Personalizing prostate cancer education for patients using an EHR-Integrated LLM agent

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Clinician workload often limits timely patient education and personalized support in cancer care.
• Prostate cancer patients require accessible and accurate information for better treatment adherence and outcomes.

Purpose of the Study:

• To develop and evaluate MedEduChat, a Large Language Model (LLM) agent integrated with electronic health records (EHR).
• To assess the impact of MedEduChat on prostate cancer patient education, engagement, and health confidence.
• To evaluate the correctness, completeness, safety, and personalization of the LLM agent by clinicians.

Main Methods:

• A quality improvement study involving fifteen non-metastatic prostate cancer patients and three clinicians.
• Integration of the MedEduChat LLM agent with the clinic's EHR system.
• Data collection through patient interaction logs, usability scores (UMUX), and clinician evaluations of interaction history.

Main Results:

• MedEduChat achieved a high usability score (UMUX = 83.7/100).
• Patient health confidence significantly improved (Health Confidence Score rose from 9.9 to 13.9).
• Clinicians rated MedEduChat as highly correct (2.9/3), complete (2.7/3), and safe (2.7/3), with moderate personalization (2.3/3).

Conclusions:

• LLM agents like MedEduChat show significant potential for enhancing patient engagement in cancer care.
• The integration of LLMs with EHRs can effectively augment patient education and support, addressing clinician workload limitations.
• MedEduChat demonstrates a promising approach to improving the quality and accessibility of health information for prostate cancer patients.