Influence of sex differences among endoscopists in gastric cancer and precancerous condition detection rate during EGD in the artificial intelligence era: a retrospective study.

AIMS: This study investigated whether endoscopist sex is associated with the detection of gastric precancerous conditions and neoplasia during esophagogastroduodenoscopy (EGD), and whether these associations differ with or without artificial intelligence (AI) assistance. METHODS: We conducted a retrospective, single-center study of patients who underwent EGD with or without AI assistance between January 1, 2018, and June 30, 2021. The precancerous condition detection rate (PCR) and gastric neoplasia detection rate (GNR) were compared between male endoscopists (MEs) and female endoscopists (FEs) using multivariable logistic regression models, with sensitivity analyses performed to assess robustness. RESULTS: A total of 15,536 patients were included: 13,074 (84.2%) examined by 31 MEs and 2462 (15.8%) by 13 FEs. Among these, 6579 (42.3%) procedures were performed with AI and 8957 (57.7%) without. Overall, FEs were related to higher GNR (2.80% vs. 1.98%; adjusted odds ratio [aOR] 1.37, 95% CI 1.04-1.79, p = 0.023), PCR (5.56% vs. 3.50%; aOR 1.63, 95%CI 1.33-1.98 p < 0.001), and dysplasia detection rate (DDR, 1.50% vs.0.78%; aOR 1.68, 95% CI 1.12-2.45, p = 0.009) compared with MEs. In subgroup analyses, detection rates were comparable between FEs and MEs without AI assistance, whereas FEs achieved significantly higher GNR, PCR, and DDR when AI was applied. Notably, AI assistance markedly improved detection performance among FEs but not among MEs. CONCLUSION: Female endoscopists demonstrated higher detection rates of gastric cancer and precancerous conditions, and dysplasia than male endoscopists in this cohort. AI assistance was associated with improved detection performance, with a larger relative improvement observed among FEs. Because key EGD process-quality metrics (e.g., inspection time and biopsy intensity) were unavailable and AI use was not randomized, these findings should be interpreted as associations and warrant multi-center confirmation.