Two Novel Equations to Estimate Kidney Function in Persons Aged 70 Years or Older
Download the PDF here |
Ann Intern Med. 2 October 2012
Elke S. Schaeffner, MD, MS*; Natalie Ebert, MD, MPH*; Pierre Delanaye, MD, PhD; Ulrich Frei, MD; Jens Gaedeke, MD; Olga Jakob; Martin K. Kuhlmann, MD; Mirjam Schuchardt, PhD; Markus Tolle, MD; Reinhard Ziebig, PhD; Markus van der Giet, MD; and Peter Martus, PhD
"In summary, iohexol measurements revealed lower GFRs than predicted by current equations used to estimate GFR in participants aged 70 years or older who have normal or mild to moderately reduced kidney function. The newly developed BIS equations may provide more precise and accurate tools for estimating GFR in this age group."
"Several results were striking and may indicate that older adults are a unique population in which traditional assumptions are not necessarily true. We show that in elderly participants, the MDRD study equation yielded higher eGFRs across CKD stages than did the CKD-EPI and especially the Cockcroft-Gault equation. This contrasts with the situation in younger adults in whom implementation of the CKD-EPI equation has reduced CKD prevalence (13) but agrees with current results seen in older adults (39 - 40). The MDRD study equation is known to overestimate GFR in elderly persons with reduced muscle mass (41). Accordingly, prevalence rates of GFR less than 60 mL/min per 1.73 m2 estimated by the Cockcroft-Gault equation were much higher than those estimated by the MDRD study equation, as has been shown earlier in large data sets (42 - 43). The higher prevalence rates of reduced GFR compared with the estimates by the MDRD and CKD-EPI equations were surprising. The MDRD study population is considerably younger and does not include persons older than 70 years, but the CKD-EPI Collaboration study does include older adults. Potential explanations for the different prevalence rates could be that the initial CKD-EPI Collaboration study conditions were more heterogeneous than those of the BIS in terms of samples, marker used for measuring clearance, methodology and way of measuring GFR (plasma and urine clearance), and calibration of serum creatinine. This may explain a larger variation of the data, but whether it relates to bias remains speculative."
Background: In older adults, current equations to estimate glomerular filtration rate (GFR) are not validated and may misclassify elderly persons in terms of their stage of chronic kidney disease.
Objective: To derive the Berlin Initiative Study (BIS) equation, a novel estimator of GFR in elderly participants.
Design: Cross-sectional. Data were split for analysis into 2 sets for equation development and internal validation.
Setting: Random community-based population of a large insurance company.
Participants: 610 participants aged 70 years or older (mean age, 78.5 years).
Intervention: Iohexol plasma clearance measurement as gold standard.
Measurements: GFR, measured as the plasma clearance of the endogenous marker iohexol, to compare performance of existing equations of estimated GFR with measured GFR of the gold standard; estimation of measured GFR from standardized creatinine and cystatin C levels, sex, and age in the learning sample; and comparison of the BIS equations (BIS1: creatinine-based; BIS2: creatinine- and cystatin C-based) with other estimating equations and determination of bias, precision, and accuracy in the validation sample.
Results: The new BIS2 equation yielded the smallest bias followed by the creatinine-based BIS1 and Cockcroft-Gault equations. All other equations considerably overestimated GFR. The BIS equations confirmed a high prevalence of persons older than 70 years with a GFR less than 60 mL/min per 1.73 m2 (BIS1, 50.4%; BIS2, 47.4%; measured GFR, 47.9%). The total misclassification rate for this criterion was smallest for the BIS2 equation (11.6%), followed by the cystatin C equation 2 (15.1%) proposed by the Chronic Kidney Disease Epidemiology Collaboration. Among the creatinine-based equations, BIS1 had the smallest misclassification rate (17.2%), followed by the Chronic Kidney Disease Epidemiology Collaboration equation (20.4%).
Limitation: There was no validation by an external data set.
Conclusion: The BIS2 equation should be used to estimate GFR in persons aged 70 years or older with normal or mild to moderately reduced kidney function. If cystatin C is not available, the BIS1 equation is an acceptable alternative
Accurate assessment of kidney function is important for appropriate clinical care. However, most currently used estimates of creatinine clearance were not developed in populations of older adults.
Two estimates of glomerular filtration rate (GFR) were developed and validated in a study population of adults aged 70 years or older: 1 based on creatinine only and 1 based on both creatinine and cystatin C measurements. Both showed excellent agreement with directly measured GFR.
The study was cross-sectional. Only white participants with normal to moderately decreased kidney function were included.
Two newly developed estimates of GFR may provide more accurate assessment of kidney function in older adults.
Chronic kidney disease (CKD) has increasingly been considered a research and public health priority, with even some discussion of a silent epidemic (1). The initiation of an automatic reporting of the glomerular filtration rate (GFR), the best indicator of kidney function (2), has led to an increase in nephrology referrals, especially among persons identified with only mild to moderately reduced kidney function (GFR, 30 to 59 mL/min per 1.73 m2) (3). This is especially true for older adults, in whom prevalence rates vary in the literature between one third and nearly one half of the general population (4 - 9), impressive numbers that have raised controversy among experts about the clinical relevance of the CKD diagnosis in this age group. The debate has been heated by the fact that few data exist about normal kidney function in elderly persons (10). The pure assessment of creatinine-based GFR in the elderly is already problematic: Neither the Cockcroft-Gault (11) nor the 2 most frequently used estimating equations, the Modification of Diet in Renal Disease (MDRD) study equation (12) and the Chronic Kidney Disease Epidemiology (CKD-EPI) Collaboration equation (13), were developed in older adults, although the latter incorporated approximately 650 participants in this age group. These equations are based on serum creatinine levels, which are influenced by alterations in muscle mass and dietary protein intake as well as by chronic disease (common conditions in older adults). Equations based on cystatin C, an alternative marker of GFR, may be advantageous at older ages (14 - 16). However, validation studies using a reference method against a gold standard to measure GFR are scarce. Elderly persons have generally been underrepresented-even in large cross-sectional data sets of equation development for GFR (12 - 13,17 - 18)-and the need for an age-adapted equation has been stated repeatedly (19 - 22).
Accurate assessment of kidney function has several clinical implications, such as adequate adjustment of drug dosing, improved decision making in imaging testing, help in the timing of initiation of renal replacement therapy, evaluation for kidney donation, and the psychological and financial aspect of wrongly labeling someone as having CKD.
The goal of the Berlin Initiative Study (BIS) was to assess kidney function in an elderly population-based cohort by comparing existing equations with a gold standard measurement and to derive a novel estimating equation that would estimate GFR more correctly in persons aged 70 years or older. This is clinically relevant because it would lead to less misclassification of persons with either GFR of 60 mL/min per 1.73 m2 or greater or GFR less than 60 mL/min per 1.73 m2.
Using the BIS cross-sectional database of 570 evaluable persons with iohexol clearance measurements, we developed 2 new equations: 1 based on creatinine only (the BIS1) and 1 based on creatinine and cystatin C (the BIS2) to estimate GFR in elderly participants aged 70 years or older. Compared with current creatinine-based or creatinine- and cystatin C-based equations, the new BIS1 and BIS2 equations showed better precision and excellent agreement with mGFR, especially in a population with an eGFR greater than 30 mL/min per 1.73 m2 (CKD stages 1 to 3). This is important because a validated equation to estimate GFR in older adults, especially in cases of normal or only moderately reduced kidney function, has been lacking.
Several results were striking and may indicate that older adults are a unique population in which traditional assumptions are not necessarily true. We show that in elderly participants, the MDRD study equation yielded higher eGFRs across CKD stages than did the CKD-EPI and especially the Cockcroft-Gault equation. This contrasts with the situation in younger adults in whom implementation of the CKD-EPI equation has reduced CKD prevalence (13) but agrees with current results seen in older adults (39 - 40). The MDRD study equation is known to overestimate GFR in elderly persons with reduced muscle mass (41). Accordingly, prevalence rates of GFR less than 60 mL/min per 1.73 m2 estimated by the Cockcroft-Gault equation were much higher than those estimated by the MDRD study equation, as has been shown earlier in large data sets (42 - 43). The higher prevalence rates of reduced GFR compared with the estimates by the MDRD and CKD-EPI equations were surprising. The MDRD study population is considerably younger and does not include persons older than 70 years, but the CKD-EPI Collaboration study does include older adults. Potential explanations for the different prevalence rates could be that the initial CKD-EPI Collaboration study conditions were more heterogeneous than those of the BIS in terms of samples, marker used for measuring clearance, methodology and way of measuring GFR (plasma and urine clearance), and calibration of serum creatinine. This may explain a larger variation of the data, but whether it relates to bias remains speculative.
The most striking result was that incorporation of cystatin C in the equation decreased the effect of age and sex. This confirms the independence of cystatin C from age- and sex-associated conditions (14,16,36) and may thus make it the preferred laboratory variable to be included in a GFR-estimating equation in an elderly population where reduction in muscle mass is common. Advantages of cystatin C to identify a moderate decline in renal function in elderly persons have been described (44). However, the implementation of an equation that includes cystatin C may be considerably more expensive than those that include creatinine.
Among the other current equations, the Cockcroft-Gault exhibited the smallest bias compared with mGFR. This agrees with a previous study evaluating renal function by using iohexol in a sample of 222 patients aged 65 to 88 years that also presented the Cockcroft-Gault equation as an unbiased estimate (45). However, in terms of precision, the Cockcroft-Gault equation was disappointing because of 23% misclassification.
One strength of the study is that it used primary data of mGFR by using 1 gold standard in a fairly large population-based sample of vulnerable persons with a mean age of 78.5 years. Data in the literature support the notion that plasma clearance has the potential for higher precision than urinary clearance (46 - 47). We only found 1 other study that measured iohexol clearance in a large sample of elderly patients (48). However, this study investigates hospitalized patients only and differs considerably in terms of methodology of clearance measurements. A further strength of the BIS is its design of prospective data collection, including isotope dilution mass spectrometry-traceable creatinine as well as standardized cystatin C values. Iohexol samples were collected by the same well-trained staff and analyzed in Berlin within 7 days of collection. These methodological advantages and the potential homogeneity of the BIS population may have contributed to the high precision of the BIS2 equation. Given the high cost of cystatin C analysis, we also derived an alternative creatinine-based equation (BIS1).
Several limitations deserve mention. First, our data have not been validated in an external validation sample. An equation always performs best in the data set from which it was derived (49), especially with regard to bias. Therefore, a future external validation by an independent investigator is needed. Second, the subsample of 610 participants was not randomly drawn from the entire cohort of 2073 but was asked to have iohexol measured after enrollment in the BIS, potentially introducing selection bias. When the MDRD study equation was used (Appendix Table 6), the prevalence of eGFR less than 60 mL/min per 1.73 m2 in our iohexol subsample was approximately 7% less than the prevalence rates in other community-dwelling elderly cohorts (7,50). The rate of participants with albuminuria was similar to another representative sample (51). Thus, we were able to measure GFR in older adults with normal or mild to moderately reduced GFR, which are the population that has generated the most debate about possible CKD misclassification. Further, we believe that potential responder bias would not have had a considerable effect on equation development. For reasons of feasibility and precision in an elderly cohort with a mean age of 78.5 years, we measured only plasma clearance and not urine clearance. One could argue that when plasma clearance is used, only an overestimation of true GFR exists, especially in persons with reduced kidney function (52). However, early studies suggest that tubular secretion of iohexol is negligible (53 - 54). An overestimation would increase the magnitude of the difference between our mGFR and eGFRs. Third, BIS includes only white participants with mild to moderately reduced kidney function; thus, we cannot necessarily extend these results to other ethnicities or to patients with more severe kidney function. Fifth, we use iohexol and not iothalamate, which was the gold standard for development of the MDRD study and CKD-EPI equations and thus cannot rule out differences in measurement because of diverse markers. Finally, the high prevalence rates of GFR less than 60 mL/min per 1.73 m2 may not reflect true CKD. Because this is a cross-sectional analysis, we do not have repeated creatinine measurements in our sample to indicate chronicity. The high prevalence rates may continue the debate about current cutoffs used to define CKD in older adults (55). Whereas our cross-sectional analysis cannot answer this question, we hope that longitudinal follow-up data from BIS will help resolve whether CKD staging in older adults needs to be redefined.
In summary, iohexol measurements revealed lower GFRs than predicted by current equations used to estimate GFR in participants aged 70 years or older who have normal or mild to moderately reduced kidney function. The newly developed BIS equations may provide more precise and accurate tools for estimating GFR in this age group.
Total Iohexol Subsample
The goal to include 600 persons in the iohexol subsample was reached: 610 of 2073 participants of the BIS agreed to have iohexol clearance measured. Iohexol clearance measurement was feasible in older adults, and no participant had adverse events. From the 610 participants originally measured, 40 were excluded because of corrupt measurement of the gold standard (27 because of an incomplete number of iohexol measurement points; 12 because of an insufficient fit of the Schwartz model for iohexol measurements; and 1 with an outlying mGFR of 300 mL/min per 1.73 m2, as determined from the iohexol measurements), leaving a sample of 570 for the final analysis (Appendix Table 3).
Table 1 shows the main characteristics of the total iohexol subsample. Mean age was 78.5 years, most were men (57.2%), one quarter had diabetes, more than three quarters had hypertension, and nearly one third was overweight (body mass index >30 kg/m2). Associations with eGFR (Table 1) and mGFR (Appendix Table 4) were similar. Overall mean mGFR was 60.3 mL/min per 1.73 m2, whereas all means of GFR estimated by using existing equations showed values that were 2.5 to 11.5 mL/min per 1.73 m2 higher.
Equation Development and Description of the BIS Equations
Appendix Table 5 documents the development of the BIS equations and shows the performance of the models developed in the learning sample. The final BIS2 equation for estimating GFR includes serum creatinine, serum cystatin C, sex, and age: BIS2 = 767 x cystatin C-0.61 x creatinine-0.40 x age-0.57 x 0.87 (if female) (Appendix 2). For practicability, we developed a creatinine-based equation that did not include cystatin C: BIS1 = 3736 x creatinine-0.87 x age-0.95 x 0.82 (if female).
No further variables were included in the equation, according to the criteria described above. Moreover, using 2 different slopes for creatinine (as was done for the CKD-EPI equation) did not improve our equation (Appendix Figure).
Comparison of eGFR Equations and mGFR in the Validation Sample
Figure 1 shows the boxplot of mGFR and eGFR, according to the eGFR equations. The mGFR yielded a median of 61 mL/min per 1.73 m2, as did both new BIS equations. Apart from Cockcroft-Gault adjusted for body surface area (62 mL/min per 1.73 m2) and the CysC2 equation (63 mL/min per 1.73 m2), all other equations considerably overestimated mGFR (MDRD, 73 mL/min per 1.73 m2; CKD-EPI, 73 mL/min per 1.73 m2; CysC1, 70 mL/min per 1.73 m2; and CysC3, 69 mL/min per 1.73 m2) (Appendix Table 6).
Age dependence of GFR estimated by the BIS1 and BIS2 in the study sample is displayed in Figure 2 and shows a very similar pattern of eGFR decline with increasing age. Above the age of 75 years, mean eGFR decreased below the cutoff of 60 mL/min per 1.73 m2.
Table 2 illustrates the bias of existing equations using different statistical parameters. Apart from the BIS2 equation, all other equations had a much larger proportion of false-negatives (wrongly considered >60 mL/min per 1.73 m2) than false-positives (wrongly considered <60 mL/min per 1.73 m2). The total number of misclassifications was smallest for BIS2, followed by the CysC2 equation and BIS1 (33, 43, and 49, respectively, vs. at least 58 for the other equations). Except for the comparison of BIS1 with CKD-EPI, all other statistical tests showed significantly fewer total misclassifications. Accordingly, the P15 and P30 values were largest for the BIS equations, followed by the CysC2 equation.
Appendix Table 7 compares the BIS models with published models more detailed in terms of bias and correlation. Among the creatinine-based equations, the CKD-EPI and new BIS1 equations had the best performance (models 2 and 6). The Cockcroft-Gault equation showed very small bias. However, the overall finding was that cystatin C was superior to creatinine with respect to correlation with mGFR (models 7, 8, 9, and 10 vs. 1, 2, 4, and 5).
Tables 3, 4, and 5 compare reclassification rates for different equations. The BIS1 classifies 55 more participants to a GFR less than 60 mL/min per 1.73 m2 than does the CKD-EPI, and the BIS2 equation classifies 53 more participants to a GFR less than 60 mL/min per 1.73 m2 than does the CysC3 equation (Tables 3 and 4). Table 5 shows that both BIS equations do not differ systematically (15 participants with higher GFRs estimated by BIS2 and 13 participants with higher GFRs estimated by BIS1).