Attitudes toward large language model-based Artificial Intelligence systems as an information source for shared decision-making in radiation oncology

Le résumé fourni par la source

BACKGROUND: Implementing structured shared decision-making (SDM) requires high-quality, reliable patient information. In radiation oncology, patients often have limited knowledge and misconceptions about therapy and side effects, affecting their decision-making. Large language model-based AI systems (LLMs) may help by providing evidence-based information in accessible language, but successful implementation depends on the willingness of patients and health care professionals (HCPs) to adopt these technologies. METHODS: A survey was conducted among patients undergoing radiation therapy and HCPs between 03/2024 and 02/2025. Data was collected using structured electronic questionnaires (32 items for patients, 35 for HCPs). The survey assessed sociodemographic characteristics, the status of SDM in oncology, sources of information relevant to SDM, and current and anticipated LLM applications. Data were analyzed using descriptive statistics and logistic regression analysis. RESULTS: The internet was the prime information source for patients (n = 400). Regarding current use of LLMs, a large discrepancy between patients and HCPs (n = 200) was observed (18.2% vs 69.5%). Although 77% of HCPs believed that patients will rely on LLMs in the future, only 29.1% of patients agreed. Most patients (65.8%) stated that even as LLMs improve, they will continue to trust physicians more; 46% of HCPs shared this view. Only 16.5% of patients were convinced that LLMs provide all relevant data for SDM in cancer care. Familiarity with technology was the strongest predictor of LLM use among patients. CONCLUSION: Only a minority of radiation oncology patients currently use LLMs, and many remain skeptical about their future role-contrasting with the more optimistic expectations of HCPs.

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