Impact of AI-Assisted Endoscopy on Secondary Review and Referral Decisions in Gastric Cancer Screening: A Multicenter Real-World Study.

OBJECTIVES: Artificial intelligence (AI)-assisted endoscopy may support gastric cancer screening, but its impact on structured secondary human review remains unclear. We evaluated secondary review modifications and referral decisions before and after AI implementation. METHODS: This retrospective multicenter study included 5013 screening esophagogastroduodenoscopy examinations performed before (n = 2021) and after (n = 2992) AI implementation. The primary outcome was secondary review modifications. Secondary outcomes included referral for further diagnostic evaluation. Multivariable logistic regression analyses adjusted for age, sex, Helicobacter pylori eradication status, prior-year endoscopy, and endoscopist experience were performed. Interaction analyses according to endoscopist experience and inverse probability of treatment weighting analyses were conducted. No external reference standard was available to verify secondary-review modifications or referral decisions. RESULTS: Secondary review modifications increased from 1.2% in the pre-AI period to 3.7% in the AI period (p < 0.01). A significant interaction was observed between AI implementation and endoscopist experience (p for interaction < 0.01), with increased modifications primarily among nonexpert endoscopists. Cancer-related modifications remained uncommon but increased after AI implementation (0.6% vs. 0.2%, p = 0.04). Referral rates decreased from 4.9% to 2.8% (p < 0.01), and AI implementation was independently associated with lower odds of referral (adjusted OR = 0.53; 95% confidence interval = 0.36-0.77). CONCLUSIONS: AI-assisted endoscopy was associated with lower referral rates; however, secondary review modifications persisted after AI implementation, particularly among nonexpert endoscopists. These findings describe workflow changes and do not establish the clinical validity of the modifications or the appropriateness of referral decisions.