Five‐Year Unified Risk Prediction Model for Gastric and Esophageal Cancer Based on Nationwide Health Check‐Up Data in Japan: J‐GESS Score
Kenta Watanabe, Sho Fukuda, Dai Kubota, Nobutake Yamamichi, Yu Takahashi, Yoshitaka Watanabe, Kyoichi Adachi, Norihisa Ishimura, Tomoyuki Koike, Hideyuki Sugawara, Kiyotaka Asanuma, Yasuhiko Abe, Takashi Kon, Eikichi Ihara, Kazuhiro Haraguchi, Yoshihiro Otsuka, Rie Yoshimura, Yugo Iwaya, Takuma Okamura, Noriaki Manabe, Akira Horiuchi, Mio Matsumoto, Kengo Onochi, So Takahashi, Tatsuki Yoshida, Yosuke Shimodaira, Katsunori Iijima
Abstract
BACKGROUND AND AIM: In Japan, gastric cancer (GC) and esophageal squamous cell carcinoma (ESCC) account for nearly all fatal upper gastrointestinal malignancies. A unified model that simultaneously predicts the incidence of these two cancers could help optimize endoscopic screening and postindex surveillance strategies. METHODS: We conducted a retrospective multicenter cohort study of 33 422 individuals who underwent screening esophagogastroduodenoscopy (EGD) at 17 health check-up institutions across Japan. Clinical and endoscopic data from the index EGD were used to develop a 5-year risk prediction model for GC or ESCC. Significant predictors were identified using Cox regression, and a scoring system (J-GESS score) was constructed by scaling β coefficients. Model performance was evaluated using the 5-year inverse probability of censoring weighted (IPCW) Brier score, calibration, and c indices. Internal validation used bootstrap resampling (500 iterations) to obtain optimism-corrected estimates. Decision curve analysis (DCA) assessed clinical utility at 5 years. RESULTS: At 5 years, the IPCW-estimated cumulative incidence of GC or ESCC was 0.84% (234 events). Independent predictors included older age, male sex, ever smoking, and endoscopic gastric mucosal atrophy. The model showed good calibration at 5 years (calibration-in-the-large 0.005; slope 1.055; identical after optimism correction) and robust discrimination (Harrell's c index 0.81; Uno's 5-year c index 0.81). DCA demonstrated positive net benefit across clinically relevant thresholds. CONCLUSION: Using routinely collected health check-up data, we developed a unified, practical, and internally validated 5-year prediction model for GC and ESCC. The J-GESS score enables effective risk stratification and may inform postindex endoscopic surveillance and screening interval optimization in Japan.