335 Can ChatGPT Choose the Right Bariatric Operation? a Comparison of Large Language Model and MDT Decision Making

Abstract Aim Large Language Models (LLMs) like ChatGPT show promise in simulating medical reasoning. However, their role in complex surgical decision-making, such as providing MDT-level recommendations, remains unclear. We assessed how ChatGPT compares with an experienced bariatric MDT in recommending procedures for weight loss surgery. Method Fifty consecutive bariatric patients from a UK tertiary centre were retrospectively analysed. Structured clinical data were input into ChatGPT-4o using a detailed prompt to recommend an operation: sleeve gastrectomy (LSG), Roux-en-Y gastric bypass (LRYGB), one-anastomosis gastric bypass (OAGB), or no surgery, with justification. LLM outputs were screen-recorded and extracted verbatim. Recommendations were compared with both MDT decisions and procedures performed. Concordance was assessed across multiple domains, including by operation type. Results The MDT recommended LRYGB in 80% (40/50) and LSG in 20% (10/50). ChatGPT matched MDT recommendations in 44% (22/50) of cases, with better alignment when LSG was chosen (80%) than LRYGB (35%). ChatGPT matched the procedure actually performed in 50% of cases. In 30% of patients, the surgery differed from the MDT plan, often aligning more closely with ChatGPT’s recommendation. This may reflect intraoperative findings or evolving surgical judgement and underscores how LLM outputs can mirror real-world variability. Conclusions ChatGPT shows partial concordance with MDT decisions, particularly in recommending LSG. However, it under-recommends LRYGB, even for metabolically complex patients. While not a replacement for MDTs, LLMs may offer insights into decision boundaries and case ambiguity. Screen-recorded outputs enable live demonstration and audience engagement during presentation.