CHAT-RT study: ChatGPT in radiation oncology-a survey on usage, perception, and impact among DEGRO members.

Abstract

Background: Radiation oncology is increasingly turning to Artificial Intelligence (AI) - and in particular Chat Generative pre-trained transformer (ChatGPT) - for decision support, patient education, and workflow efficiency. Despite promising gains, questions about accuracy, General Data Protection Regulation (GDPR)-compliance and ethical use persist, especially in high-stakes cancer care. To clarify real-world attitudes and practices, we surveyed members of the German Society of Radiation Oncology (DEGRO) on their use, perceptions, and concerns regarding ChatGPT across clinical, research, communication, and administrative tasks.

Methods: An anonymous online survey was implemented via LimeSurvey platform and distributed to all members of the DEGRO in Germany, Austria, and Switzerland between April and June 2024. The 40-item questionnaire-covering demographics, radiotherapy experience, and ChatGPT's clinical, research, communication, and administrative applications-was developed through a narrative literature review, ChatGPT-assisted drafting, back-translation, expert validation, and pilot testing. Fully completed responses were used for descriptive statistics and analysis.

Results: Of 213 respondents, 159 fully completed the survey. Participants were predominantly based in Germany (92.5%), worked in university hospitals (74.2%), and identified as radiation oncologists (54.7%), with a broad range of radiotherapy experience (< 1 year: 7.5%; >15 years: 24.5%). Awareness of ChatGPT was high (94.9%), yet actual use varied: 32.1% never used it, while 35.2% employed it regularly for administrative tasks and 30.2% for manuscript drafting. Mid-career clinicians (6-10 years' experience) showed the greatest enthusiasm-44% agreed it saves time and 72% planned further integration-though all career stages (71.7% overall) expressed strong interest in formal training. Satisfaction was highest for administrative (94.6%) and manuscript support (91.7%) but lower for technical queries (66.7%). Major concerns included misinformation (69.2%), erosion of critical thinking (57.9%), and data-privacy risks (57.2%).

Conclusion: Our survey demonstrates high awareness and adoption of ChatGPT for administrative and educational tasks, alongside more cautious use in clinical decision-making. Widespread concerns about misinformation, critical-thinking erosion, and data privacy-especially among early- and mid-career clinicians-underscore the need for targeted AI training, rigorous validation, and transparent governance to ensure safe, effective integration into patient care.