HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Extract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Thienpont, L. M. Search for Related Content PubMed PubMed Citation Articles by Thienpont, L. M.

Calculation of Measurement Uncertainty—Why Bias Should Be Treated Separately

Laboratory for Analytical Chemistry, Faculty of Pharmaceutical Sciences, Gent University, Gent, Belgium

Address correspondence to the author at: Laboratory for Analytical Chemistry, Faculty of Pharmaceutical Sciences, Gent University, Harelbekestraat 72, B-9000 Gent, Belgium, Fax 32-9-264.81.98, E-mail linda.thienpont{at}ugent.be

To the Editor:

The concept of measurement uncertainty (u), as laid down in the Guide to the Expression of Uncertainty in Measurement (GUM) (1), is gaining more and more importance in the field of laboratory medicine. This concept does not include bias in the calculation of u, but instead requires it to be reported separately. Many are dissatisfied with that approach and try to "fix" GUM by treating bias as a variance and propagating it as such(2). I contend that the calculation of u should adhere to the original GUM document. Assume that an analytical procedure with a CV_A of 10% and a bias of +10% is used for the measurement of a serum component, x, in a group of healthy individuals with a CV_G of 20% (CV_G, group biological variation, consisting of within- and between-individual biological variation). These values yield a distribution with a total CV (CV_T) of 22.4% [CV_T = (CV_G² + CV_A²)]. The test results are used to detect "hypo-x" and "hyper-x" values at particular clinical-decision limits. For simplification, let us assume that these limits are set at ±3 CV_G, which is equal to the mean of x ± 60%. Fig. 1 shows the distributions of component x for the unbiased (solid line) and biased situations. The biased situation is presented as actually observed values (dashed line) and those predicted by the treatment of bias as variance [u = (CV_A² + bias²) = 14.1%] (dotted line). When bias is treated as a variance, CV_A increases from 10% to 14.1%, and the distribution of values for component x becomes slightly broader [CV_T = (CV_G² + CV_A² + bias²) = 24.5%]. This choice slightly increases the number of hypo-x and hyper-x cases. In reality, however, the population is moved by the bias to the hyper-x side. Consequently, the number of hypo-x and hyper-x cases significantly decreases and increases, respectively. This example shows that the treatment of bias as variance leads to an erroneous prediction of the influence of test performance in clinical practice. Therefore, bias should not be included as a variance component in calculations of measurement uncertainty but should be reported separately.

View larger version (19K): [in this window] [in a new window]   Figure 1. Distribution of values of a component x in a group of healthy individuals with a CVG of 20%, as measured with a CVA of 10% for an unbiased situation (solid line) and a biased situation (hypo-x and hyper-x are indicated at ±3 CVG, equivalent to ±60%). The biased situation (bias = +10%) is presented as actually observed values (dashed line) and as values predicted by treating bias as a variance [u = (CVA2 + bias2) = 14.1%] (dotted line). Note that the lower panel is an enlargement of the upper panel.

Acknowledgments

Grant/Funding Support: None declared.

Financial Disclosures: None declared.

References

1. . International Organization for Standardization (ISO). Guide to the expression of uncertainty in measurement 2nd ed. 1995 International Organization for Standardization (ISO) Geneva. .
2. Haesselbarth W. Accounting for bias in measurement uncertainty estimation. Accred Qual Assur 2004;9:509-14.

This Article Extract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Thienpont, L. M. Search for Related Content PubMed PubMed Citation Articles by Thienpont, L. M.