The role of artificial intelligence in postoperative clinical decision-making for pancreatic cancer: a pilot study

Abstract

Postoperative follow-up and treatment decisions after pancreatic cancer surgery are routinely determined by multidisciplinary tumor boards (MDTs), which represent the mainstay of clinical decision-making in these patients. However, this process is associated with increased workload, time consumption, and financial burden. Previous studies have demonstrated improved treatment planning, guideline adherence, and patient outcomes with MDTs, albeit with substantial time, cost, and administrative demands. The present study aims to compare MDT-based decision-making with an artificial intelligence (AI)–assisted model in the postoperative management of patients undergoing surgery for pancreatic cancer and to explore the potential utility of such a model. Materials and methods – An AI-based model was developed using clinical data from 67 patients discussed in multidisciplinary tumor boards between October 2020 and December 2023, in conjunction with current treatment guidelines. The model's recommendations were subsequently compared with MDT decisions in an independent cohort of 15 patients who underwent surgery in 2024. Results – The overall concordance rate between AI-generated recommendations and MDT decisions was 80%. The Cohen's kappa coefficient was 0.625 (95% CI: 0.278–0.972), indicating moderate agreement beyond chance. Three discrepant cases were further analyzed to explore potential reasons for discordance. Discussion – The findings suggest that AI-assisted decision-support systems may approximate MDT recommendations in postoperative pancreatic cancer management. However, observed discrepancies highlight the continued importance of expert clinical judgment and contextual interpretation in complex decision-making scenarios. Conclusion – AI-based models may serve as supportive tools in postoperative clinical decision-making by potentially reducing workload and time burden, but should complement rather than replace multidisciplinary expert evaluation. © 2026 Yigman, Demirel, Halil Ozata, Çelik, Toprak, Karahan, Adsay, Bilge and Tellioğlu.