Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1992
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T06:55:29.764915Z
status: published
lens: g_space
sector: health
headline: Development and validation of an explainable model for predicting lymph node metastasis in early gastric cancer: a multicenter retrospective study with external and prospective validation
dek: Accurate preoperative prediction of lymph node metastasis (LNM) is essential for treatment selection in early gastric cancer (EGC), particularly when considering endoscopic submucosal dissection. Existing approaches show limited sensitivity for metastatic disease and frequently lack external or prospective validation. We aimed to develop and validate an interpretable machine-learning model for individualized preoperative prediction of LNM in EGC. In this multicenter study, four cohorts from three tertiary center…
gain_title: An interpretable XGBoost model estimated individual risk of lymph node metastasis in early gastric cancer and held up in external and prospective validation, which may help choose between endoscopic and surgical treatment.
problem_title: (none)
trace_subject: (none)
gain_reading: An interpretable XGBoost model estimated individual risk of lymph node metastasis in early gastric cancer and held up in external and prospective validation, which may help choose between endoscopic and surgical treatment.
gain_evidence: maintained robust performance in the external (AUC = 0.8583; PR-AUC = 0.6079) and prospective validation cohorts | may support personalized decision-making between endoscopic and surgical management in EGC
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built a machine-learning model to estimate before surgery whether early gastric cancer has spread to lymph nodes. They trained it on 3412 patients from tertiary centers, of whom 467 had lymph node metastasis, and compared six algorithms. XGBoost performed best.

The result matters because treatment selection, particularly whether endoscopic submucosal dissection is appropriate, depends on accurate nodal risk estimates, and existing approaches were described as lacking sensitivity and validation. The model's discrimination fell from training to validation, and the source reports performance metrics rather than patient outcomes from using the tool.
limitation: PR-AUC values in validation cohorts (about 0.51 to 0.61) were considerably lower than the training AUC suggests, indicating room for improvement in identifying metastatic cases.
tag: Evidence-backed gain
key_points: Four cohorts from three tertiary centers were used, including a prospective cohort enrolled from September 2023 to December 2024. | XGBoost reached a training AUC of 0.9289, an external AUC of 0.8583, and a prospective AUC of 0.8766. | A preoperative pathology validation cohort from January 2025 to January 2026 gave an AUC of 0.8653 and a PR-AUC of 0.6050. | Key predictors were lymphovascular/neural invasion, invasion depth, tumor location, differentiation, and tumor size. | SHAP was used for interpretability and a web-based calculator was developed.
rundown: Patients with pathologically confirmed early gastric cancer who underwent gastrectomy with lymph node dissection were enrolled retrospectively for training and external validation. Variables were selected with LASSO regression, and SMOTE addressed class imbalance, since only 13.7% of training patients had metastasis.

Performance was assessed using AUC and PR-AUC, with SHapley Additive exPlanations used to explain predictions. The prospective cohort had an AUC of 0.8766 and PR-AUC of 0.5137, lower than the training AUC.

The authors released a web-based calculator intended to give transparent individualized risk estimates to inform the choice between endoscopic and surgical management.
sources:
- peer_reviewed | European Journal of Surgical Oncology | https://doi.org/10.1016/j.ejso.2026.112310 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

d646c1a8d75e8616b34b6d8ee3c0828b63711ceb9810fb0a05900b035cde6c04

previous

0000000000000000000000000000000000000000000000000000000000000000