MEASUREMENT UNCERTAINTY


Bias


Bias (in Standard Deviations from 0)


No bias 95%


2.5% 2.5%


Less than 95%


More than 2.5%


0


0.5 1 2 3


% normal population falling out of the reference range


5


10 20 50 80


Fig 2. The impact of bias on reference ranges derived in the laboratory. Image adapted from Coskun A.3


assume all significant bias is removed. This is variably possible in medical laboratories. There may be facility to apply a correction factor to final patient results if a bias is identified, and unable to be corrected for. This requires that a carefully considered correction is applied and it doesn’t result in negation of regulatory approval of methods or inappropriate alteration of metrologically traceable material including calibrator assigned values.


Emerging topics regarding bias and measurement uncertainty


n Lot-to-lot variability In the laboratory, lot-to-lot variation only checks an incoming lot against the current lot. This runs a risk of introducing, or not detecting, long-term drifts. If bias is deemed acceptably small and correction is not required, the bias that is infrequently present due to lot changes, calibration and short-term systematic shifts is detected within uRW


as part of


long-term imprecision. This was a major part of the discussion held in the 5th CELME meeting in Prague (October, 2023) where practical applications of APS models were discussed.


n Bias between analysers and method uncertainty


When using a common mean and SD across multiple analysers in a laboratory or laboratory network the apparent bias between analyser-specific means is managed according to ISO/TS 20914:2019.


mean target values into a single estimate of uRW


Incorporating small differences in IQC will artificially inflate uRW


estimates


if not handled properly. To be able to provide a single combined uncertainty for the analytical platform, that can


reasonably be applied to a patient sample run on any one of the analysers, we can calculate an overall combined uncertainty. We use the pooled standard deviation method discussed in the previous article, incorporating the differences in means across analysers.


This is a multi-step process (Fig 3) starting with: n The mean, and SD of the IQC for each QC level on each analyser (step 1).


n The SD from step 1 squared gives the variance of the same material (step 2).


n Separately calculate the variance of the means across the analysers (step 3).


n Using the pooled SD method combine the individual components of step 2 (step 4).


n Combine the step 3 result with the step 4 result by the root sum of squares method (step 5). n Combine step 5 with ucal combined uncertainty.


for overall


n Total error versus MU: the battle continues


From the summary example shown in Figure 3, it can be seen that, as you would expect, the combined uncertainty across all analysers is larger than any single uncertainty. This is due to the inclusion of all long-term imprecision, but importantly for the topic here it also includes the small biases across analyser means. This approach isn’t, I don’t think, widely


There are a number of reasons why the TE and MU approaches appear to be in opposition. Some of those are long-lasting and come from the battle between a top-down approach and a bottom-up approach. Total error is seen as a top-down approach, all error (random and systematic) being contained within a single metric. MU was historically seen


Table 1. The percent bias present in the assay and its impact on the proportion of the normal population relative to the reference intervals derived assuming bias > 0.


applied, but it is a very common question asked of our uncertainty and how it is calculated. The fact that the uncertainty of the group is larger than any individual method raises concerns with some people about overestimating the uncertainty and making our assays look worse than they are. However, ultimately, by comparing the MU to the performance specification we have, as discussed in a previous article, if this result achieves specifications, we are satisfied it is clinically appropriate. If the maximum MU is exceeded, the usual troubleshooting and root cause is required to identify which of the methods is not performing within specification. In this scenario, it may be due to it being far away from the group (Analyser B in Figure 3 for example is furthest from the other two) or it may be due to excessive random variability.


Measurement uncertainty as a metrological concept subscribes to the same metrological concepts as those outlined in the GUM that assume all significant bias is removed. This is variably possible in medical laboratories


WWW.PATHOLOGYINPRACTICE.COM JUNE 2024 17


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