MEASUREMENT UNCERTAINTY
and statistical methods are still in development for application to measurement uncertainty. The delta value relates to individual test results and the probability of false or indeterminate results, and relates findings back to the sensitivity, specificity and prevalence of the disease in the original study. The use of imprecision data, used in the same way as for quantitative tests, can be applied when a categorical result is reported. ISO/TS 20914:2019 acknowledges this approach and suggests some terminology such as ‘probably negative/positive’ to incorporate this understanding. It is worth noting the same ISO guideline recognises that the use of such an ‘uncertainty zone’ only has value around the cut-off and doesn’t add value the further from the cut-off the result is found.
Qualitative criteria where no numeric data are used involves evaluating the presence or absence of a specific feature or characteristic through visual observations or qualitative indicators.
from numerical results and examples are provided for how to approach it. This is to be applied when categorical results are expressed when a measured result is compared to a pre-determined decision threshold. As such, the measurement procedure is consistent with the definition of quantity as used throughout most guidance documents. Qualitative criteria where no numeric data are used involves evaluating the presence or absence of a specific feature or characteristic through visual observations or qualitative indicators. This type of analysis focuses on categorising or assessing the test item without assigning numerical values.
Challenges with uncertainty in categorical results
The second point to note is the way the result is reported. Measurement uncertainty in numerical (quantitative) results defines the dispersion of results around the result measured, with a given level of confidence. A categorical result cannot do that. There are no reported values and limited statistical models to account for uncertainty. There are, however, some including Bayes’ theorem, the Normal distribution method, and the information theoretic approach.1 Uncertainty in this context relates to uncertainty in the interpretation of the result. It conveys the level of confidence in the result. It can also provide guidance for opportunities for quality improvement. To express MU in qualitative results we use terms such as conditional probability, likelihood ratios, and entropy; themselves
24
taking advantage of metrics we have readily available including clinical sensitivity, specificity and prevalence. Bayes’ theorem provides the positive
predictive value (PPV). It is the probability of a disease in the presence of a positive test result. The probability of the absence of disease with a negative result is the negative predictive value (NPV). Bayes’ theorem considers assay sensitivity, specificity and prevalence. The Normal distribution method uses quantitative measurements to categorise patients as positive or negative. Both methods use the likelihood ratio (LR), calculated by dividing the true-positive rate (TPR) by the false-negative rate (FNR) for a positive result, and by dividing the true- negative rate (TNR) by the false-positive rate (FPR) for a negative result. An LR of 1 provides no additional information to the pretest odds of disease. Less than 1 means there is a decreased probability of disease, while greater than 1 increases the probability. Common terms such as weak support, moderate support, moderately strong support, strong support, very strong support, and extremely strong support can be used to describe LRs. A high LR (>10) indicates a ‘very strong support’. Finally, the somewhat more abstract information theory approach discusses entropy as a measure of uncertainty in binary outcomes.
Grey zone
Concepts including the grey zone and delta zone are discussed in some documents,2
but computational
Qualitative guidelines The Evaluation of Qualitative Test Performance (CLSI EP12-A) provides mathematical models for determining clinical sensitivity, clinical specificity, PPV, NPV and other probabilities. All such metrics can then be applied to MU quantification. They are all calculated from a 2 × 2 contingency table and use Bayesian probability as their basis. By using a method-precision experiment for qualitative tests near the pre- determined cut-off, the C5-C95 interval can be calculated with 95% confidence to assesses the consistency of strong positives and negatives.
Other disciplines as guides to handle qualitative results In the chemical field, discussion of measurement uncertainty in qualitative results is extensive. Organisations like Eurachem have played a significant role in producing authoritative guidance on measurement uncertainty evaluation, method validation, and external quality assessment. These guidelines have helped establish best practices in analytical measurement and promote harmonisation of practices in the industry.
The Eurachem/CITAC guide on the assessment of performance and uncertainty in qualitative chemical analysis provides valuable guidance on how to assess qualitative results and evaluate the associated uncertainties.3 It emphasises the importance of considering both quantitative and qualitative criteria in the performance of qualitative tests. Calculation methods for measurement uncertainty are discussed in the document. It mentions the use of Bayesian probability, specifically in the context of determining clinical sensitivity and specificity. Bayesian probability is
AUGUST 2024
WWW.PATHOLOGYINPRACTICE.COM
Ajay Kumar Chaurasiya CC BY-SA 4.0 Wikimedia Commons
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
