ORIGINAL INVESTIGATIONS Pathogenesis and Treatment
Calibration and Random Variation of the Serum Creatinine Assay as Critical Elements of Using Equations to Estimate Glomerular Filtration Rate Josef Coresh, MD, PhD, Brad C. Astor, PhD, Geraldine McQuillan, PhD, John Kusek, PhD, Tom Greene, PhD, Frederick Van Lente, PhD, and Andrew S. Levey, MD ● Equations using serum creatinine level, age, sex, and other patient characteristics often are used to estimate glomerular ﬁltration rate (GFR) in both clinical practice and research studies. However, the critical dependence of these equations on serum creatinine assay calibration often is overlooked, and the reproducibility of estimated GFR is rarely discussed. We address these issues in 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 assayed for serum creatinine level a second time during November 2000. Variation in serum creatinine level was assessed in 1,919 NHANES III participants who had serum creatinine measured on two visits a median of 17 days apart. Linear regression was used to compare estimates. Calibration of serum creatinine varied substantially across laboratories and time. Data indicate that serum creatinine assays on the same samples were 0.23 mg/dL higher in the NHANES III than MDRD study. Data from the College of American Pathologists suggest that a difference of this magnitude across laboratories is not unusual. Conversely, serum creatinine assays an average of 2 weeks apart have better precision (SD of percentage of difference in estimated GFR, 15%; 90% of estimates within 21%). Errors in calibration make little difference in estimating severely decreased GFR (<30 mL/min/1.73 m2), but result in progressively larger differences at higher GFRs. Both clinical and research use of serum creatinine or equations to estimate GFR require knowledge of the calibration of the serum creatinine assay. © 2002 by the National Kidney Foundation, Inc. INDEX WORDS: Glomerular ﬁltration rate (GFR); prediction equations; serum creatinine.
ERUM CREATININE is the most widely used assay to measure the presence and progression of chronic kidney disease.1 Equa-
From the Departments of Epidemiology and Biostatistics, The Johns Hopkins University School of Hygiene and Public Health; Department of Medicine, The Johns Hopkins University School of Medicine; Welch Center for Prevention, Epidemiology and Clinical Research, The Johns Hopkins Medical Institutions, Baltimore; Division of Health Examination Statistics, National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville; and the National Institute of Diabetes, Digestive and Kidney Diseases, The National Institutes of Health, Bethesda, MD; Cleveland Clinic Research Foundation, Cleveland, PA; and the Department of Medicine, Tufts University School of Medicine, Division of Nephrology, New England Medical Center Hospital, Boston, MA. Received June 4, 2001; accepted in revised form November 30, 2001. Supported in part by a grant from the National Kidney Foundation as part of the Kidney Disease Outcomes Quality Initiative; and National Institutes of Health grants no. DK48362 and RR00722 (J.C.); grant no. UO1 DK35073 (T.G.); grants no. RO1 DK53869 and UO1 DK35073 (A.S.L.); and grant no. T32HL7024 (B.C.A.). Address reprint requests to Josef Coresh, MD, PhD, 2024 E Monument, Baltimore, MD 21205. E-mail: [email protected]
© 2002 by the National Kidney Foundation, Inc. 0272-6386/02/3905-0002$35.00/0 doi:10.1053/ajkd.2002.32765 920
tions based on serum creatinine level, age, sex, and other variables perform much better at predicting glomerular ﬁltration rate (GFR) than serum creatinine level alone.2 The Cockcroft-Gault equation was developed to estimate creatinine clearance,3 but has been used to estimate GFR with relatively good precision.4 Recently, several equations have been developed using the Modiﬁcation of Diet in Renal Disease (MDRD) study to estimate GFR directly.2,5 Any equation using serum creatinine level to estimate GFR is critically dependent on the calibration and reproducibility of the serum creatinine assay. Understanding the principles of estimating GFR from serum creatinine level is critical to ongoing efforts to increase the diagnosis and treatment of patients with chronic kidney disease before they reach kidney failure. The most widely used method to measure serum creatinine is based on the modiﬁed kinetic rate Jaffe reaction implemented in many autoanalyzers. The picric acid Jaffe reaction was recognized very early to overestimate serum creatinine level because of “noncreatinine chromogens” in healthy individuals by approximately 20% to 30% relative to high-performance liquid chroma-
American Journal of Kidney Diseases, Vol 39, No 5 (May), 2002: pp 920-929
tography and mass spectroscopy measurements.6,7 Conversely, the amount of noncreatinine chromogens in urine is negligible. This differential overestimation in serum leads to a systematic underestimation of creatinine clearance (Ccr) computed from serum (Scr) and urine creatinine (Ucr) and urine volume (V) measurements (Ccr ⫽ UcrV/ Scr). By coincidence, in healthy individuals, this error is approximately similar to the difference between true creatinine clearance and GFR because of the tubular secretion of creatinine (average, 12% ⫾ 10% [SD]) among 110 individuals with a normal GFR studied by Lemann et al8 and 24% ⫾ 24% among 62 older adults studied by DeSanto et al.9 Thus, overestimation of serum creatinine level because of noncreatinine chromogens provided a convenient compensation for the tubular secretion of creatinine. Calculated creatinine clearance yielded a low estimate of true creatinine clearance, but a better estimate of GFR. Advances in clinical chemistry led to the development of the modiﬁed kinetic rate Jaffe reaction, as well as enzymatic methods for measuring creatinine, which can be calibrated to avoid measurement of noncreatinine chromogens. Calibration refers to systematic standardization of the assay compared with reference materials. Ideally, assays can be calibrated to a deﬁnitive method for the measurement of serum creatinine by isotope dilution mass spectrometry, reference laboratories, or methods that have been compared with the deﬁnitive method. Calibration to measure “true” serum creatinine yields serum creatinine values that are 0.2 to 0.3 mg/dL lower than methods that include noncreatinine chromogens and creatinine clearance estimates that are noticeably higher than measured GFR. As a result, such calibration would be problematic in a clinical setting if timed urine collections were used to estimate creatinine clearance as a direct measure of GFR without explicitly correcting for creatinine secretion. This convenient compensation has been detrimental to the standardization of serum creatinine assays. The College of American Pathologists (CAP) conducts regular surveys comparing serum creatinine assay methods across clinical chemistry laboratories. Analysis of 1994 survey data and fresh frozen plasma assays in 700 laboratories compared with deﬁnitive methods at the Na-
tional Institutes of Standards and Technology indicates that differences in calibration of serum creatinine assays account for 85% of the difference between laboratories in serum creatinine measurements, overwhelming random measurement error. The laboratories surveyed also overestimated serum creatinine by an average of 13% compared with a reference method. Variations across laboratories in both calibration and overestimation were greater for serum creatinine than for any of the other 10 analytes examined. In comparison, reproducibility of serum creatinine measures within a laboratory was much better than for many other analytes. The average coefﬁcient of variation for serum creatinine was 8%.10 Large differences in calibration of the serum creatinine assay across laboratories and often within a laboratory over time adversely impact clinicians’ abilities to: (1) diagnose cases of mild chronic kidney disease, and (2) interpret changes in serum creatinine levels over time in the same patient. This also adversely affects the conduct of research and patient education about chronic kidney disease. We show the magnitude of the difference in serum creatinine calibration between two leading laboratories and its impact on estimates of GFR for both a general population sample of participants in the Third National Health and Nutrition Examination Survey (NHANES III) and patients with kidney disease in the MDRD study. CAP survey data are used to show how calibration can change over a decade. Finally, we estimate the reproducibility of estimates of GFR using equations including serum creatinine and other patient characteristics. METHODS Serum creatinine was measured as part of both the MDRD study (1989 to 1991) and NHANES III (1988 to 1994). Both studies were funded by The National Institutes of Health and reported good quality control, providing an opportunity to show the issues in applying an equation developed for estimating GFR based on serum creatinine assays in one laboratory to serum creatinine assays obtained in another laboratory. Repeated analysis of frozen serum from both studies allows for examination of interlaboratory differences for both a general population sample with a low mean serum creatinine level and a group of patients with kidney disease with a greater mean serum creatinine level. Serum collected during the screening period of the MDRD study from 222 patients was thawed and analyzed again during November 2000 in both of the core laboratories for the MDRD and NHANES III studies. Similarly, serum from 354 participants
CORESH ET AL
in the NHANES III was thawed and analyzed in both laboratories. In the NHANES III, serum was collected at the Mobile Examination Center and creatinine measurements were performed at the White Sands Research Center laboratory by means of the modiﬁed kinetic Jaffe reaction11 using an Hitachi 737 analyzer (Boehringer Mannheim Corp, Indianapolis, IN) and reported using conventional units (1 mg/dL ⫽ 88.4 mol/L). The coefﬁcient of variation for creatinine determination was 2.7% at 1.7 mg/dL, 2.1% at 3.5 mg/dL, and 2.0% at 4.4 mg/dL during the study, with stable quality control. Data for physiological variation in creatinine were obtained in a sample of 1,921 participants who had a repeated creatinine measurement. The assay had very stable quality control measures during the duration of the study (Laboratory Procedures Manual 1996). In the MDRD study, serum was collected as part of study visits, and creatinine measurements were performed at the Cleveland Clinic laboratory by means of the modiﬁed kinetic Jaffe reaction11 using a Beckman Astra CX3 autoanalyzer (Brea, CA). The coefﬁcient of variation for creatinine determination was 4.3% at 1.0 mg/dL and 1.5% at 5.4 mg/dL during the study, with stable quality control. CAP Comprehensive Chemistry Survey data from both laboratories were requested with permission of the laboratories and reviewed. Data were available from 1992 to 1999. The CAP Comprehensive Chemistry Survey includes approximately 7,000 laboratories and is conducted to assess and increase the accuracy of laboratory assays. Participating laboratories receive ﬁve samples of lyophilized human plasma quarterly. Laboratories receive a report comparing their assay results with those of other laboratories using the same method. Individual values are considered within the acceptable range if they are within 0.3 mg/dL or 15% of whichever is greater of the mean for the same method. Results of gold standard methods for the measurement of serum creatinine are not reported back as a part of this survey because assays of lyophilized samples may differ from assays of fresh or frozen plasma. Such differences have been termed matrix effects and may differ across different analytical methods.10,12
GFR Estimation Estimated GFR was calculated using the abbreviated MDRD equation5 because it uses fewer variables while performing nearly as well as other equations developed in the MDRD study that use more serological measures.2 The equation is as follows: Estimated GFR ⫽ 186.3 * 共sCr兲⫺1.154 * age⫺0.203 * 共0.742 if female兲 * 共1.21 if black兲 This equation also is equivalent to: Estimated GFR ⫽ exp共5.228 ⫺ 1.154 * ln共sCr兲 ⫺ 0.203 * ln共age兲 ⫺ 0.299 if female ⫹ 0.192 if black兲 GFR also was estimated using the Cockcroft-Gault (CG) equation estimate of creatinine clearance (CrCl) adjusted for body surface area (BSA) as follows:
CG C cr ⫽ 共140 ⫺ age兲/sCr * weight/72 * 1.73/BSA where weight is measured in kilograms and BSA is estimated using the DuBois formula.13 Estimated GFR was analyzed as both a continuous measure and divided into ranges as 15 to 29, 30 to 59, 60 to 89, and 90⫹ mL/min/1.73 m2.
Statistical Analysis Serum creatinine measurements on the same samples at different laboratories or times were plotted against one another and compared using linear regression. Residuals were graphed, and outliers were identiﬁed. Differences between paired serum creatinine measurements were calculated, along with corresponding SDs, SEs, and 95% conﬁdence intervals. Because analyses supported a relatively constant bias independent of serum creatinine level, the mean difference was used as the primary measure. All data analyses were conducted using STATA (College Station, TX) and SAS (Cary, NC) version 8.
Figure 1 shows the comparison of original serum creatinine measurements in the MDRD study to duplicate measurements on thawed samples in the Cleveland Clinic and White Sands laboratories during 2000. Overall, agreement is very good, with regressions very similar to the line of identity. It is noteworthy that 10 samples had a difference of 1.0 mg/dL or more between at least two of the three measurements, greater than 4 SDs of the difference without these samples. These outliers were deleted from further analysis because they were believed to be caused by sample evaporation, insufﬁcient sample mixing, or other handling issues, rather than differences between serum creatinine assays. Most of these outliers were abnormal for both the repeated values; six values being low for the Cleveland Clinic repeated measurements and high for the White Sands repeated measurements, consistent with inadequate sample mixing. Figure 2 shows results similar to Fig 1 for original and repeated serum creatinine measurements on 345 NHANES samples. Three outliers in which two of the three serum creatinine values differed by 1 mg/dL or more were excluded from analysis for the same rationale. Table 1 lists serum creatinine assay values after excluding outliers. The mean serum creatinine level in the MDRD study was 2.635 mg/dL. Repeated assays of these samples had a higher mean at the Cleveland Clinic (difference, 0.0778 mg/dL) and a lower mean at the White Sands
Fig 1. Scatter plot of serum creatinine (sCr) values assayed in 2,000 (new) in the (A) Cleveland Clinic and (B) White Sands laboratories compared with values measured during the MDRD study (old). Outliers, deﬁned as values in which any pair of serum creatinine measurements differs by 1 mg/dL or more, are marked with ⴙ and excluded from the regression and further analysis in both panels.
laboratory (difference, ⫺0.117 mg/dL). The NHANES samples had a lower mean serum creatinine level of 1.095 mg/dL. Repeated assays of these samples had a lower mean at the Cleveland Clinic (difference, ⫺0.141 mg/dL) and an even lower mean at the White Sands laboratory (difference, ⫺0.358 mg/dL). Repeated assays allowed for two independent calculations of the mean difference between the old White Sands
and Cleveland Clinic measurements. These two calculations were very similar and averaged to an estimate of 0.23 ⫾ 0.01 (SE) mg/dL higher serum creatinine values at the White Sands laboratory during the NHANES III than at the Cleveland Clinic laboratory during the MDRD study. This estimate agrees very well with the mean difference between CAP samples measured in the two laboratories between 1992 and 1995 (49
CORESH ET AL
Fig 2. Scatter plot of serum creatinine (sCr) values assayed in 2,000 (new) in the (A) Cleveland Clinic and (B) White Sands laboratories compared with values measured during the NHANES III study (old). Outliers, deﬁned as values in which any pair of serum creatinine measurements differs by 1 mg/dL or more, are marked with ⴙ and excluded from the regression and further analysis in both panels.
samples; mean difference, 0.24 ⫾ 0.01 [SE] mg/dL). The average difference was limited to this period because a plot of the difference between each laboratory and the mean of all serum creatinine methods showed that the White Sands laboratory method was recalibrated in 1995 (Fig
3). The difference in calibration is large, but is lower than the limit of acceptability for proﬁciency testing for serum creatinine measurements, which is ⫾0.3 mg/dL or ⫾15%, whichever is greater.12 Figure 3 also shows that the bias is larger than the variability in individual serum
Table 1. Serum Creatinine Measurements in the MDRD and NHANES III Samples During the Original Study and Samples Thawed in 2000 (New) and Estimates of the Calibration Difference Between the Cleveland Clinic and White Sands Laboratories MDRD Study Samples
CC WS New between-lab difference Original calibration difference
2.635 (0.109) — —
2.713 (0.114) 2.518 (0.118) 0.195 (0.016)
NHANES III Samples Within-Lab Difference Over Time
0.078 (0.011) —
New between-lab difference ⫺ Within-lab difference over time (CC) ⫹ Within-lab difference over time (WS) ⫽ 0.195 ⫺ 0.078 ⫹ (⫺0.358) ⫽ 0.240 (0.017) Average original calibration difference Average difference from CAP
— 1.095 (0.025) —
0.954 (0.025) 0.737 (0.025) 0.217 (0.008)
Within-Lab Difference Over Time
— ⫺0.358 (0.007)
New between-lab difference ⫺ Within-lab difference over time (CC) ⫹ Within-lab difference over time (WS) ⫽ 0.217 ⫺ 0.078 ⫹ (⫺0.358) ⫽ 0.219 (0.012) 0.230 (0.012) 0.237 (0.015)
NOTE. SEs in parentheses. Abbreviations: CC, Cleveland Clinic; WS, White Sands.
creatinine measurements, but smaller than the variability of the average of ﬁve measurements. We also developed a linear calibration equation, which did not assume a slope of 1.0. The slope of all regression lines in Figs 1 and 2 were close to 1.0 (range, 0.967 to 1.068). However, 95% conﬁdence intervals were very tight and did not include 1.0 for all but Fig 2A. Averaging regression results from the two redundant assays results in a calibration equation of: Old MDRD ⫺ old NHANES serum creatinine ⫽ ⫺0.030 ⫺ 0.081 * old NHANES creatinine. The equivalent regression using CAP data resulted in an inter-
Fig 3. Serum creatinine values in the White Sands and Cleveland Clinic laboratories minus the mean serum creatinine level in all laboratories participating in the CAP survey for that quarter. The same ﬁve samples are analyzed in all participating laboratories each quarter. Lines indicate the average of the difference for each quarter at White Sands and Cleveland Clinic during 1992 to 2000.
cept of ⫺0.16 and slope of ⫺0.025. Although regressions differed, they yielded a correction of ⫺0.24 and ⫺0.23 at the mean serum creatinine level in MDRD study samples (2.63 mg/dL). Analysis including outliers resulted in a similar overall constant correction of ⫺0.24 despite different values in the two redundant calculations. In all further analyses, the constant correction of ⫺0.23 mg/dL was used. The magnitude of the difference in calibration of serum creatinine assays between the two laboratories (0.23 mg/dL) would result in a large difference in estimates of GFR regardless of the
equation used. The magnitude of this recalibration is similar to the difference in serum creatinine levels between men and women. For a creatinine level of approximately 1.0 mg/dL in the NHANES III, this difference would result in a GFR that is approximately 23% higher after recalibration. For a 60-year-old white man, the MDRD-estimated GFR with a serum creatinine level of 1.0 mg/dL in the NHANES III would be 81 mL/min/1.73 m2 before recalibration and 110 mL/min/1.73 m2 after recalibration. A similar man with a serum creatinine level of 1.5 mg/dL would have an estimated GFR of 51 mL/min/ 1.73 m2 before recalibration compared with 61 mL/min/1.73 m2 after recalibration. The impact of a constant calibration bias decreases with increasing serum creatinine level: a creatinine level of 2.0 mg/dL in a 60-year-old man would lead to GFR estimates of 36 and 42 mL/min/1.73 m2 before and after recalibration. A creatinine level of 3.0 mg/dL would lead to GFR estimates of 23 and 25 mL/min/1.73 m2 before and after recalibration. Figure 4 shows the decreasing effect of a constant calibration bias with increasing serum creatinine level for individuals with very different creatinine production rates, a 40-year-old black man and a 70-year-old white woman. The effect of the calibration bias on individuals with a high serum creatinine level is negligible. However, the effect is substantial in the high-normal range of serum creatinine levels, which is the
CORESH ET AL
relevant range for diagnosing moderately impaired kidney function (GFR ⬍ 60 mL/min/1.73 m2). Table 2 lists serum creatinine thresholds for detecting a GFR of 60 mL/min/1.73 m2 implied by the MDRD and Cockcroft-Gault equations (calculations using the published equation yield values calibrated to the Cleveland Clinic laboratory). Both equations support using a lower threshold in older individuals. The threshold is similar across equations for white 60-year-old men at 1.3 mg/dL and 70-year-old women at 1.0 mg/dL. However, the decrease with age is steeper for the Cockcroft-Gault equation, resulting in higher values for younger individuals (1.47 mg/dL for a 30-year-old white man using the MDRD equation versus 1.83 mg/dL using the Cockcroft-Gault equation). In addition, the MDRD equation suggests that the threshold for blacks should be higher: between 0.17 and 0.27 mg/dL higher, depending on age and sex. All these values fall within the range in which common calibration errors have an important effect. A sample of 1,919 NHANES III participants had repeated creatinine data a mean of 17.5 ⫾ 8.0 (SD) days after the initial examination. Repeated creatinine measurements showed good agreement with initial creatinine measurements. The percentage of difference between the two measurements had a mean of 0.2% ⫾ 9.7% (SD). The percentage of difference was independent of the time difference between visits and the absolute creatinine level. The impact of this level of
Fig 4. Estimated GFR for two sample individuals using the MDRD abbreviated equation in a laboratory with no bias and a laboratory with a 0.23-mg/dL bias in estimating serum creatinine levels compared with the Cleveland Clinic.
Table 2. Serum Creatinine Corresponding to an Estimated GFR of 60 mL/min/1.73 m2 by the MDRD and Cockcroft-Gault Equations MDRD Equation EuropeanAmerican
30 40 50 60 70 80
1.47 1.39 1.34 1.30 1.26 1.23
1.13 1.08 1.03 1.00 0.97 0.95
1.73 1.65 1.58 1.53 1.49 1.46
1.34 1.27 1.22 1.18 1.15 1.12
1.83 1.67 1.50 1.33 1.17 1.00
1.56 1.42 1.28 1.13 0.99 0.85
NOTE. MDRD estimated GFR ⫽ exp(⫺5.228 ⫺ 1.154 * ln(sCr) ⫺ 0.203 * ln(age) ⫺ 0.299 if female ⫹ 0.192 if black or estimated GFR ⫽ 186.3 * (sCr) ⫺ 1.154 * age ⫺ 0.203 * (0.742 if female) * (1.21 if black). Cockcroft-Gault CrCl ⫽ (140 ⫺ Age)/sCr * (weight/72) * (0.85 if female) * BSA/1.73 m2. Calculations in this table assume a weight of 72 kg and BSA of 1.73 m2.
variation caused by biological and individual week-to-week variability in the context of a single well-calibrated laboratory is listed in Table 3. Estimated GFR categories are deﬁned based on draft guidelines for deﬁning stages of kidney disease (severely [15 to 29 mL/min/1.73 m2] and moderately [30 to 59 mL/min/1.73 m2] decreased GFR categories are combined, mildly decreased GFR [60 to 89 mL/min/1.73 m2] is shown separately, whereas normal and higher estimates of GFR [⬎90 mL/min/1.73 m2] are subdivided at 30-mL/min/1.73 m2 increments). Repeated GFR estimates based on a second serum creatinine measurement show some regression to the mean; repeated measures of low initial values are likely to be higher, and repeated measures of high values are likely to be lower. The SD of the percentage of difference between the two estimated GFRs remained relatively constant at approximately 15%, reﬂecting greater imprecision at higher estimated GFRs. The absolute value of the percentage of difference in GFR between two estimates shows agreement without regard to the direction of the difference; 50th and 90th percentiles of the absolute value of the percentage of difference were 10.7% and 21.2%. These values are similar to the 12% and 30% reported for the same statistics comparing estimated and measured GFRs in the validation of the four-variable MDRD equation.5
Given appropriate calibration of the serum creatinine assay, estimates of GFR can be reproducible (SD of estimates 2 weeks apart, 15%). For example, at moderately decreased GFRs (GFR, 30 to 59 mL/min/1.73 m2), estimates of GFR based on serum creatinine level are precise enough to allow for diagnosis and clinical management. At this GFR range, a follow-up estimated GFR will be within 27% and a measured iothalamate GFR will be within 30% of the original estimate most of the time, corresponding to 13.5 and 15 mL/min/1.73 m2 at a GFR of 50 mL/min/1.73 m2. This level of precision allows for a high level of conﬁdence that individuals with a GFR in this range have kidney disease. Averaging two or more estimates of GFR should lead to further increases in the precision of estimates. Serum creatinine thresholds for diagnosing moderate chronic kidney disease (GFR ⬍ 60 mL/min/1.73 m2) vary by age, sex, race, and laboratory used. These thresholds are far lower than a serum creatinine level of 2.0 mg/dL, which often has been used as the clinical threshold to diagnose kidney disease. For European American women, the MDRD equation suggests the thresholds are close to 1.0 mg/dL, often considered a normal serum creatinine level. A GFR of 60 mL/min/1.73 m2 in a 30-year-old European American woman corresponds to a Table 3. Percentage of Difference Between Two Estimates of GFR Among the NHANES III Participants Who Had Two Measurements of Serum Creatinine a Median of 17 Days Apart
Estimated GFR at Baseline
Estimated GFR Category (mL/min/1.73 m2)
30-59 60-89 90-119 120-149 150⫹ Overall
98 573 816 311 117 1919
Difference* Between Second Visit and Baseline Estimated GFR (%)
Absolute Value of % Difference
49.7 7.6 77.6 8.0 103.8 8.5 131.7 8.0 170.2 15.7 101.6 28.9
7.8 4.7 0.5 ⫺4.6 ⫺9.1 0.7
15.7 15.2 13.4 14.0 12.7 14.7
9.9 10.7 9.4 13.1 13.0 10.7
27.0 24.2 20.5 23.4 29.0 21.2
*(Second estimated GFR ⫺ baseline estimated GFR)/baseline estimated GFR * 100.
serum creatinine level of 1.13 mg/dL (could be reported as 1.2 to 1.5 mg/dL in different laboratories). Analysis of the entire NHANES III data shows that very few of these women (0.3% among 25- to 35-year-olds) have a serum creatinine level greater than this level (data not shown). For individuals with a serum creatinine level less than 2.0 mg/dL, the existence of substantial variation in the calibration of serum creatinine assays across laboratories complicates the estimation of GFR from serum creatinine and patient variables. We show the impact of differences in serum creatinine calibration between two leading clinical chemistry laboratories. The bias in estimating GFR was more than 25% at a GFR of 100 mL/min/1.73 m2 and decreased to 16% at a GFR of 50 mL/min/1.73 m2. The bias would be approximately 8% at a GFR of 25 mL/min/1.73 m2. Therefore, although calibration is relatively unimportant in severely decreased GFR (⬍30 mL/min/1.73 m2), it is important in estimating moderate decreases in GFR (30 to 60 mL/min/ 1.73 m2) and critical in estimating mild decreases in GFR (60 to 90 mL/min/1.73 m2). It is important to consider the extent to which differences shown between two speciﬁc laboratories represent the calibration in other laboratories. Analysis of the CAP Comprehensive Chemistry Survey indicates that calibration bias in measuring serum creatinine is very common.10 Among 700 participating laboratories, serum creatinine was the analyte most frequently showing a signiﬁcant calibration bias; a detectable calibration bias was present in 97% of laboratories and ranged from 4% to 30%. The average calibration bias for creatinine was 13% higher than the reference method. Overall, calibration biases contributed to 85% of the total error variance between measured serum creatinine and the reference value. Thus, removal of calibration errors would substantially improve the validity of serum creatinine measurements, reducing the total error, measured as the root mean square error, from 15% to 6%.10 Appropriate calibration of the serum creatinine assay would result in urinary creatinine clearance estimates that overestimate GFR because of creatinine secretion. This overestimate can be corrected by explicitly multiplying measured urinary creatinine clearance by a factor that compensates for creatinine secretion. This
CORESH ET AL
approach can be applied uniformly across laboratories and is preferable to attempting to incorporate such a correction into the serum creatinine calibration. In the MDRD validation study, this correction factor was 0.842. Use of urinary creatinine clearance is limited further by errors in collection, which often result in poorer estimates of GFR than estimates based on serum creatinine level and a prediction equation.14 Equations for estimating GFR can be readily implemented into existing laboratory information systems to report estimated GFR values and ﬂag abnormally low values. Such an application of the equation also would allow for explicit correction of the laboratory for the best estimate of its current serum creatinine calibration. Application of equations for estimating GFR also should allow for substantial progress in understanding the epidemiological characteristics of chronic kidney disease in the population. A number of large studies, including the NHANES III, have measured serum creatinine. Appropriate estimation of GFR in these studies will allow for the study of factors associated with low GFRs cross-sectionally and the subsequent risk for kidney and cardiovascular complications prospectively. ACKNOWLEDGMENT The authors thank the White Sands and Cleveland Clinic laboratories, which facilitated the conduct of this study by conducting assays and providing access to their quality control data; investigators and participants of the MDRD study and NHANES III, which provided data for this study; as well as their funding institutes, the National Institute of Diabetes, Digestive and Kidney Disease and the National Center for Health Statistics.
REFERENCES 1. Levey AS: Measurement of renal function in chronic renal disease. Kidney Int 38:167-184, 1990 2. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D: A more accurate method to estimate glomerular ﬁltration rate from serum creatinine: A new prediction equation. Modiﬁcation of Diet in Renal Disease Study Group. Ann Intern Med 130:461-470, 1999 3. Cockcroft DW, Gault MH: Prediction of creatinine clearance from serum creatinine. Nephron 16:31-35, 1976 4. Gault MH, Longerich LL, Harnett JD, Wesolowski C: Predicting glomerular function from adjusted serum creatinine. Nephron 62:249-256, 1992 5. Levey AS, Greene T, Kusek JW, Beck GJ, and MDRD Study Group. A simpliﬁed equation to predict glomerular ﬁltration rate from serum creatinine. J Am Soc Nephrol 11:155A, 2000 (abstr A0828)
6. Perrone RD, Madias NE, Levey AS: Serum creatinine as an index of renal function: New insights into old concepts. Clin Chem 38:1933-1953, 1992 7. Coresh J, Toto RD, Kirk KA, Whelton PK, Massry S, Jones C, Agodoa L, Van Lente F: Creatinine clearance as a measure of GFR in screenees for the African-American Study of Kidney Disease and Hypertension pilot study. Am J Kidney Dis 32:32-42, 1998 8. Lemann J, Bidani AK, Bain RP, Lewis EJ, Rohde RD, the Collaborative Study Group of Angiotensin-Converting Enzyme Inhibition in Diabetic Nephropathy: Use of the serum creatinine to estimate glomerular ﬁltration rate in health and early diabetic nephropathy. Am J Kidney Dis 16:236-243, 1990 9. DeSanto NG, Coppola S, Anastasio P, Coscarella GXCG, Bellini L, Santangelo R, Massimo L, Siciliano A: Predicted creatinine clearance to assess glomerular ﬁltration rate in chronic renal disease in humans. Am J Nephrol 11:181-185, 1991 10. Ross JW, Miller WG, Myers GL, Praestgaard J: The
accuracy of laboratory measurements in clinical chemistry: A study of 11 routine chemistry analytes in the College of American Pathologists Chemistry Survey with fresh frozen serum, deﬁnitive methods, and reference methods. Arch Pathol Lab Med 122:587-608, 1998 11. Kasiske BL, Keane WF: Laboratory assessment of renal disease: Clearance, urinalysis, and renal biopsy, in Brenner BM, Rector FC Jr (eds): The Kidney. Philadelphia, PA, Saunders, 1996, pp 1137-1173 12. College of American Pathologists: Chemistry Survey 1996 Sets C3, C4, C5, C6, C7-B. Northﬁeld, IL, College of American Pathologists, 1996 13. DuBois D, Dubois EF: A formula to estimate the approximate surface area if height and weight be known. Arch Intern Med 17:863-871, 1916 14. Walser M: Assessing renal function from creatinine measurements in adults with chronic renal failure. Am J Kidney Dis 32:23-31, 1998