In this large population based study involving 1471 subjects aged 65 to 91 years and randomly selected from the general population in the area of Ulm (Germany) prevalence of chronic kidney disease stage 3–5 varied considerably between 14.6% on a CysC-based estimating equation to 33.0% and 34.3% based on CKD-EPI and MDRD estimating equation, respectively. A steep increase with age was found with all three equations. Although the cystatin C based definition of CKD resulted in a lower prevalence compared to the creatinine based ones, other measures of renal damage such as albuminuria were more prevalent in those defined by CysC-eGFR. The association of gender with CKD was quite different, depending on the definition used: women had a higher risk for CKD with the MDRD.
The prevalence of CKD in our elderly population is roughly in line with data from others studies when considering the Cr-based estimating equations . In a different population based study in subjects aged 50–75 years also conducted in Germany and using the MDRD, however, the prevalence was slightly higher in comparable age strata . Prevalence of CKD in the Three-City study including 8705 community-dwelling elderly aged 65 years or older was 13.7% and 12.9% with the MDRD and CKD-EPI equation, and the long-term risk estimates were quite similar with both equations . In a study from the US conducted in octogenarians (mean age 86 years) in which beside CKD-EPI and also two cystatin C based equations were used to define CKD, the patterns were quite similar to our results . In this study of Shastri et al. the CKD-EPI resulted in prevalence of 51.3% and with the one-variable CysC-based equation in a prevalence of 33.0%. In our study, in the respective age category of 80+ years the CKD-EPI lead to a prevalence of 54.4% and the CysC-based equation to 30.7%. In a study from the US conducted in elderly patients aged 65 years and over the prevalence of CKD based on the MDRD was 44% .
Notably we found a higher prevalence of CKD in women when using the MDRD estimating equation. It is consistent with results from a study conducted in a population of elderly from another State in Germany  and has been a consistent feature in a systematic review . Beside gender differences in physiological factors the lower prevalence may also be a function of the correction factor for females in the Cr-based equations. Men are de facto also at higher risk for ESRD compared to women, which is also indirect evidence for a higher burden of the underlying CKD, which finally may progress to ESRD .
In addition it is known that especially the MDRD-based eGFR overestimates renal function in general in elderly subjects  and part of the high prevalence may be accountable to it. In a population of volunteers (mean age 61.3 years, SD 8.6) MDRD underestimated eGFR in a healthy population compared to CysC-based ones . Cr-based eGFR may show a large intra- and inter-individual variation due to the dependence of Cr concentration on many factors such as muscle mass, dietary intakes and other diseases . Recently the CKD-EPI formulation had been proposed to work better in the context of epidemiological studies . CKD-EPI seems to misclassify fewer low-risk patients compared to MDRD . Another study reported that compared to MDRD the CKD-EPI leads to higher estimates in young people and to lower estimates in elderly. The authors suggested an age dependent threshold for CKD .
We found a stronger association of established risk factors, comorbid conditions, and biochemical markers related more or less to risk for kidney disease with CysC-based definition of CKD. The only exception was uric acid which was stronger for Cr-based eGFR. The association of metabolic risk factors such as HDL and insulin resistance (although the latter was not available in our study) had been described with early impairment of renal function, irrespective of markers of renal damage .
Several studies consistently found a better prognostic value of cystatin C or CysC-based estimating equations with all-cause mortality [24, 25], CVD , and ESRD [24, 27]. Adding CysC- to Cr-based measures to improve predictive accuracy was also suggested by some authors , however, this approach was not superior in other studies . A meta-analysis also concluded that the diagnostic accuracy for measuring renal function is in favour for CysC when compared to Cr. However, the study included a very heterogeneous patient population and only few elderly .
Currently there is a hot debate about the definition of CKD in elderly based on a fixed threshold [29, 30]. The argument is that screening will identify many false positives, especially among elderly women, most of them having no evidence of kidney disease. The decrease in kidney function might be due to changes in kidney structure associated with aging  and be considered physiological. However, in a collaborative meta-analysis including more than 100,000 subjects from 14 studies the shape between eGFR and risk of all cause- and CVD-mortality was similar in age groups below and above age 65 years and the test for interaction between eGFR and age were not significant in most studies .
Although we found a strong association of CKD stage 3–5 with various cardiovascular diseases and risk factors, we found a statistically significant association with history of diabetes only in bivariate analyses, consistently for all three estimating equations. The lack of statistically significant associations with diabetes has also been reported by other investigators [14, 22].
When looking on the results the following limitations should be considered. We did not have GFR measurement, so we cannot determine the extent to which either CysC or Cr reflects the glomerular filtration rate as determined by a gold standard (i.e. inulin clearance). However, it is very difficult to obtain true GFR measurements in such a large population. Also, we only used a cross-sectional design, so the temporal association between the described risk factors and CKD cannot be assessed. Furthermore, we only had a single measurement of CKD, whereas usually for definition of CKD two measures within a time period of 3 months should be employed. The latter, however, is a limitation present in most of the cited epidemiological studies, but may indeed cause an increase in prevalence numbers. In addition, we used only one among the many suggested CysC-eGFR equations. Current available data on CysC-eGFR equations are still very limited. Results from this study does not exclude the possibility that CysC-based eGFR overestimates GFR in the elderly. In this case, wide use of this equation may expose this population to drug overdosing and adverse effects. However, it is notably, that the proportion with no signs of renal damage is highest in the group with no CKD based on the CysC-CKD definition. Therefore, further studies should evaluate the CysC-eGFR equations in large diverse populations and further standardization of assays is needed; especially with the Siemens Cystatin C method calibration has changed during the last five years; a downward shift in calibrator had occurred between 2006 and 2009 . This has to be considered if results obtained from older lots are compared to the current study.