Advanced

Accuracy diagrams : a novel way to illustrate uncertainty of estimated GFR

Björk, Jonas LU ; Grubb, Anders LU ; Sterner, Gunnar LU ; Bäck, Sten Erik LU and Nyman, Ulf LU (2017) In Scandinavian Journal of Clinical and Laboratory Investigation 77(3). p.199-204
Abstract

Most studies that validate GFR equations present accuracy results stratified by measured GFR (mGFR; diagnostic correctness) or by estimated GFR (eGFR; diagnostic predictiveness) only, without a clear distinction in interpretation. The accuracy of a GFR equation is normally reported in percent (e.g. P30), but is often misinterpreted when stratified by eGFR. The aim of the study was to develop new accuracy measures and diagrams that allow straightforward interpretations and illustrations of the uncertainty in eGFR in clinical practice. We applied quantile regression to the distribution of estimation errors for two creatinine-based GFR equations, LM-REV and CKD-EPI, in a clinical cohort (n = 3495) referred for GFR measurement... (More)

Most studies that validate GFR equations present accuracy results stratified by measured GFR (mGFR; diagnostic correctness) or by estimated GFR (eGFR; diagnostic predictiveness) only, without a clear distinction in interpretation. The accuracy of a GFR equation is normally reported in percent (e.g. P30), but is often misinterpreted when stratified by eGFR. The aim of the study was to develop new accuracy measures and diagrams that allow straightforward interpretations and illustrations of the uncertainty in eGFR in clinical practice. We applied quantile regression to the distribution of estimation errors for two creatinine-based GFR equations, LM-REV and CKD-EPI, in a clinical cohort (n = 3495) referred for GFR measurement (plasma clearance of iohexol). Measures of bias and precision and accuracy intervals (AIs) were expressed in mL/min/1.73 m2. Diagrams with AIs were chosen as a novel way to present the error margin in eGFR at a pre-specified certainty level. It was shown that creatinine-based equations are still quite inaccurate in that large estimation errors could not be ruled out with satisfactory certainty. As an example, the 75% AI for the most accurate equation, LM-REV, was approximately ±10 mL/min/1.73 m2 at eGFR = 45 mL/min/1.73 m2, whereas it ranged between −13 and +20 mL/min/1.73 m2 at eGFR = 90 mL/min/1.73 m2. Accuracy intervals presented in diagrams can be used to illustrate the uncertainty of eGFR. Future validation studies should assess the variability in the predictiveness of eGFR across populations and clinical settings using tools and performance measures that are easy to interpret.

(Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
bias (epidemiology), clinical decision-making, predictive value of tests, Renal insufficiency, reproducibility of results, statistical regression
in
Scandinavian Journal of Clinical and Laboratory Investigation
volume
77
issue
3
pages
199 - 204
publisher
Informa Healthcare
external identifiers
  • scopus:85014540352
  • wos:000399482100006
ISSN
0036-5513
DOI
10.1080/00365513.2017.1292362
language
English
LU publication?
yes
id
f3491d88-4478-407c-9e1e-0f418004da72
date added to LUP
2017-03-15 08:19:36
date last changed
2018-01-07 11:55:21
@article{f3491d88-4478-407c-9e1e-0f418004da72,
  abstract     = {<p>Most studies that validate GFR equations present accuracy results stratified by measured GFR (mGFR; diagnostic correctness) or by estimated GFR (eGFR; diagnostic predictiveness) only, without a clear distinction in interpretation. The accuracy of a GFR equation is normally reported in percent (e.g. P<sub>30</sub>), but is often misinterpreted when stratified by eGFR. The aim of the study was to develop new accuracy measures and diagrams that allow straightforward interpretations and illustrations of the uncertainty in eGFR in clinical practice. We applied quantile regression to the distribution of estimation errors for two creatinine-based GFR equations, LM-REV and CKD-EPI, in a clinical cohort (n = 3495) referred for GFR measurement (plasma clearance of iohexol). Measures of bias and precision and accuracy intervals (AIs) were expressed in mL/min/1.73 m<sup>2</sup>. Diagrams with AIs were chosen as a novel way to present the error margin in eGFR at a pre-specified certainty level. It was shown that creatinine-based equations are still quite inaccurate in that large estimation errors could not be ruled out with satisfactory certainty. As an example, the 75% AI for the most accurate equation, LM-REV, was approximately ±10 mL/min/1.73 m<sup>2</sup> at eGFR = 45 mL/min/1.73 m<sup>2</sup>, whereas it ranged between −13 and +20 mL/min/1.73 m<sup>2</sup> at eGFR = 90 mL/min/1.73 m<sup>2</sup>. Accuracy intervals presented in diagrams can be used to illustrate the uncertainty of eGFR. Future validation studies should assess the variability in the predictiveness of eGFR across populations and clinical settings using tools and performance measures that are easy to interpret.</p>},
  author       = {Björk, Jonas and Grubb, Anders and Sterner, Gunnar and Bäck, Sten Erik and Nyman, Ulf},
  issn         = {0036-5513},
  keyword      = {bias (epidemiology),clinical decision-making,predictive value of tests,Renal insufficiency,reproducibility of results,statistical regression},
  language     = {eng},
  month        = {02},
  number       = {3},
  pages        = {199--204},
  publisher    = {Informa Healthcare},
  series       = {Scandinavian Journal of Clinical and Laboratory Investigation},
  title        = {Accuracy diagrams : a novel way to illustrate uncertainty of estimated GFR},
  url          = {http://dx.doi.org/10.1080/00365513.2017.1292362},
  volume       = {77},
  year         = {2017},
}