APPENDIX 2. Kidney Function and Associated Conditions in the United States: Methods and Findings From the Third National Health and Nutrition Examination Survey (1988 to 1994)
The Third National Health and Nutrition Examination Survey (NHANES III) data offer the first opportunity to study the prevalence and number of people with chronic kidney disease in a nationally representative sample of the United States. An initial analysis from NHANES III showed that the prevalence of elevated serum creatinine was higher among non-Hispanic blacks than non-Hispanic whites and among older compared to younger individuals.1 The present analysis was undertaken to describe the distribution of estimated glomerular filtration rate (GFR) in the US population. Estimated GFR was calculated using an equation based on each participant’s creatinine, age, sex, and race. The associations of estimated GFR with age, high blood pressure, anemia, and other metabolic and functional abnormalities are also examined to display the range of abnormalities associated with decreased kidney function. The prevalence of microalbuminuria and proteinuria by age, sex, race, and diabetes are tabulated to show the frequency with which these abnormalities are present in the population.
The NHANES III survey, conducted during 1988 to1994 by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention, provides cross-sectional, nationally representative data on the health and nutritional status of the civilian, non-institutionalized US population. 657, 658 Non-Hispanic blacks, Mexican-Americans, as well as the elderly and children were deliberately oversampled in this survey. This oversampling makes it possible to obtain reliable estimates of the distribution of creatinine in the two largest minority groups of the civilian, non-institutionalized US population as well in a broad range of age groups. Standardized questionnaires were administered in the home, followed by a detailed physical examination at a Mobile Examination Center.
Serum was collected at the Mobile Examination Center and creatinine measurements were performed by the modified kinetic Jaffe reaction659 using a Hitachi 737 analyzer (Boehringer Mannheim Corp, Indianapolis, IN) and reported using conventional units (1 mg/dL = 88.4 µmol/L). The coefficient of variation for creatinine determination ranged from 0.2% to 1.4% during the 6 years of study. Data on physiologic variation in creatinine were obtained in a sample of 1,921 participants who had a repeat creatinine measurement. The percent difference between the two creatinine measurements, a mean of 17 days apart, had a mean of 0.2% and standard deviation of 9.7%.5
Estimation of GFR
Estimation of GFR using an equation requires that the calibration of the serum creatinine assay be the same as that in the laboratory where the equation was developed. In NHANES III, serum creatinine was measured in the White Sands Laboratory, where quality control data shows stable calibration over time. The mean serum creatinine for 20 to 39-year-old participants without hypertension or diabetes was 1.14 mg/dL for men and 0.91 mg/dL for women. College of American Pathologists Survey data, released with permission of both laboratories, show that creatinine values in the White Sands laboratory measured during 1992 to 1995 using the Hitachi 737 instrument averaged 0.2 to 0.3 mg/dL higher than values in the Cleveland Clinic measured using the Beckman Astra and Synchron instruments. The latter values were similar to the overall mean of all laboratories for creatinine. These lower values were also close to a gold standard HPLC assay for creatinine in a small validation study.660 This concern lead to a direct comparison of the two laboratories using frozen samples from 212 Modification of Diet in Renal Disease (MDRD) Study participants and 342 Third National Health and Nutrition Examination Survey (NHANES III) participants which were assayed for serum creatinine a second time in each of the study laboratories during the year 2000.661 The GFR estimates in this report are based on creatinine values which were recalibrated using these results. This correction resulted in an estimated median GFR value of 119 mL/min/1.73 m2 at age 20 years (5th and 95th percentiles of 88 and 180 mL/min/1.73 m2). These values are a little lower than published data on normal GFR among young adults. Whether the equations for estimating GFR require further refinement in the normal GFR range is uncertain, but the associations observed support the utility of this estimated GFR. Individuals with very low creatinine values had an estimated GFR that was higher than physiologically plausible. These NHANES participants were assigned a GFR value of 200 mL/min/1.73 m2 as an upper limit to avoid undue influence (0.5% of men, 2.3% of women, 0.7% of non-pregnant women). Pregnant women accounted for approximately half of the women with an estimated GFR>200 mL/min/1.73 m2. Statistics focused on percentiles of the distribution to further decrease the influence of such outliers.
A random spot urine sample was obtained from each participant aged 6 years and older, using a clear catch technique and sterile containers. Urine samples were placed on dry ice and shipped overnight to a central laboratory where they were stored at -20°C. Urinary albumin concentration was measured by solid-phase fluorescent immunoassay.662 Urine albumin was not measured in specimens which contained blood or which tested positive for hemoglobin using qualitative test strips (Multistix). Urine creatinine concentration was measured by the modified kinetic rate Jaffe method using a Beckman Synchron AS/ASTRA analyzer. The inter-assay coefficients of variation for low (1.0 mg/L) and medium (15 mg/L) urine albumin quality control standards were 16% and 10%, respectively. The urinary albumin to urinary creatinine ratio is reported in mg/g. Sex specific cutoffs were used to define microalbuminuria and albuminuria in a single spot urine.
Our estimates reflect the prevalence of albuminuria based on a single untimed urine specimen and include individuals with persistent albuminuria and individuals with intermittent albuminuria. Repeat measurements were obtained in a subset of 1,241 NHANES III participants within 2 months of the initial examination. Agreement between the initial and repeat tests classified as normal, micro, and macro albuminuria was 91.2% (kappa 0.59). Microalbuminuria persisted in the second visit in 57% and macroalbuminuria was present in another 4% of the 110 participants with microalbuminuria on the first exam. The variation in persistence by age group and sex was: 45% at 20 to 39 (n = 22), 59% at 40 to 59 (n = 32), 70% at 60 to 79 (n = 43), and 44% at ≥ 80 years (n = 9), 65% among men (n = 48), and 52% among women (n = 62). Among 1,099 individuals without microalbuminuria at the first visit 5% (n = 56) had microalbuminuria or albuminuria on the second visit.
A 22 analyte biochemistry panel, including serum creatinine, was performed with a Hitachi Model 737 multi-channel analyzer (Boehringer Mannheim Diagnostics, Indianapolis, IN).
Blood pressure measurements were obtained three times during the home interview and another three times during the examination and averaged. Individuals were classified as hypertensive if they had a mean blood pressure ≥ 140 mm Hg systolic, or ≥ 90 mm Hg diastolic, or reported being currently prescribed medication for hypertension treatment.102
Diabetes was defined by history as well as blood glucose values. The primary analysis stratified individuals based on a history of diagnosed diabetes mellitus since this information was available for nearly all individuals and could be used by physicians for risk stratification. Ancillary analyses examined the impact of using the American Diabetes Association (ADA) criteria663 for diabetes mellitus in the subset of individuals who fasted at least 8 hours.
Dietary history was collected using a food frequency questionnaire.
The complex survey design of NHANES III incorporates differential probabilities of selection. To derive national estimates, sampling weights are used to adjust for non-coverage and non-response. All prevalence estimates were weighted to represent the civilian, non-institutionalized US population and to account for over sampling and non-response to the household interview and the physical examination.658 All data analyses were conducted using STATA svy commands for analyzing complex survey design data with 49 strata and 98 primary sampling units.664 A total of 16,589 participants out of 20,050 (82.7%) examined had both blood pressure and serum creatinine data were used as the starting sample for all analyses. The missing data rate was higher in older individuals (individuals missing data were 4 years older), among men than women (17.9% versus 16.7%), and lower among Mexican-Americans and other ethnic groups (14%) than among non-Hispanic whites (19%) and non-Hispanic blacks (18%). These differences were primarily due to missing phlebotomy data. To minimize bias the combined Mobile Examination Center and home exam weights were divided by the proportion of participants missing creatinine data in each of the design age, sex, and race ethnicity strata. This corrects differences in missing data across sampling strata but assumes that data are missing randomly within strata. Missing data rates for other covariates among these individuals varied from 0% for serum albumin to 4.3% for urinary albumin. Survey weights were not further adjusted for missing data in these variables.
Estimated GFR was calculated using the abbreviated MDRD Study equation using the corrected serum creatinine (SCr, mg/dL) as follows:
Estimated GFR (mL/min/1.73 m2) = 186 × (SCr ) -1.154 × (Age) -0.203 × (0.742 if female) × (1.21 if African American)
This equation is also equivalent to:
Estimated GFR (mL/min/1.73 m2) = exp (5.228 - 1.154 × ln (SCr) - 0.203 × ln (Age) - (0.299 if female) + (0.192 if African American)
Estimated GFR was analyzed both as a continuous measure and divided into ranges as described in the guidelines summary of stages of chronic kidney disease.
Continuous analysis of estimated GFR used quantile regression to avoid undue influence of outliers.664 The medians, as well as 95th and 5th percentiles of each covariates (for example, percentiles of blood hemoglobin), were regressed on estimated GFR to show how the middle and top and bottom ends of the covariate distribution varied across the range of GFR. The shape of the association of each covariate with median estimated GFR was modeled using a fifth order polynomial to allow for deviation from a linear association. The regression was further adjusted for age to avoid confounding by age since older individuals have a much lower GFR than younger individuals and older age is also associated with abnormalities in many other covariates. To allow for non-linear associations with age, age adjustment used a fifth order polynomial. The regression was then used to predict values across the range of GFR while fixing age to 60 years. Regressions were weighted using the sampling weights but quantile regression did not allow for explicit incorporation of survey strata into calculation of standard errors. The results are presented in graphical format as regression along with 95% confidence intervals for selected points in the age-adjusted regression. Regressions include all of the relevant data but the graphs are displayed for estimated GFRs between 15 and 150 mL/min/1.73 m2 where the results are most meaningful.
Categorical analysis of estimated GFR divided estimated GFR into four categories according to the proposed stages of chronic kidney disease (≥ 90, 60 to 89, 30 to 59, and 30 to 15 mL/min/1.73 m2). The prevalence of abnormality in each category was calculated for two cutoff values. For example, with blood hemoglobin as the covariate, the cutoffs were <11 g/dL and <13 g/dL. This shows the prevalence of more and less severe abnormalities. Prevalence estimates were age adjusted using logistic regression to avoid confounding by age. Logistic regressions incorporating sample weights and the complex survey design were fit separately for each outcome (for example serum albumin <3.4 g/dL coded as 0/1) with a fifth order polynomial in GFR to fit non-linear associations. The model adjusted for age by including a fifth order polynomial in age. The regression was then used to predict the prevalence for a 60-year-old person with all other covariates unchanged. Bar graphs show this age adjusted prevalence and 95% confidence intervals by estimated GFR.
GFR Estimation in the Canadian Multicentre Cohort
This calculation was made using the abbreviated MDRD formula similarly to methods in NHANES III. Some of the figures label this estimate as “mL/min,” although it should more correctly be labeled “mL/min/1.73 m2.”