Epidemiologic studies in Western countries have shown that prevalence of stage 3 CKD, defined as a GFR less than 60 mL/min/1.73 m2, is about 10% in the general population [1, 2, 11, 27]. But there are large differences between studies regarding the prevalence of CKD which can be explained not only by the population characteristics, but also by the difference in the method used to estimate GFR . Most of the recent data have been obtained with the MDRD study equation using a well calibrated serum creatinine [2, 28, 29]. However, the use of this equation is not free from criticisms. We and others, have demonstrated that this equation tends to strongly underestimate GFR in healthy populations and, more generally, in patients with normal or near normal creatinine values [12, 30, 31]. Thus, the Levey’s group has recently proposed a new equation, the CKD-EPI equation, which is expected to be better especially in the higher GFR range (over 60 ml/min/1.73 m2). Indeed, the CKD prevalence is significantly lower according to the CKD-EPI equations for the present population cohort than if the MDRD equation is used (9.8% versus 13%). Among the patients screened, 140 (3.3%) were classified as having stage 3 CKD with the MDRD study equation but not with the CKD-EPI equation. This is important from an epidemiological point of view. We have thus confirmed that a higher prevalence of stage 3 CKD when the MDRD equation is used [11, 32–36]. The higher prevalence of stage 3 CKD found with the MDRD study equation compared to the prevalence observed with the newer CKD-EPI equation may be explained by the systematic underestimation of GFR obtained with this first equation, especially in women [12, 30, 35, 37]. Interestingly, we confirmed that difference between the MDRD and the CKD-EPI equations seems to decrease with age, notably in subjects older than 70 years [34–38]. These results may be explained by the performances of the two equations which have been showed to be very similar (or at least less different) in elderly patients . Indeed, it has been shown that the MDRD equation yields systematically lower eGFR results (for a given age and creatinine level) than the CKD-EPI equation except in the very old patient . This is confirmed by our own data showing that all patients presenting discrepant results between MDRD and CKD-EPI equations are actually positively screened only by the MDRD equation, except in one patient who was the oldest of the sample (96 years old).
In their initial CKD-EPI article , Levey et al. have also compared prevalence of CKD in the NHANES study. If we consider the same definition of CKD (only based on eGFR), these authors have found that prevalence of CKD in the NHANES study was 10.06% with the CKD-EPI equation and 14.9% with the MDRD study equation (Appendix Table 9 in ) in subjects aged between 60 and 69 years. The discrepancies observed between the two equations are comparable to discrepancies observed in our study.
One can conclude that the performance of the CKD-EPI equation to determine the prevalence of CKD in epidemiological studies is better than the MDRD study equation [11, 32]. However, this equation is not free from criticism. The CKD-EPI notably remains dependent on the limitations of serum creatinine and its precision is imperfect. Cystatin C is often presented as a better marker of kidney function, especially because its concentration is less influenced by muscular mass than serum creatinine . New equations based on standardized cystatin C have been recently proposed by the CKD-EPI consortium. These equations have been built and internally validated from an impressive sample (5592 subjects in the developing dataset and 1119 in the validation dataset). The authors demonstrated better performance of the equations combining creatinine and cystatin C compared to the CKD-EPI equation . Recent studies confirmed the good performances of the two cystatin C-based equations in specific populations [24, 40]. This better performance is also supported by the studies that show a better estimate of mortality risk when using cystatin C compared to creatinine-based equations [41, 42].
In our population of volunteers, the prevalence of stage 3 CKD is strongly discordant using creatinine- or cystatin C-based equations. Indeed, if the prevalence of CKD is as high as 13% and 9.8% using the MDRD or the CKD-EPI study equations, the prevalence will decrease to 5%, for the equations based on cystatin C only or on cystatin C and creatinine. Interestingly, the seminal study on the cystatin C-based equations published by Inker et al. showed a better performance of the CKD-EPI mix equation to estimate GFR . In our study, we found however no difference in the prevalence of stage 3 CKD results between the CKD-EPI Cys or the CKD-EPI mix. This may be related to the way the authors calibrated the cystatin C measurement, which was not unquestionable. Also, the bias of the equations influences the epidemiologic results and we note that in the CKD-EPI study (Table 3 in ), bias of the CKD-EPI Cys is similar to the bias of the CKD-EPI mix equation in the range of interest for the CKD screening (i.e. around 60 mL/min/1.73 m2).
In our cohort, when we considered the ability of each equation to detect CKD, kappa statistics showed very good agreement (κ = 0.84) between the CKD-EPI and the MDRD equations. Agreement is still acceptable, even if lower (κ = 0.71), between the CKD-EPI Cys and the CKD-EPI mix. However, agreement between CKD-EPI (or MDRD) and CKD-EPI Cys is low (0.39 and 0.32, respectively) although agreement is slightly better between CKD-EPI (and MDRD) and the CKD-EPI mix equations (0.59 and 0.48, respectively). These results underlined the powerful value of each biomarker in the equations: more concordant results between equations based exclusively on creatinine (MDRD versus CKD-EPI), poor concordant results between equations based on creatinine and on cystatin C only (CKD-EPI or MDRD versus CKD-EPI Cys), and intermediate concordance results when the mix equation is compared to others. Better concordance between the CKD-EPI and the two new cystatin C-based equations compared to the MDRD study equation is also probably explained by the mathematical construction of the equation (different exponent applied to creatinine according to creatinine level) and by the fact that the MDRD study equation has been developed on a very different sample  although the three other equations have been developed from comparable cohorts [11, 23].
Bias and precision results must be interpreted with caution in our study. Indeed, the systematic difference between the MDRD and the CKD-EPI equations seems low (3 mL/min/1.73 m2). It can be suggested from Figure 1 that this bias is however very dependent on the GFR level. The systematic difference between the MDRD and the CKD-EPI equations increases with increasing GFR values especially at high values. Comparing the three CKD-EPI equations, we also found a positive correlation between the difference in equations and the mean of the two equations, which means that the differences between equations slightly increase with eGFR levels. This correlation is however low (r2 between 0.02 and 0.04, p < 0.001) and linear although a clear non linear relationship with a knot join-point around X = 80 mL/min/1.73 m2 does exist when MDRD is compared to CKD-EPI and CKD-EPI Mix. This fact also means that discrepancies between the MDRD and the CKD-EPI are systematic and almost in the same direction, i.e. MDRD giving a positive screening result and CKD-EPI giving a negative one. Such a systematic deduction cannot be done when discrepancies are found between the CKD-EPI Cys and the CKD-EPI mix equations.
The prevalence of stage 3 CKD may be higher in women if the CKD-EPI, and especially if the MDRD study equation is used [2, 11]. Contrary to the CKD-EPI and the MDRD study equations, the prevalence of stage 3 CKD with the cystatin C-based equations is the same in both men and women. In the absence of mGFR, this discrepancy in results according to gender cannot be explained. Regarding the underestimation of women’s GFR by the MDRD equation , the higher prevalence of stage 3 CKD in women must be interpreted with caution. Differences in prevalence of CKD according to gender observed between the three CKD-EPI equations deserve further studies.
Diabetes is a well-known risk factor associated with CKD. As expected, CKD prevalence was higher in diabetic patients than in non-diabetic patients. Interestingly, the difference between the prevalences with creatinine-based versus the cystatin C-based equations appears less important than in the general population. Once again, in the absence of measured GFR, we can only describe such differences. Moreover, we must be careful in our interpretation of this result, as diabetic patients have also different clinical characteristics than non-diabetics, like a higher mean BMI.
CKD prevalence is also greatly influenced by age. As expected, we observed an increasing prevalence of stage 3 CKD according to age [2, 23, 37, 45, 46]. At every age range, the prevalence of stage 3 CKD remains significantly lower with the CKD-EPI Cys and the CKD-EPI mix, except in the higher age category but results must be interpreted with caution due to the low sample number.
It is useful to analyze the characteristics of patients with discrepant results. Generally, concordant results for the 4 equations were found in 3702 subjects (88%). If we based our CKD definition on eGFR less than 60 mL/min/1.73 m2 with all four equations, the prevalence of stage 3 CKD would be 3.1%. Our first comment must be emphasized: a perfect concordance between all four equations does not necessarily mean that the results of these equations are correct. Indeed, in the absence of mGFR, it remains possible that all four equations misclassified the patients. The same argument must be applied in the following discussion on discrepant subjects. In 131 subjects, the discrepant result is the positive screening with the MDRD study equation. As already discussed, there is little doubt that the MDRD effectively underestimates GFR and thus yields a substantial proportion of false positive screening results. The most numerous discrepant patients are those with a positive screening result with the creatinine-based equations (MDRD and CKD-EPI) but with a negative screening result with the two cystatin C-based equations. These 218 patients are slightly older compared to the general population (67 ± 7 y versus 63 ± 7 y). Because mGFR is not really known, such a discrepancy remains difficult to explain. However, this last result could be interpreted in the light of recent studies suggesting that the combined creatinine and cystatin-C based equations better estimate GFR in the elderly subject [24, 45]. Moreover, it must be underlined that the proportion of diabetic and hypertensive subjects was the same in the whole population. The positive screening result using cystatin C-based equations could thus be the most accurate, especially in the elderly . In 63 patients, the CKD-EPI Cys was the only one to give a negative screening result. These subjects were slightly older and more often women (71.4%). On the contrary, the CKD-EPI Cys is the only equation to give a positive screening result in 49 subjects. These subjects were older and had a higher BMI. Once again, in the absence of mGFR, we can only note the discrepancies without asserting which equation actually gives the better results. The possible influence of body mass index on cystatin C levels deserves future study . Further studies are actually important in these discrepant patients who are also more frequently hypertensive and diabetic.
There are several limitations to our study. First, the main limitation is linked to the fact that we have not measured GFR with a reference method. Therefore, even if we have indirect arguments to affirm that the CKD-EPI and MDRD equations overestimate the prevalence of CKD, such an assertion is conclusive only if a reference method to measure GFR is used. In our cohort, we can only identify clinical characteristics of discrepant patients. In the same vein, our Jaffe assay has lesser precision than enzymatic assays, although IDMS traceable. Second, our population may not be representative of the Belgian population because only volunteers were included. We actually noted a higher proportion of hypertension and diabetes than expected in the general population. Therefore, our results are not generally applicable for epidemiological considerations but more of an illustration of potential discrepancies in the CKD epidemiology due to difference in estimating GFR [21, 32]. Third, we have no data on the ethnicity. As the ethnicity factor for each equation is different, this could be source of bias. But in the Province of Liège, Caucasians are by far the dominant ethnic group. Therefore, it is doubtful that the differences observed in our study are due to ethnic factors. Fourth, as in several epidemiological studies, our subjects have been tested only one time. A strict definition of CKD requires that two or three samples confirm the first result after at least three months . With this correct definition, CKD prevalence would likely still be lower [48–52]. Last, we have defined stage 3 CKD as a GFR less than 60 mL/min/1.73 m2. The definition of CKD is however subject to debate and we have recently questioned this definition, notably in elderly population [31, 37, 46].