subgroup with BMI ≥30 kg/m² (average BSA 2.08 m²), the differences in GFR between CKD EPI and de-indexed equations were 15.5 mL/min. Our results are in agreement with other reports. In the group of subjects with the average BSA >2 m², Redal-Baigorri et al. [32] found the difference between indexed and de-indexed equations of 20 mL/min, and Firedman et al. noticed a difference of 29 mL/min in the group of subjects with the average BSA of 2.27 m² [27]. López-Martínez et al. [34] reported that indexation of measured GFR for BSA underestimated GFR of 10 mL/min in 25% of patients. Accordingly, in the BMI ≥30 kg/m² subgroup, prevalence of CKD stage ≥3A was higher when we used CKD EPI than other equations. Similar differences were observed in the BMI 25–30 kg/m² subgroup (average BSA of 1.9 m²). However, in the BMI <25 kg/m² subgroup, prevalence of CKD was 1.8% lower when we used the CKD EPI equation than de-indexed equations. On contrary, the prevalence of GHF in obese subjects was significantly lower when we used CKD EPI versus de-indexed equations. Our results are in line with the report of Wurzner et al. [8, 39] who found in two different populations that the prevalence of GHF increases as the BMI category increases. Rothberg et al. [35] found that de-indexed equations are more reliable in very obese subjects in diagnosing GHF than the CKD EPI equation. In the BMI ≥30 kg/m² subgroup, we found 70% less GHF when using CKD EPI versus de-indexed equations which is completely in agreement with results of López-Martínez et al. [34] who observed difference of 65%. On contrary, in the BMI <25 kg/m² subgroup, there were 85–90% more GHF when we used the CKD EPI equation versus de-indexed equation. All these results strongly suggest that the CKD EPI equation could not be considered reliable for the whole population, particularly not for obese subjects.

Our results are in line with other reports, showing that difference between indexed and de-indexed equations increases as BMI, i.e., BSA increases regardless if GFR was measured or calculated [36, 37]. Redal-Baiggori et al. [33] found that increase of BSA is associated with underestimation of GFR if GFR is indexed for BSA. Indexation for BSA underestimated GFR in higher BMI categories because BMI strongly correlates with BSA. Importantly, indexation for BSA is associated with underestimation of GFR only in obese subjects which was also shown by Lemoine et al. [40]. These results are in line with studies where GFR was measured. In subjects with BSA <1.8 m², Redal-Baiggori et al. [32] found small difference (1.57 mL/min; LoA: 0.86–2.28). Interestingly, we found similar results in the subgroup BMI <25 kg/m² when comparing CKD EPI versus CKD DBi, reflecting an average BSA of 1.73 m² in this subgroup. In the group of subjects with BSA >1.8 m², Redal-Baiggori et al. [33], who measured GFR, observed significant differences (−12.31 mL/min; LoA: 13.94 to −10.68), which was again in line with results we found in our obese subjects who had BSA >1.8 m² (−13.6 mL/min; LoA: 13.9 to −12.8) and are in agreement with reports of Wuerzner et al. [39]. It should be underlined that Redal-Baiggori et al. [33] found the same differences and bias between indexed and de-indexed estimated GFR as they found between indexed and de-indexed measured GFR, supporting that bias is associated with BSA as the normalizing factor [32].

Disagreement between all equations increases as GFR increases. López-Martínez et al. [34] reported that incorrect GFR results were achieved in more than 25% of subjects with GFR >120 mL/min regardless of the equation used. We found the same. Bland-Altman graphs (online suppl. Table S4; Fig. S1A, B) are showing that bias is becoming more negative as eGFR increases, particularly when eGFR is >120 mL/min which is in line with other authors, indicating the importance of choosing the correct equation for the particular BMI group in diagnosing GHF [32-35].

Also, data on the hemodynamics of the kidneys in the consumption of protein-rich foods and the hemodynamics of the deterioration of renal function and progression to the terminal stage of CKD in obese individuals are scarce. A study by Anastasio et al. [41] indicates that in nonalbuminuric obese individuals, the GFR value (measured by the inulin method) was reduced compared to healthy subjects, while the eGFR value was normal. These results indicate the complexity of renal hemodynamics in obese individuals and still insufficiently studied renal mechanisms and points to the need to compare eGFR with the gold standard, which is difficult to do in most epidemiological studies.

Our study has several limitations. First, serum creatinine was determined only once, and we did not include other elements impotent for making diagnosis of CKD, i.e., albuminuria; thus, we can discuss only eGFR stages and not CKD. However, our aim was not to precisely diagnose CKD but to analyze reliability and agreements among equations in obese subjects. Second, we analyzed data from only one population, inhabitants from a rural area, and all participants were Caucasians; thus, our results could not be extrapolated to other populations. Third, we did not measure GFR and show data only as estimated/calculated GFR. This is the case in most large epidemiological studies, due to its complexity and the cost of the procedure of measured GFR. The main concern with the CKD EPI equation in obese subjects is indexation on the unique 1.73 number. Obese subjects have BSA significantly higher than 1.73 m²; thus, we suggested that de-indexed equations with individual BSA could be recommended instead of the CKD EPI equation so long as direct measurement of GFR becomes more reliable, cheaper, and simple. However, it would be important to analyze which GFR standardization would better predict the decline per year of GFR. In this cohort, we are going to analyze this important item during the 10-year period, hoping to report these results in the following study. Our study has several important strengths. Data of more than 2,000 subjects were analyzed. This is the first report on kidney function and BSA in rural population from this European region; we used very strict definition of GHF. Finally, we used adequate statistical methods for estimation of agreements among various equations, which were not always the case in other reports [30].

In conclusion, according to our results, the CKD EPI equation is unreliable in subjects with BMI ≥30 kg/m². De-indexed equations are equally reliable in all BMI categories, and based on these observations, it could be concluded that de-indexed equations could be recommended instead of the CKD EPI equation in, at least, epidemiology surveys that include obese subjects so long as direct measurement of GFR becomes more reliable, cheaper, and simple. This statement is in agreement with conclusion of other authors who consider that both CKD EPI and MDRD equations could be improved if not indexed to BSA [21, 32-34, 42].

We are grateful to all collaborators of the research project who helped in field and laboratory work and to farmers who agreed to participate.

All participants signed a written informed consent completed an extensive questionnaire containing personal and family history data and were clinically examined. This study was performed in line with the principles of the Declaration of Helsinki. The study was approved by the Ethical Committees of the School of Medicine, University of Zagreb, and the Croatian Public Health Institute. Reference number: 108-0000000329.

The authors have no conflicts of interest to declare.

This study was supported by several grants: Endemic Nephropathy in Croatia – epidemiology, diagnosis and etiopathogenesis (Ministry of Science and Education and Sports 108-0000000329), epidemiology of arterial hypertension in Croatia (EH-UH 1) (Ministry of Science and Education and Sports 0108109), and the Croatian Science Foundation “Epidemiology of arterial hypertension and salt intake in Croatia” (EH-UH 2) (IP-06-2016).

Marija Domislovic and Viktor Domislovic participated in the field work, conducted statistical analyses, and prepared the first draft and final version of the manuscript. Mirjana Fucek conducted all biochemistry tests, participated in data cleaning, and contributed to preparing the final version of the manuscript. Ana Jelakovic and Lana Gellineo participated in the field work, in data entering and cleaning, and contributed to preparing the final version of the manuscript. Zivka Dika coordinated field work and contributed to preparing the final version of the manuscript. Bojan Jelakovic designed the study, organized research projects, participated in the field work, contributed to interpretation of statistical analyses, and prepared the first draft and final version of the manuscript.

The data that support the findings of this study are available on request from the corresponding author. Data are a part of ENAH study, and due to its proprietary nature and ethical reasons, supporting data may not be publicly available. Additional information on data and access conditions is available upon request to project leader Bojan Jelakovic.