Development and temporal validation of a pathology-based model incorporating the lymph node ratio to predict recurrence in stage III colorectal cancer.
- 1Department of General Practice, Affiliated Hospital of Xuzhou Medical University, Xuzhou, China.
Abstract
Postoperative recurrence remains a major determinant of long-term outcomes in stage III colorectal cancer (CRC). The extent to which the lymph node ratio (LNR), when combined with routinely reported pathologic features, improves individualized recurrence-risk estimation remains uncertain. We aimed to develop and temporally validate a pathology-based recurrence model incorporating LNR and routinely available clinicopathologic features in stage III CRC. We retrospectively analyzed 400 patients with pathologic stage III CRC who underwent curative-intent resection at a single institution. Patients treated from January 2020 to August 2023 formed the training cohort (n=300), and those treated from September 2023 to March 2024 formed a temporally separated validation cohort (n=100). Recurrence-free survival (RFS) was the primary endpoint. Missing covariate data were handled using 30 multiple imputations. Least absolute shrinkage and selection operator (LASSO)-Cox selection was repeated in each imputed training dataset, and variables selected in at least 24 of 30 datasets were entered into pooled multivariable Cox models. LNR was modeled continuously in the primary analysis. Model performance was assessed using Harrell's C-index, time-dependent receiver operating characteristic curves, calibration, decision curve analysis, risk stratification, and incremental comparisons beyond tumor (T)/node (N) stage. Eight variables met the prespecified stability-selection threshold, and a parsimonious five-variable model retained tumor size, perineural invasion, continuous LNR, tumor budding, and tumor deposits. Each 0.1 increase in LNR was associated with a higher recurrence risk [hazard ratio (HR) 1.20, 95% confidence interval (CI): 1.09-1.32]. The model achieved C-indices of 0.750 (95% CI: 0.705-0.793) in training and 0.746 (95% CI: 0.637-0.848) in temporal validation. Time-dependent areas under the curve (AUCs) were 0.789, 0.778, and 0.808 at 1, 2, and 3 years in training and 0.818 and 0.776 at 1 and 2 years in validation. Calibration was acceptable, although estimates in the validation cohort were imprecise. Adding LNR to a model containing T/N stage and other pathologic factors resulted in only modest increases in discrimination in both cohorts. A pathology-based prognostic model incorporating continuous LNR with routinely available pathologic features showed moderate discrimination for postoperative recurrence in stage III CRC. The model may complement TNM staging for risk stratification, but the incremental discriminative contribution of LNR beyond T/N stage and other pathologic factors was modest. Independent multicenter validation with longer follow-up is required before clinical implementation.
Related
- Incorporating natural language processing-derived features from pathology reports into machine learning models to predict differentiated thyroid cancer recurrence
- Development and Internal Validation of a Prediction Model Incorporating Respiratory Muscle Function for CAT Score Deterioration in Stable COPD
- Development and internal validation of a nomogram incorporating C-reactive protein-to-albumin ratio for predicting esophageal stricture after endoscopic submucosal dissection in early esophageal neoplasia.
- A clinicopathological and hematological biomarker-based nomogram for overall survival in early-onset colorectal cancer: development, external validation, and translational implications
- Development and validation of a nomogram incorporating vascular morphological parameters for prediction of peripherally inserted central catheter-related thromboembolism in cancer patients.
BibTeX
@article{gu2026development,
title = {{Development and temporal validation of a pathology-based model incorporating the lymph node ratio to predict recurrence in stage III colorectal cancer.}},
author = {Jiahui Gu and Qiuyu Chen and Jiaying Xiong and Ruiling Yang},
journal = {Journal of gastrointestinal oncology},
year = {2026},
month = sep,
doi = {10.21037/jgo-2026-0718},
url = {https://europepmc.org/article/MED/42996590},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Development and temporal validation of a pathology-based model incorporating the lymph node ratio to predict recurrence in stage III colorectal cancer.
AU - Jiahui Gu
AU - Qiuyu Chen
AU - Jiaying Xiong
AU - Ruiling Yang
PY - 2026
DA - 2026/09/28
JO - Journal of gastrointestinal oncology
DO - 10.21037/jgo-2026-0718
UR - https://europepmc.org/article/MED/42996590
AB - Postoperative recurrence remains a major determinant of long-term outcomes in stage III colorectal cancer (CRC). The extent to which the lymph node ratio (LNR), when combined with routinely reported pathologic features, improves individualized recurrence-risk estimation remains uncertain. We aimed to develop and temporally validate a pathology-based recurrence model incorporating LNR and routinely available clinicopathologic features in stage III CRC. We retrospectively analyzed 400 patients with pathologic stage III CRC who underwent curative-intent resection at a single institution. Patients treated from January 2020 to August 2023 formed the training cohort (n=300), and those treated from September 2023 to March 2024 formed a temporally separated validation cohort (n=100). Recurrence-free survival (RFS) was the primary endpoint. Missing covariate data were handled using 30 multiple imputations. Least absolute shrinkage and selection operator (LASSO)-Cox selection was repeated in each imputed training dataset, and variables selected in at least 24 of 30 datasets were entered into pooled multivariable Cox models. LNR was modeled continuously in the primary analysis. Model performance was assessed using Harrell's C-index, time-dependent receiver operating characteristic curves, calibration, decision curve analysis, risk stratification, and incremental comparisons beyond tumor (T)/node (N) stage. Eight variables met the prespecified stability-selection threshold, and a parsimonious five-variable model retained tumor size, perineural invasion, continuous LNR, tumor budding, and tumor deposits. Each 0.1 increase in LNR was associated with a higher recurrence risk [hazard ratio (HR) 1.20, 95% confidence interval (CI): 1.09-1.32]. The model achieved C-indices of 0.750 (95% CI: 0.705-0.793) in training and 0.746 (95% CI: 0.637-0.848) in temporal validation. Time-dependent areas under the curve (AUCs) were 0.789, 0.778, and 0.808 at 1, 2, and 3 years in training and 0.818 and 0.776 at 1 and 2 years in validation. Calibration was acceptable, although estimates in the validation cohort were imprecise. Adding LNR to a model containing T/N stage and other pathologic factors resulted in only modest increases in discrimination in both cohorts. A pathology-based prognostic model incorporating continuous LNR with routinely available pathologic features showed moderate discrimination for postoperative recurrence in stage III CRC. The model may complement TNM staging for risk stratification, but the incremental discriminative contribution of LNR beyond T/N stage and other pathologic factors was modest. Independent multicenter validation with longer follow-up is required before clinical implementation.
SN - 2078-6891
SN - 2219-679X
ER -