Reproducible and Interpretable H&E-based Histopathological Subtypes of Non-metastatic Lymph Nodes Refine Risk Stratification in Colorectal Cancer
In brief
Non-metastatic node patterns stratify colorectal cancer survival, with validation scores near 0.75
In 754 patients across three cohorts, researchers classified non-metastatic lymph nodes into three immune-related patterns, with strong agreement between pathologists. A model combining these patterns separated patients by overall survival in temporal and geographic validation cohorts, with performance scores near 0.75, and added information beyond standard staging; whether it improves treatment decisions remains unknown.
- Journal
- Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc (Q1)
- Published
- 9 October 2026
- Study design
- Cohort / observational study
- Evidence level
- Level 3, Low (CEBM 3b)
- Authors
- Zhen Zhang, Yueyang Zhang, Xu Liu, Yafang Li, Wei Cui, Hong Su, et al.
- PMID
- 42855076
- DOI
- 10.1016/j.modpat.2026.101096
Why clinicians should know about it
- Picked for Pathology and Forensic Medicine (top studies of the week, 11 October 2026): Large multicentre cohort, reproducible LN histopathology subtypes
Abstract
Lymph node status is fundamental for staging, prognostic assessment, and treatment decision-making in colorectal cancer (CRC); however, current pathological evaluation largely focuses on metastatic involvement and overlooks the biological heterogeneity of non-metastatic lymph nodes (LNs). Here, we developed a reproducible histopathological classification system for non-metastatic LNs and established an interpretable machine-learning model for patient-level risk stratification in CRC. In this multicentre cohort study, we analyzed H&E-stained whole-slide images of non-metastatic LNs from patients with CRC undergoing curative-intent surgery across three independent cohorts. Non-metastatic LNs were classified into stimulated, unstimulated, and depleted subtypes based on semiquantitative assessment of proliferative and depletion-related morphological features. The biological characteristics of these subtypes were investigated using immunohistochemistry, transcriptomic profiling, and clinicopathological correlation analyses. Patient-level distributions of lymph node histopathological subtypes were integrated into multiple machine-learning algorithms, with model performance evaluated in internal and external validation cohorts. A total of 754 patients and 11,678 LN whole-slide images were included, comprising 10,410 non-metastatic and 1,268 metastatic LNs. Histopathological subtype assignment showed high interobserver reproducibility (Fleiss' κ=0.879, 95% CI 0.810-0.938). Stimulated LNs demonstrated enhanced immune activation features, whereas depleted LNs exhibited stromal remodeling, innate inflammatory activation, and impaired adaptive immune signaling. These histopathological subtypes were strongly associated with previously defined molecular LN subtypes. Among 12 machine-learning approaches, the random survival forest model achieved the best prognostic performance, with C-indices of 0.877 in the training cohort, 0.777 in cross-validation, 0.744 in temporal validation, and 0.748 in geographic validation. The resulting risk stratification system consistently discriminated overall survival across independent cohorts and clinically relevant subgroups. We established a reproducible H&E-based histopathological classification of non-metastatic LNs in CRC and developed an interpretable machine-learning model that captures lymph node immune-stromal heterogeneity and provides additional prognostic information beyond conventional staging parameters.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.