An official website of the United States government
Publication Information
Affiliation
1. Division of Nephrology, Department of Medicine, University of Mississippi Medical Center,
Jackson, Mississippi
2. Department of Physiology, University of Mississippi Medical Center, Jackson, Mississippi
3. Department of Population Health, Bower School of Population Health, University of
Mississippi Medical Center, Jackson, Mississippi
4. Department of Data Science, Bower School of Population Health, University of Mississippi
Medical Center, Jackson, Mississippi
5. Department of Epidemiology and Welch Center for Prevention, Epidemiology, and Clinical
Research, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD,
United States
6. Division of Nephrology, Department of Medicine, Mayo Clinic, Rochester, Minnesota
7. The Mind Center, University of Mississippi Medical Center, Jackson, Mississippi
8. Division of Cardiology, Department of Medicine, University of Mississippi Medical Center,
Jackson, Mississippi
9. Department of Surgery, University of Mississippi Medical Center, Jackson, Mississippi
Authors
Butler J, Butler KR, Dossabhoy NR, Guallar E, Hall ME, Lirette ST, Rule AD, Shafi T, Thomas Mosley T, Tio MC, Vaitla P, Wynn JJ, Zhu X
Abstract
Background
Although the population-level differences between estimated and measured glomerular filtration
rate (eGFR and mGFR) are well-recognized, the magnitude and potential clinical implications of
individual-level differences are unknown.
Methods
Using data from four community-based cohorts with mGFR (N=3,223), we calculated eGFR
from serum creatinine alone (eGFRCR) and with cystatin C (eGFRCR-CYS). Using quantile
regression, we assessed eGFR's individual-level inaccuracy by calculating a 95% prediction
interval (PI), defined as the distribution of 95% of the mGFRs at a given eGFR. We also
assessed eGFR's reliability using standard metrics, including bias (median mGFR–eGFR). All
GFR results are presented as ml/min/1?73m2
.
Findings
The participants' mean age was 59 years, 32% were Black individuals, 55% were women, and
the mean mGFR was 68. The bias for eGFRCR was small (–0.6); however, the individual-level
inaccuracy was large. The 95% PI for eGFRCR 90 was 56 to 120; for eGFRCR, 60 was 36 to 87;
for eGFRCR, 45 was 27 to 70, and; for eGFRCR 30, it was 17 to 54. eGFRCR bias among Black
(3.7) and White (–2.9) individuals was substantially smaller than the width of the PI (97?5th–
2?5th percentiles) in Black (50) and White (57) individuals. Substantial misclassification was
noted; 37% of individuals with eGFRCR 45 to 59 and 29% with eGFRCR <30 had mGFR above
those thresholds. eGFRCR-CYS had similar individual-level inaccuracy; PIs were wide, and among
those with eGFRCR-CYS 45 to 59, the agreement with mGFR was only 49%.