External validation and comparison of five prediction models for lymph node metastasis after noncurative endoscopic submucosal dissection for early gastric cancer

Full abstract

Read the full abstract

BACKGROUND: Additional gastrectomy is recommended after noncurative endoscopic submucosal dissection (ESD) for early gastric cancer (EGC) despite the low incidence of lymph node metastasis (LNM). The eCura system, W-eCura score, and 11-point score reportedly predict LNM risk; however, their comparative performance remains unclear. We aimed to externally validate and directly compare five risk stratification approaches in patients with noncurative ESD. METHODS: We retrospectively evaluated 634 consecutive patients with EGC who underwent noncurative ESD at a single high-volume center. Five prediction models were evaluated: the eCura system (7-point score and three-risk classification), W-eCura score (7-point score and three-risk classification), and 11-point score. All pathological findings were re-evaluated by an outcome-blinded expert pathologist. Model performance was assessed by discrimination, calibration, and decision-curve analysis (DCA). RESULTS: LNM was identified in 26 (4.9%) patients. All models showed comparable discrimination (c-statistics, 0.71 to 0.72). However, calibration-in-the-large and calibration slope were poorer in the W-eCura score and 11-point score (- 1.43 to - 0.96 and 0.26 to 0.64) than in the eCura system (- 0.43 to - 0.40 and 0.83 to 0.84). Using DCA, the eCura system and W-eCura score consistently showed higher net benefits than did the 11-point score across clinically relevant threshold probabilities. Recalibration improved both calibration and net benefits across all models. CONCLUSIONS: Despite similar discriminative performance, the eCura system demonstrated relatively favorable calibration and potential clinical utility in this Japanese cohort. Given their simplicity and structural similarity, three-risk classifications of the eCura system and W-eCura score, when appropriately calibrated, may support decision-making regarding additional gastrectomy.