Effective patient education relies heavily on individualized explanations from clinical professionals, who can contextualize treatment options while addressing patients’ psychological and social concerns. However, clinicians often have limited time and capacity to engage in these in-depth educational dialogues. When healthcare resources are strained, understanding evolving patient communication needs is frequently among the first priorities to be reduced. As a result, many cancer treatment pathways lack a comprehensive and timely educational component, leaving patients without adequate support during some of the most stressful periods of their lives.
Institutions such as Mayo Clinic have implemented alternative educational resources, including e-learning modules with videos and images to explain diagnoses, tests, and treatment options. While these tools provide valuable baseline information, they cannot fully address the highly personalized questions patients and their family caregivers often have. Common questions include: “What is my survival rate?”, “What are the risk factors of polycythemia?”, or “Should I stop drinking whole milk? I’ve heard dairy products affect prostate cancer.” These questions arise not only during clinic visits, but also during family discussions, after treatment completion, and when patients consider genetic risk or screening needs for relatives.
Large language models (LLMs) offer a new opportunity to bridge this educational gap by functioning as conversational companions for cancer patients. By leveraging capabilities in information summarization, explanation, and real-time natural language interaction, LLM-based chatbots tailored to specific cancer domains can provide personalized, context-aware responses to patient inquiries. Beyond delivering medical information, these systems also have the potential to offer emotional reassurance and continuity of support outside traditional clinical encounters.
Unlike conventional rule-based chatbots, LLM agents enable more nuanced and individualized engagement. They can respond dynamically to patient concerns, enhancing clinical education through timely, conversational interactions. By helping patients better understand what to expect—both physically and emotionally—after diagnosis or during treatment, LLM-based tools may improve patient confidence and preparedness.
In a recent study conducted at Mayo Clinic, we developed an LLM-based agent called MedEduChat, designed to connect directly with prostate cancer patients’ clinical profiles. MedEduChat integrates patient-specific information, including cancer type and stage, laboratory results, treatment history, symptoms, medication use, family medical background, and demographic data. For patients who have already undergone treatment, the system also incorporates relevant treatment-related details to contextualize responses.
The MedEduChat prompting framework combines closed-domain patient data with semi-structured guidance informed by a sustainable patient education model. The agent was designed around three core objectives:
- Providing detailed explanations of health outcomes,
- Enhancing patient learning, and
- Promoting engagement by soliciting user feedback on interaction quality.
We evaluated MedEduChat with 15 prostate cancer patients and closely observed the personalized questions they raised during interactions. Patients described the system as a valuable educational intermediary. One participant noted, “MedEduChat is like a gatekeeper for my questions—it helps ease the anxiety of waiting for answers, even when the answer isn’t fully resolved yet.” Another reflected on a complex medication-related question: “I asked about Zanubrutinib, a drug rarely used in prostate cancer. I didn’t expect a yes-or-no answer, but the explanation helped me understand what the drug does and how it differs from treatments related to my cancer.”
These examples illustrate how AI can move beyond static educational materials to provide personalized, interactive support for cancer patients. LLM-based chatbots can be integrated into existing clinical workflows to address patient questions in a timely, relevant, and scalable manner. By helping patients navigate complex medical information and emotional uncertainty, AI-enabled patient education reflects a broader shift toward more responsive, patient-centered technologies in healthcare.
Written by: Wei Liu, PhD, DABR, FAAPM, FAIMBE, Professor of Radiation Oncology, Consultant of Department of Radiation Oncology, Co-Lead of Workstream in Data Science, AI, and LLMs of Mayo Clinic Comprehensive Cancer Center in Arizona, Research Director of Division of Medical Physics, Mayo Clinic College of Medicine and Science, Mayo Clinic, Phoenix, AZ
Read the Abstract