MEASUREMENT UNCERTAINTY
Bias and measurement uncertainty: a duo rarely handled correctly
In this further article in his series on measurement uncertainty, Stephen MacDonald moves on to look at bias, and considers its nature, detection, clinical and statistical significance, and correction.
Bias (or its absence) and trueness are often considered the same thing and described as systematic error. At its simplest, bias is the difference between a value that is observed and one that is expected. If that expected value is a ‘known’ value, it is also a representation of its trueness.
Bias is either constant across the
measurement range or is influenced by the measurand value, so called proportional bias. Short- and long- term biases can be introduced through common laboratory activities including lot changes of reagents and calibrators. A solution may be to recalibrate the assay. Over calibration may itself introduce bias, and impact metrological traceability.
Clinically significant bias should be eradicated by manufacturers and suppliers of methods. This is achieved by evidencing metrological traceability of the method (ISO/TS 20914:2019, 6.6). Ideally, calibration steps are documented and provided to laboratories through the instructions for use, although this is rarely realised. Despite removal of all clinically significant bias, the uncertainty of that correction (ubias) remains and is a source to be considered in the ISO/TS 20914:2019 framework. Absence of ubias data may have significant implications. Local ongoing quantification of bias is routinely
achieved using certified reference material studies, external quality assessment (EQA) and internal quality control (IQC) peer review.
The nature of bias and its detection
The method of bias detection and correction determines the impact
Assigned value
Uncertainty of assigned value
of uncertainty in patient results. Proportional bias, when comparing results, has the appearance of increasing the slope of the regression line so that as the measurand concentration increases so does the apparent bias. It is often characterised as a percentage of the measurand value. For example, a measurand may have a proportional bias of 5%. If the measurand value (the true value) is 100 units, the proportional bias will add 5 units to that and give a value of 105 units. If the measurand value is increased to 200 the difference between expected and realised value becomes 10 units (200 + 5% = 210 units). Constant bias is different in that the absolute value of the bias is consistent, irrespective of the concentration of the measurand. It is represented by an absolute number,
Achieved value Standard error
Absolute (unmeasurable bias) Potential bias Potential bias
Overlap of distributions
Fig 1. Different potential values for bias based on uncertainty contained within reference values and values derived in the laboratory being used for comparison.
WWW.PATHOLOGYINPRACTICE.COM JUNE 2024 15
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56
