2671 - Trust as a Determinant of Acceptability of AI Chatbots for Clinical Trial Education: A Qualitative Focus Group Study
Presenter(s)
J. Zeng1, Z. Xu2, A. Yaghoubi3, V. Goddla4, B. Zhai5, Y. H. Chen6, T. Brown7, J. Maues8, B. Silverman9, N. E. Martin10, E. Sharon11, D. E. Kozono12, L. Lehmann13, D. Dligach3, D. S. Bitterman1, and A. Revette11; 1Department of Radiation Oncology, Mass General Brigham/Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, 2Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA, 3Loyola University, Chicago, IL, 4Harvard College, Cambridge, MA, 5Brigham & Women's Hospital, Boston, MA, 6Dana Farber Cancer Institute, Boston, MA, 7Boston, MA, 8GRASP Cancer, Baltimore, MD, 9Mass General Brigham, Boston, MA, 10Department of Radiation Oncology, Brigham and Women’s Hospital, Dana-Farber Cancer Institute and Harvard Medical School, Boston, MA, 11Dana-Farber Cancer Institute, Boston, MA, 12Department of Radiation Oncology, Dana-Farber Cancer Institute/Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, 13Harvard Medical School, Boston, MA
Purpose/Objective(s): Educating patients about clinical trials continues to be challenging. LLM-mediated conversational agents are increasingly used to support clinical trial education yet more research is needed on integration into clinical and research workflows. Patient-centered LLM development is essential for effective clinical translation. We conducted focus groups to explore stakeholder attitudes toward AI chatbots in clinical trial education, with the primary objective of identifying and characterizing stakeholder acceptance.
Materials/Methods: Three patient focus groups (21 participants) and one clinical research coordinator focus group (6 participants) were conducted using semi-structured guides exploring perceptions of AI chatbots for trial education as well as review a preliminary chatbot prototype. Sessions were audio-recorded, transcribed verbatim, and analyzed using applied thematic analysis. A hybrid inductive–deductive coding approach was used to develop a codebook. Two investigators led coding and analysis, with discrepancies and uncertainties discussed in recurrent team meetings. Themes were organized around several key domains, and here we present themes associated with trust.
Results: Across stakeholder groups, trust was a critical dimension of perspectives on overall AI chatbot acceptance for clinical trial education. Stakeholders advocated for a hybrid model, where AI-driven accuracy and privacy are integrated into continued human-led engagement. Participants identified the pre-existing clinician-patient relationship as the essential anchor for introducing AI tools. Additionally, having an option to bypass the chatbot and/or connect to a live person if the chatbot could not provide the necessary answers was preferred. While concerns regarding bias and sycophantic behavior were noted, participants suggested that formal institutional 'sign-off' and endorsed vetting protocols serve as essential proxies for trust, supported by verifiable citations for all AI-generated responses. The most salient privacy concerns were around data storage and sharing, with stakeholders wanting transparency and autonomy in their ability to opt in and out of data sharing. Historical medical mistrust and digital literacy gaps were highlighted as barriers requiring specialized, transparency-first engagement strategies for marginalized groups.
Conclusion: Trust in AI chatbots for clinical trial education is not inherent and must be intentionally designed through transparency, human oversight, institutional accountability, and clear privacy protections. Implementation strategies should prioritize clinician endorsement, visible citation practices, and explicit disclosure of chatbot governance. These findings inform the responsible integration of AI tools into oncology trial education and highlight design features necessary for ethical deployment in clinical practice.