- Design
- Multicentre cohort study with machine-learning prognostic model
- Population
- 754 patients with colorectal cancer; 11,678 lymph node whole-slide images
- Primary outcome
- Overall survival risk stratification by lymph node histopathological subtype
- Effect
- Interobserver kappa 0.879; random survival forest C-index 0.744 temporal and 0.748 geographic validation
A multicentre cohort study examined H&E whole-slide images of 11,678 lymph nodes (10,410 non-metastatic and 1,268 metastatic) from 754 patients with colorectal cancer after curative-intent surgery.
Non-metastatic nodes were classified as stimulated, unstimulated or depleted on semiquantitative features of proliferation and depletion. Agreement between observers was high (Fleiss' kappa 0.879, 95% CI 0.810 to 0.938). Stimulated nodes showed immune activation; depleted nodes showed stromal remodelling and impaired adaptive immune signalling. Among 12 machine-learning models, a random survival forest performed best, with C-indices of 0.877 in training, 0.744 in temporal validation and 0.748 in geographic validation.
This is a research classification, not a reporting standard. The training performance overstates what validation showed, and prospective confirmation is needed before it changes how nodes are signed out.
- Nodal morphology beyond metastasis may carry prognostic information.
- Agreement was high, which suggests the subtypes can be taught and applied.
- External performance (C-index about 0.75) is modest and below the training figure.
- No change to current staging or reporting is supported yet.
Why it matters
It suggests negative nodes are not all equal, a question usually closed once metastasis is excluded.
Don't overread it
A retrospective model with modest external discrimination; it adds to, and does not replace, staging.
The statistics, in plain English
A C-index of 0.5 is chance and 1.0 is perfect discrimination. The fall from 0.877 in training to about 0.75 in outside cohorts is the realistic expectation, because models fit best to the data they learned from.
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