Multidisciplinary tumor board decisions and artificial intelligence-generated recommendations in general surgery: a retrospective observational study.

Artificial intelligence (AI)-based decision support systems are increasingly explored in surgical oncology. However, their concordance with multidisciplinary tumor board (MTB) decisions, particularly in complex gastrointestinal malignancies, remains insufficiently characterized. This retrospective observational study included 47 patients who underwent oncological surgery and were discussed at a multidisciplinary tumor board during an eighteen-month period. For each patient, concordance between MTB-established treatment decisions and subsequent AI-generated recommendations (ChatGPT-based) was assessed. Concordance was categorized as discordant (0), partially concordant (1), or fully concordant (2). Discordant cases were further analyzed across predefined domains, including staging discrepancies, resectability assessment, interpretation of metastatic disease burden, and treatment sequencing. Full concordance was observed in 18 cases (38.3%), partial concordance in 22 cases (46.8%), and discordance in 7 cases (14.9%). Overall, 85.1% of cases demonstrated at least partial concordance. Discordance was more frequent among male patients (5 of 19; 26.3%) compared with female patients (2 of 28; 7.1%). Qualitative analysis revealed that discordance most commonly arose from differences in resectability assessment and staging interpretation. AI recommendations tended to favor broader surgical candidacy, whereas MTB decisions more frequently excluded surgery based on nuanced clinical and contextual factors. AI-generated recommendations show substantial overlap with MTB decisions in most cases but diverge at critical surgical decision points, particularly regarding resectability and staging. These findings suggest that AI may serve as a complementary decision-support tool rather than a substitute for multidisciplinary clinical judgment for oncological surgery care.