b'Scientific Session Abstracts SATURDAY APRIL 20, 2024 4: CHATGPT ABLE TO GENERATE ADEQUATE RESPONSES TO PATIENT FAQS IN VASCULAR SURGERY Sean Perez MD, Elsie Ross MD, MS, RPVI, FAHA,Mahmoud Malas MD, MHS, RPVI, FACSDivision of Vascular & Endovascular Surgery, Department of Surgery,University of California San Diego (UCSD), La Jolla, CaliforniaBackground: Information on the internet regarding vascular surgery can be unreliable and/or difficult for patients to understand. Large language models (LLMs) have been popularized recently with widespread uptake and many use cases. ChatGPT, the most popular and readily available LLM currently, was trained on a large data set derived from the internet and uses it to respond to user-generated prompts, including medical questions, with easy-to-understand language. Given this unique ability, patients may use it as a resource to seek additional information on a particular diagnosis or surgical treatment within the field of vascular surgery. However, the adequacy of its responses regarding topics within vascular surgery is not well understood. The goal of this study was to assess the ability of ChatGPT to respond to patient frequently asked questions (FAQs) related to vascular surgery as judged by board-certified vascular surgeons.Methods: Patient FAQs were gathered to use as ChatGPT prompts via internet search from vascular surgery societies, institution websites, Facebook patient support groups, and device manufacturer FAQ webpages. Prompts and their responses were randomly selected to compile a 10-question survey to be completed by board-certified vascular surgeons at a single academic institution. The 10 questions covered topics including peripheral arterial, carotid, aortic, and venous pathologies, endovascular principles, and dialysis access (Table I).ChatGPT responses were graded on a Likert-type scale as 4-Comprehensive, 3-Accurate but inadequate, 2-Correct but with some incorrect information, and 1-Completely incorrect. Scores were binarized as to redefine scores 2 as Acceptable and remaining scores as unacceptable for patient-facing responses uses.Results: Survey completion was 100%. Only one response to a question regarding dialysis access by ChatGPT received a unanimously agreed upon comprehensive score by the reviewers. Most reviewers found all 10 ChatGPT generated responses to be acceptable (Table II). FAQs on the topics of dialysis access, PAD, and vein pathology were deemed to be the most comprehensive. One reviewer graded 7 out of the 10 responses as unacceptable (scoresof 1 or 2).Conclusion: ChatGPT is a useful tool for patients seeking basic information regarding vascular pathology or procedures on the internet. Utilizing ChatGPT prior to an initial consultation could serve as helpful primer for patients, 40SOUTHERN CALIFORNIA VASCULAR SURGICAL SOCIETY'