Grading phyllodes tumours rests on features experts agree on only moderately

Design

Interobserver agreement study with pre- and post-training assessment

Population

19 subspecialist breast pathologists; 54 images for cellularity and atypia, 23 for border

Primary outcome

Agreement (Krippendorff's α, Fleiss's κ) on stromal cellularity, atypia and border

Effect

Post-training two-tier κ: cellularity 0.675, atypia 0.427, border 0.615; training made most readers more conservative

The distinction between fibroadenoma and benign, borderline and malignant phyllodes tumour depends on stromal cellularity, stromal atypia and the tumour border. This study measured how consistently 19 subspecialist academic breast pathologists assessed those features on standardised images, before and after structured virtual training and a washout of at least three weeks.

Before training, agreement on cellularity was moderate on a three-tier (α 0.727) and two-tier scale (κ 0.675); on atypia it was poor on three tiers (α 0.570) and moderate on two (κ 0.446); on border it was moderate (κ 0.532). After training, cellularity and atypia barely moved (κ 0.675 and 0.427); border agreement improved to κ 0.615 but remained moderate. What did change was calibration — most pathologists became more conservative, assigning lower cellularity (15 of 19) and atypia (13 of 19) to the same images.

If experts agree only moderately, a single reader in a general laboratory should not treat a borderline call as secure. For atypia, agreement was rated moderate on a two-tier call but poor on three tiers, and a second opinion on borderline and malignant calls is the proportionate response, because the grade drives margin width and the decision to re-excise.

• Grade stromal cellularity and atypia using the WHO descriptors (mild, moderate, marked); for atypia, agreement was rated moderate on the marked-versus-not-marked call but poor on three tiers.
• Seek a second opinion before signing out a borderline or malignant phyllodes tumour.
• Describe the border explicitly as infiltrative or well circumscribed and state if it is uncertain.
• Sample the tumour-stroma interface generously; the border call is the least reliable feature.
• Tell surgeons when a grade is borderline so re-excision decisions weigh that uncertainty.

Why it matters

The grade decides margin width and re-excision, yet the features behind it are not reliably reproducible even among specialists.

Don't overread it

Agreement was measured on selected static images, not whole slides, and may differ in routine practice.

The statistics, in plain English

Kappa and alpha measure agreement beyond chance: 0 is chance, 1 is perfect, and values around 0.4–0.7 are usually called moderate. A κ of 0.43 for atypia means pathologists agreed only somewhat better than chance would predict, even after training.

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.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.