Today's lung carcinoid paper reports inter-observer agreement for the WHO 2021 classification applied to biopsies at κ 0.276. Three specialist pathologists, revising the same slides, agreed only slightly more than chance would predict.
That number deserves to sit uncomfortably, because the classification is not thereby wrong. It is a statement about what morphology can carry on a small biopsy, where mitotic counting and necrosis assessment — the two criteria separating typical from atypical carcinoid — are precisely the features a biopsy samples worst.
The practical discipline is to say so. A report that gives a confident category on a biopsy where the criteria cannot be reliably applied transfers a false certainty to the person acting on it. Naming the limitation, or deferring the grade to resection, is not hedging; it is the accurate report. Where a reproducible adjunct exists — and today's paper offers one at κ 0.673 — it earns its cost precisely because it removes the observer from the answer.
- Distinguish 'the features are absent' from 'the sample cannot show the features'
- On a small biopsy, defer a grade that depends on mitotic count or necrosis rather than estimating it
- Where an adjunct has published agreement figures, they are as relevant as its sensitivity
- Audit your own agreement on a category you report often; kappa is measurable within a department
- A classification with poor reproducibility is still useful — it just cannot bear a decision on its own
Why it matters
Reproducibility is usually treated as an academic property of a classification rather than a fact about the report you signed this morning.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for pathology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free