Abstract Aims Multidisciplinary team meetings are key to providing optimal care to breast cancer patients. However, increasing caseloads and complexity have lengthened these meetings. Large language models (LLM) can efficiently process large amounts of information and respond with conversational prose (Gokul 2023). This study aims to evaluate the ability of LLMs, specifically Anthropic’s Claude 3 Opus, to accurately predict MDT outcomes within Breast MDT. Methods A prospective diagnostic concordance and validation study, assessing Claude’s ability to interpret clinical information and provide guideline-based recommendations for management. We will provide Claude with the same clinical information that is presented at MDT and prompt it to generate treatment plans based on current Association of Breast Surgery and National Institute of Health and Care Excellence guidelines. We will review and generate a concordance score comparing Claude’s output with the MDT decision, marked against critical elements of patient care and validated by the MDT. Results Preliminary results demonstrate that Claude outputs well-structured recommendations and signposts its use of the guidelines to justify its proposed treatment plans. In addition, its conversational nature enables it to provide clear clinical reasoning for decisions. We anticipate our results to demonstrate good concordance between Claude and actual MDT outcomes. Conclusion If LLMs such as Claude can generate accurate, guideline-based treatment plans, this could significantly reduce the time required to discuss each case at Breast MDT. This could allow the team to allot more time to discuss complex cases, resulting in enhanced overall patient care.
Oliver-Blaney et al. (Fri,) studied this question.