AI meets gastroenterology practice: mapping large language model use across Europe – EuroGI-AI project

Le résumé fourni par la source

Background and aims The use of large language models (LLMs), such as ChatGPT, is rapidly expanding in clinical medicine but data in gastroenterology are limited. We aimed to assess the prevalence and patterns of LLM use among European gastroenterologists, evaluate perceived benefits and risks and identify barriers and priorities for future integration. Methods A Europe-wide online survey was conducted between June and August 2025. A 24-item questionnaire was distributed via national gastroenterology societies and social media. Questions on specific applications, reliability and clinical impact were shown only to respondents with prior LLM experience. Results A total of 966 gastroenterologists from 34 European countries responded, 55.3% reported using LLMs, mainly weekly (54.4%) or daily (20.2%). Common uses included education (59.2%), clinical decision support (58.2%) and scientific writing (39.1%). Among academic users, summarisation (80.1%) and grammar refinement (74.4%) were most frequent. Most users rated LLMs as moderately or highly reliable (81.0%), and 84.3% believed they improve outcomes. Nearly two-thirds (64.5%) felt LLMs can support endoscopy practice. Inaccuracies were reported by 42.7% and were associated with lower trust (ρ=–0.40, p<0.001). Although only 7.8% had formal training, 90.0% supported structured education and 85.2% favoured curricular integration. Most respondents would recommend LLM use (91.4%) and expect broader adoption within 5 years (93.4%). Reported barriers included institutional regulations (45.7%), cost (36.1%), complexity (29.4%) and time constraints (27.9%). Conclusion LLMs are increasingly integrated into clinical and academic workflows in European gastroenterology. Structured training, specialty-specific development and clear governance are essential for safe implementation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.