Prevalence of Low Glomerular Filtration Rate in Nondiabetic Americans: Third National Health and Nutrition Examination Survey (NHANES III)

Materials and Methods

Results

Of 17,030 NHANES III participants aged ≥20 yr who completed the interview and the examination, 558 were excluded on the basis of ethnic status and 2856 because of diabetes mellitus or incomplete diabetes assessment, leaving 13,616 eligible records. Of these, 365 were incomplete for one or more items required for the calculation of the four different clearance estimates (missing data), leaving 13,251 participants for analysis. Eligible participants with complete and incomplete data are compared in Table 1. Further biochemicaldata for analyzed cases are shown in Table 2. Figures 1 to 4 show the prevalence of levels of renal function by MDRD GFR, Cockcroft-Gault creatinine clearance, Couchoud cutpoints, and 100/(serum creatinine [mg/dl]), respectively. For each age-gender stratum, the number of participants in each decade, and the proportion of those participants whose clearance level fell into each of the different categories is enumerated and illustrated. For the sake of brevity, 95% confidence intervals (CI) are shown for MDRD GFR only (Table 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Nondiabetic adults in NHANES III. Weighted distribution of level of predicted GFR (MDRD categories) in ml/min per 1.73 m² by age (in decades) and stratified by race and gender.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Nondiabetic adults in NHANES III. Weighted distribution of level of creatinine clearance (Cockcroft-Gault [CG] categories) in ml/min by age (in decades) and stratified by race and gender.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Nondiabetic adults in NHANES III. Weighted distribution of level of predicted GFR (Couchoud categories) in ml/min per 1.73 m² by age (in decades) and stratified by race and gender. For (L/creatinine) ×100, the cutpoint of 80 ml/min corresponds to serum creatinine of 1.25 mg/dl; 60 ml/min to 1.67 mg/dl; and 30 ml/min to 3.33 mg/dl.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4. Nondiabetic adults in NHANES III. Weighted distribution of level of predicted GFR (reciprocal of creatinine categories) in dl/mg by age (in decades) and stratified by race and gender. For (L/creatinine) × 100, the cutpoint of 80 ml/min corresponds to serum creatinine of 1.25 mg/dl; 60 ml/min to 1.67 mg/dl; and 30 ml/min to 3.33 mg/dl.

By MDRD GFR, 58% (95% CI, 56 to 60%) of adult nondiabetic (black or white) Americans had a GFR below 80 ml/min per 1.73m², 13% (95% CI, 11 to 14%) below 60 ml/min per 1.73m², and 0.26% (95% CI, 0.19 to 0.33%) below 30 ml/min per 1.73m². Preserved MDRD GFR was more common in older black participants compared with white counterparts. In whites, MDRD GFR was below 80 ml/min per 1.73m² in more than 80% of participants over the age of 40 yr, and below 60 ml/min per 1.73m² in more than 40% of participants over the age of 60 yr. In blacks, MDRD GFR was below 80 ml/min per 1.73m² in more than 50% of participants over the age of 40 yr and below 60 ml/min per 1.73m² in more than 30% of participants over the age of 60 yr. Women tended to have lower MDRD GFR than men, and this effect was most marked in white participants.

The analysis using the Cockcroft-Gault formula produced similar results. However, compared with the corresponding MDRD GFR results, fewer 30- to 60-yr-olds were estimated to have clearance below 80 ml/min, and more participants beyond the age of 60 y were estimated to have a clearance less then 60 ml/min. By the Cockcroft-Gault formula, 39% (95% CI, 37 to 41%) of adult nondiabetic (black or white) Americans had a clearance below 80 ml/min, 14% (95% CI, 12 to 16%) below 60 ml/min, and 0.81% (95% CI, 0.46 to 1.2%) below 30 ml/min.

Compared with the clearance formulae, the Couchoud cutpoints (Figure 3) identified fewer people as having low GFR in all race-gender-age categories. As with the MDRD GFR and Cockcroft-Gault estimates, the proportion of participants in each stratum of low GFR increased with advancing age. Using reciprocal of serum creatinine–produced estimates in which the proportion of women with low clearance was consistently lower than the equivalent proportion for men.

We examined the agreement between the validated prediction equations in categorizing individual participants. The proportion of people in a given MDRD GFR category, who would be categorized into the same numerical category by Cockcroft-Gault creatinine clearance, varied between 35 and 95% and was dependent on level of GFR (Table 4). Higher levels of agreement were noted for both black and white participants at the extremes of each scale (e.g., for white participants, 95% of those with MDRD GFR <30 ml/min per 1.73m² were also found to have Cockcroft-Gault creatinine clearance <30 ml/min; 80% of white participants with MDRD GFR >80 ml/min per 1.73m² had Cockcroft-Gault creatinine clearance in the same range). More differences were noted in the assignment into the 30 to 59 ml/min per 1.73m² and 60 to 80 ml/min per 1.73m² groups, but in general, the recategorization was into an adjacent group, with the exception of 16 participants with MDRD GFR in the 60 to 79 ml/min per 1.73m² category who were found to have Cockcroft-Gault clearances less than 30 ml/min.

Discussion

On the basis of the NHANES III survey conducted between 1988 and 1994, we have described clearance levels for adult nondiabetic black and white Americans. The results obtained are broadly similar for both the MDRD equation and the Cockcroft-Gault formula; large proportions of people in middle age and beyond had low GFR by these estimates. The proportion of participants in each stratum of low GFR increased with advancing age, independent of the method of assessment of GFR.

We have provided age-, gender-, and race-specific estimates of the prevalence of low clearance for four different methods of clearance prediction. The large sample size compared with the Framingham study (15) has allowed us to provide prevalence estimates for different degrees of renal insufficiency; we used three different cutpoints, representing decreasing clearance, for each measure of renal function. We also used three different validated methods of estimating GFR and documented the differences between results expected with each of these methods and for non–gender-specific creatinine cutpoints. We studied nondiabetic participants, as we sought to produce estimates that could be generalized to other populations in whom the prevalence of diabetes might differ from that observed in the United States, and to inform the management of nondiabetic renal disease. The large number of black participants allows the determination of more precise estimates for black Americans than was possible with the Framingham data (15). Reporting black and white participants separately also facilitates generalization to populations of predominantly black and white individuals.

The large proportion of NHANES III participants with a low GFR was an unexpected finding. The Boston Longitudinal Study of Aging documented mean creatinine clearance (by 24-h urine) of 97 ml/min in healthy octogenarians (31,32) and a decline in clearance of 0.75 ml/min per yr; these data are widely cited as representing normal healthy aging (33,34). This rate of decline in renal function was similar to that observed with each year of life in the NHANES III cross-sectional data (MDRD GFR, 0.6 ml/min per 1.73m² per yr; Cockcroft-Gault creatinine clearance, 1 ml/min per yr [data not shown]). The analysis of the predominantly white Framingham population by Culleton et al. (15), which included diabetic participants, suggested that overall 8.9% of men and 8.0% of women had abnormal renal function (defined by creatinine exceeding the 95^th percentile for a healthy subset). Jones et al. (17) noted prevalence of serum creatinine above 1.5 mg/dl in 9.0% of white men and 1.9% of white women (diabetics and nondiabetics were not reported separately). Nissenson et al. (16) estimated the prevalence of gender-specific increases in creatinine (>1.2 mg/dl in women; >1.4 mg/dl in men) using all diabetic and nondiabetic health maintenance organization patients whose creatinine had been measured as the denominator. Acknowledging that this methodology likely leads to overestimation of prevalence, they reported that 9.3% of men and 5.6% of women had at least one creatinine value above these cutpoints in the period studied. In contrast, in the nondiabetic population that we studied, 54% of white men and 68% of white women were estimated to have GFR below 80 ml/min per 1.73m² by the MDRD formula; 25% and 53%, respectively, to have creatinine clearances below 80 ml/min by the Cockcroft-Gault formula. Our results in terms of serum creatinine values are, however, in keeping with previous estimates; for example, we found that creatinine above 1.25 mg/dl occurred in 23% of white men and 3.1% of white women, above 1.67 mg/dl in 1.6% and 0.4%, respectively (Figure 4).

We did not anticipate the finding of such large proportions of the nondiabetic population having clearance estimates below 80 ml/min or ml/min per 1.73m², and we sought alternative explanations for these results. We consider three potential explanations. First, the prediction equations may not perform well in the general population. The MDRD formula was derived in patients with CRI (23), and the Cockcroft-Gault formula in hospitalized white men (24); both have been validated in CRI (23) and in hypertensive African Americans (25,35) but not in community-based samples. However, by categorizing the data so that all patients with normal or near-normal clearance are counted in the >80 ml/min (or ml/min per 1.73m²) stratum, any effect of lack of precision of clearance estimates at the higher (normal) end of the range (underrepresented in validation studies) is minimized. The association we observed between age and clearance may be an artifact arising from the age term, which occurs in both the Cockcroft-Gault and MDRD formulae, a problem sometimes referred to as mathematical coupling. To examine this issue, we performed the additional analyses on the basis of the Couchoud cutpoints (which are gender- but not age-specific) and on reciprocal creatinine (without adjustment for age or gender). As expected, age effects are less prominent in these analyses, but still occur. Comparison of the results using Couchoud cutpoints (Figure 3) with those using reciprocal creatinine cutpoints (Figure 4) confirms that use of creatinine-based cutpoints that are not gender-specific will consistently underestimate renal insufficiency in women, and, if applied, they would lead to substantial underrecognition of low GFR in women. Failure to take gender into account when interpreting serum creatinine may account for the increased risk of late initiation of dialysis in women (36). We believe that using the Couchoud cutpoints might result in overestimation of GFR in older people. Reduction of creatinine production with advancing age is well documented (37); therefore, cutpoints in serum creatinine that disregard age are inherently biologically implausible. Couchoud et al. (26) reported that they had validated their cutpoints in age-specific subgroups, but data were not provided, and it is unlikely that this study had sufficient power to examine this question. In summary, though there is no direct evidence of validity of the formulae in community-based samples, they are not known to be consistently biased (summarized in reference 25), and they represent the best available option for the estimation of clearance in large studies.

The second possible explanation for our findings involves interlaboratory variation. Creatinine measurements from different laboratories, even using the same technique, may differ systematically by as much as 0.1 to 0.2 mg/dl (38–40). A difference of this magnitude between the laboratory used in the validation of the MDRD and Cockcroft-Gault equations and that used in NHANES III would have a large effect on these results. Using measured creatinine, 58% of the total adult nondiabetic black and white US population had MDRD GFR below 80 ml/min per 1.73m², 13% below 60 ml/min per 1.73m², and 0.26% below 30 ml/min per 1.73m². These results were derived from analyses performed according to the analytic guidelines recommended by the NCHS for NHANES III. However, if NHANES III–measured creatinine were systematically as much as 0.2 mg/dl (17.7 μmol/L) higher than MDRD-measured creatinine, these proportions would decrease to 21%, 4%, and 0.17%, respectively. The effect of this correction is most marked in terms of altering the classification of patients at the 80–ml/min per 1.73m² cutpoint. However, even with a correction of this magnitude, the proportion of the population with clearance below 80 ml/min per 1.73m² is still unexpectedly high. Furthermore, if the results presented here are biased (compared with the laboratory used in MDRD validation) by interlaboratory variation, one might anticipate that creatinine results similar to those from the NHANES III reference laboratory would be obtained by at least some clinical laboratories. Results from such laboratories would, if translated into clearances, result in the identification of large proportions of patients with low clearances. Though the importance of issues of measurement methodology and calibration in research studies has been recognized (40), the importance of these factors in the identification of low GFR in community practice has not previously been emphasized. We suggest that recommending that laboratories report a calculated clearance estimate with each creatinine measurement (41) is premature until further study has examined the implications of widespread introduction of this practice.

A third possibility is that limitations of study design have led to biased estimation of clearance. The predictions of clearance reported here are based on single random creatinine measurements without reference to hydration status or diet. We are not aware of any studies of test-retest reliability or day-to-day variation in serum creatinine, but relative stability at a sample level over years has been documented (42), and acute renal failure is a rare event (43). Though the degree of chronicity of low GFR cannot be ascertained with certainty for an individual sampled in NHANES III, it seems unlikely that large numbers of participants would have had significant creatinine elevations (above their baseline) on the day of their evaluation. Sampling was performed during the daytime, when GFR is highest (44). Sampling of noninstitutionalized individuals only, as in NHANES III, would be more likely to exclude people with higher morbidity and bias results toward higher estimates of clearance. We used fasting glucose to exclude people with undiagnosed diabetes. For half the cohort, this was measured in the afternoon. Some people with undiagnosed diabetes, who would likely have been detected had they been studied in the morning are missed when glucose is measured in this way (1.4%) (45). However, this could not have resulted in a large bias. Albumin in NHANES III was measured by the bromcresol purple (BCP) method, whereas albumin was measured by bromcresol green (BCG) in the MDRD derivation and validation. In the general population, BCG results tend to be somewhat lower than BCP, whereas BCP results tend to be lower in uremic patients (summarized in reference 46). Though this might have resulted in some bias, the absolute magnitude of this effect is of the order of a few ml/min per 1.73m² only and applies to the MDRD results only, because albumin is not included in the Cockcroft-Gault formula. Finally, some of the participants in the lowest clearance categories for each measurement method (GFR < 30 ml/min or < 30 ml/min per 1.73m²) would presumably have been receiving renal replacement therapy. This information was not captured in the dataset. However, in comparison with the point-prevalence rate of 697 per million population for ESRD reported by the United States Renal Data System (USRDS) for 1990 (the mid-year of the NHANES III study) (47), we observed prevalences in the range of 2000 to 3000 per million population (Figure 1) in the lowest clearance categories. This suggests that people with ESRD are likely to represent a substantial minority of those in the lowest categories of clearance studies based on these data.

The cross-sectional design implies that the relationship between renal function and age must be interpreted with caution. For this reason, we have reported the relationship between GFR and age as observed prevalences by decade, rather than attempting an explanatory regression analysis.

The findings of consistently higher prevalence of low GFR (by MDRD formula) in white compared with black participants is intriguing. The formula upwardly adjusts clearance estimates on the basis of black race; therefore, it may simply reflect a numerical artifact (the difference is not seen in clearance by Cockcroft-Gault formula). However, since the MDRD formula has been adequately validated in black people (25), the finding may represent a true difference. If this is the case, then given the excess prevalence of ESRD in black Americans (47), it is possible that reduced GFR of any level in a black person carries a substantially worse prognosis than the equivalent for a white person.

With these caveats, we believe that our results have important implications for our understanding of renal disease and for the organization of renal services. The difference in order of magnitude between the prevalence of low GFR and the prevalence of ESRD (e.g., for white males in their 60s, 0.2% have ESRD (1), 0.6% have MDRD GFR below 30 ml/min per 1.73m², but 27% have GFR between 30 and 59 ml/min per 1.73m²) prompts the hypothesis that not all low GFR carries the same prognosis. This may be particularly important in the evaluation of older people. Though the Boston Longitudinal Study of Aging documented a clearance of 97 ml/min in healthy octogenarians (31,32), these participants were highly selected (self-recruited middle-aged white men, generally of middle and upper socioeconomic status, without major diseases) and for this reason may not be representative of either normal aging or normal healthy aging. We have shown that low GFR may be very common in older age. Data for older people from other studies are consistent with our findings (15,37,48,49). We postulate that low GFR in some older people may not be associated with the same risks for progression that might coexist with the equivalent GFR in a younger person. This might be because the range of pathologic diagnoses is different in older people or because loss of GFR due to senescence (50) carries a better prognosis than that of other pathologic entities or because the metabolic load on the impaired kidney is also reduced with increasing years (51,52), which might ameliorate the pathologic processes (53,54) associated with single-nephron hyperfiltration in remnant kidneys. In addition, the risk of ESRD will be less than in a younger person, because of the shorter expected period of subsequent life and the competing risks of other diseases.

GFR in the 30- to 80-ml/min range in older people is often accompanied by a serum creatinine within the laboratory normal range (37,41) and may go unrecognized, uninvestigated, and unreferred (41). Only a minority of patients with an elevated creatinine are referred to nephrologists (55,56). We postulate that referred populations are selected for more progressive disease, more advanced disease, and perhaps also on the basis of fewer competing risks (56) compared with those not referred. Referral to a nephrologist is needed when a specifically treatable etiology is possible, when CRI is progressive, or when therapeutic goals cannot be met. Patients with very low GFR (perhaps below 30 ml/min; references 13,57) should also be referred because of the possibility of dialysis treatment and the probable benefits of predialysis nephrologic care (8–11). Strategies to enhance referral in these circumstances warrant immediate attention. However, it seems that low GFR in the range of 30 to 80 ml/min per 1.73m² may be much more common in the population than one would anticipate from previous studies of healthy aging. It may not be valid to generalize findings from studies in referred populations to these unreferred patients. We believe that further community-based research is needed into the prevalence of risk factors for progression of disease, the rate of progression of disease, and optimal methods of identifiying those at risk for progression to ESRD.

In summary, we believe that this work has two major implications. First, widespread implementation of predicted clearance in laboratory reports has the potential to increase identification of low GFR by primary physicians and potential referral to nephrologists by orders of magnitude. Further study is required to determine the characteristics of unreferred patients with low GFR, and the optimal further assessment of such patients is not known. For these reasons, we believe that laboratory-generated clearance reports cannot be recommended without further study. Second, low clearance may be much more prevalent than previously suspected and may not always reflect high risk of progression to ESRD. Optimal diagnostic and management strategies may differ for unreferred people with low clearance compared with those currently referred at similar clearance levels. Research into the identification of currently unreferred people who are at high risk for progression to ESRD is urgently needed.