RESULTS AND INTERPRETATION
Numerical uncertainty and categorical agreement
Do not turn an agreement coefficient into a standard uncertainty.
Identify the output first
A numerical concentration, an ordinal grade and a positive/negative classification are different outputs. Some categorical results are derived from a quantitative signal; others arise from visual interpretation or a more complex decision process. Start by describing that process.
| Output | Useful questions |
|---|---|
| Numerical result | What quantity is estimated, and how uncertain is it? |
| Binary classification | How do classifications perform relative to an appropriate reference? |
| Ordinal grade | How are disagreements distributed across ordered categories? |
| Observer assessment | How repeatable and reproducible are the ratings under the study conditions? |
For a numerical step
Evaluate analytical uncertainty at relevant signal or concentration levels, especially where the numerical step contributes to a classification. Do not assume that a numeric code assigned to a category makes the category a continuous measurand.
For categorical performance
Show the underlying cross-tabulation and denominators. Distinguish agreement from diagnostic performance. Sensitivity and specificity require an appropriate reference classification; positive and negative agreement should not be relabelled as diagnostic accuracy when that reference is absent.
A common mistake to remove
Kappa is an agreement statistic, not a quantity in the units of a patient result. It cannot be inserted directly as a standard uncertainty in a numerical MU budget. Its interpretation is also affected by the distribution of categories and marginal ratings.
Study the disagreements
Report category-specific findings, relevant uncertainty intervals and the study conditions. For ordinal ratings, explain any weighting of disagreements. For continuous measurements, use methods suited to continuous data rather than converting results into arbitrary categories.
ISO/TS 20914 explicitly includes qualitative results that contain a quantitative measurement step. That scope does not imply that every categorical decision has a single numerical expanded uncertainty.
Sources & further reading
Educational guidance, not a substitute for a locally approved procedure. Worked examples are synthetic. Reviewed against the sources indicated; licensed standards must be consulted for their full requirements.