Design: Interobserver variability study on whole-slide images
Population: 58 slides from 20 appendiceal goblet cell adenocarcinomas; 7 expert pathologists
Primary outcome: Agreement on WHO 5th edition three-tier grade
Effect: Full agreement in 4/20 cases; Fleiss kappa 0.29 (95% CI 0.14 to 0.44)

The WHO 5th edition introduced a three-tier grade for appendiceal goblet cell adenocarcinoma based on the proportion of low-grade and high-grade growth patterns. A Histopathology study, published in June 2026, tested how reproducible that is.

Seven gastrointestinal pathologists with an interest in the appendix independently graded whole-slide images of 58 slides from 20 tumours. All seven agreed on the grade for only four cases (20%). Overall agreement was fair (Fleiss kappa 0.29, 95% CI 0.14 to 0.44), and pairwise agreement ranged from none to substantial. Among the individual features, agreement was best for extracellular mucin and tumour sheets (kappa about 0.4) and worst for mild architectural disarray and necrosis (kappa 0.05 and 0.07).

Grade drives decisions on right hemicolectomy and systemic treatment, so a grade that changes with the reader is a problem. The authors suggest a two-tier system may reproduce better. Until then, reports can reduce ambiguity by stating the estimated proportion of high-grade pattern, not just the grade.

Report the estimated percentage of high-grade growth pattern alongside the WHO grade for goblet cell adenocarcinoma.
Name the specific high-grade features seen, such as tumour sheets, since these were read most consistently.
Treat necrosis and mild architectural disarray with caution as grading criteria; agreement on them was poor.
Consider intradepartmental or expert review for borderline grade 2 cases that will change surgical management.

Why it matters

A grade that experts cannot reproduce may be sending patients to different operations depending on who reads the slide.

The statistics, in plain English

A kappa of 0.29 means agreement only modestly better than chance. A pairwise median of 0.13 means a typical pair of experts agreed little more than chance would predict.