Abstract Aims Breast cancer multidisciplinary team (MDT) meetings are vital to providing optimal patient care. Discussions are complex and produce patient-individualised treatments, which can be difficult to effectively translate into patient correspondence. Due to the ability to process complex information efficiently and respond with conversational text (Gokul 2023), large language models (LLMs) represent a potential aid for patient communication. This study aims to evaluate the accuracy and accessibility of patient letters generated by Claude 3 Opus using clinical decisions from the MDT. Methods Claude will be provided with Breast MDT outcomes and prompted to summarise the decision and generate a patient letter. The letters will be evaluated by two reviewers scoring against a rubric and generating qualitative feedback. Assessments will focus on accuracy, fluency and accessibility, using a readability test (such as automated readability index). Results Initial pilot outcomes demonstrate well-structured patient letters with clear summaries of intended treatment plans. Claude outlined clinical reasoning to justify MDT decisions and suggested specific lifestyle advice and guidance on self-monitoring between appointments. As conversational dialogue is a strength of LLMs, we predict strong performance with the use of coherent and patient-friendly language. Conclusion We expect our results to support the use of Claude to assist clinicians in writing patient communication. Expected advantages include streamlining the patient correspondence process and improving patient satisfaction. Despite anticipated results, we expect that using LLMs in this way should only enhance clinicians’ efficiency in written patient communication rather than replace this aspect of their role.
Oliver-Blaney et al. (Fri,) studied this question.