Mayo Clinic Evaluates AI-Driven, EHR-Based Tool for Personalized Prostate Cancer Education
Mayo Clinic researchers have developed and evaluated an electronic health record (EHR)–integrated artificial intelligence tool that delivers personalized prostate cancer education using a large language model. The system, called MedEduChat, provides patients with accurate, individualized explanations based on their own medical records, according to findings published in Nature Portfolio Digital Medicine.
Patients diagnosed with prostate cancer often struggle to understand complex information about their disease and treatment options, while limited clinical visit time can restrict opportunities for in-depth discussion. MedEduChat was designed to help bridge this gap by offering clear, conversational education grounded in Mayo Clinic–validated clinical data.
The mixed-method usability study involved 15 prostate cancer patients at Mayo Clinic campuses in Arizona and Minnesota. Participants interacted with MedEduChat for 20 to 30 minutes. Following use of the tool, patients reported increased confidence in understanding their condition, with Health Confidence Scores rising from 9.9 to 13.9 on a 16-point scale. Usability was also rated highly, with an average score of 83.7 out of 100.
Patients reported that MedEduChat helped clarify unfamiliar or complex terminology and corrected misconceptions by drawing directly from their EHR. The conversational format allowed users to explore their diagnosis, treatment options, side effects, and follow-up expectations in an accessible way.
To assess clinical accuracy and safety, Wei Liu, PhD, a radiation oncology medical physicist, and three Mayo Clinic clinicians independently reviewed 85 anonymized question-and-response pairs generated by the system. MedEduChat’s responses were rated as highly correct (2.9 out of 3), complete (2.7 out of 3), and safe (2.7 out of 3). Reviewers also noted strong patient-readiness and moderate personalization based on factors such as age, cancer stage, and treatment history.
While overall performance was favorable, clinicians emphasized the importance of ongoing monitoring to mitigate potential errors stemming from incomplete or inconsistent EHR documentation. The research team incorporated a multilayer safety approach to address these risks and guide future improvements.
“This research demonstrates how large language models can be safely and effectively integrated into real clinical systems to improve cancer education,” Liu said. “By combining advanced AI with Mayo Clinic’s electronic health records, MedEduChat delivers personalized, accurate, and easy-to-understand explanations tailored to each patient’s medical history.”
Researchers plan to expand MedEduChat into broader clinical use across Mayo Clinic campuses in Arizona, Florida, and Minnesota, with future applications in additional cancer specialties.