EHRTutor: Enhancing Patient Understanding of Discharge Instructions
Project Overview
The document introduces EHRTutor, an innovative generative AI framework aimed at improving patient comprehension of discharge instructions through interactive question-answering. Leveraging large language models (LLMs), EHRTutor generates tailored questions derived from electronic health records and facilitates a conversational dialogue with patients to assess their understanding. Additionally, the system provides concise summaries to reinforce the learning process. Evaluation outcomes demonstrate that EHRTutor significantly surpasses traditional models in enhancing both patient education and engagement, indicating its effectiveness in promoting better health literacy. This application of generative AI exemplifies its potential to transform educational practices in healthcare settings, ultimately supporting improved patient outcomes and fostering a more informed patient population.
Key Applications
EHRTutor
Context: Patient education regarding discharge instructions, targeting patients post-discharge.
Implementation: EHRTutor integrates question generation, conversational interaction, and summarization modules based on LLMs. It utilizes a template-based approach for question generation and the ReAct framework for conversational interaction.
Outcomes: EHRTutor significantly improves patient engagement and understanding of discharge instructions, facilitating better self-managed care post-discharge. Evaluation shows a preference for EHRTutor's performance over existing methods.
Challenges: Potential issues include the LLM's propensity for hallucinations, leading to incorrect or harmful information being provided to patients, and the challenge of generating relevant and comprehensible questions.
Implementation Barriers
Technical Barrier
LLMs may produce hallucinated responses, leading to the risk of providing incorrect or misleading information to patients.
Proposed Solutions: Implement verification mechanisms and utilize template-based prompting to ensure generated content is relevant and accurate.
User Understanding Barrier
Patients may struggle with medical terminology and complex questions, which can hinder their understanding.
Proposed Solutions: Simplify language and provide contextual hints during conversations to enhance patient comprehension.
Project Team
Zihao Zhang
Researcher
Zonghai Yao
Researcher
Huixue Zhou
Researcher
Feiyun ouyang
Researcher
Hong Yu
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Zihao Zhang, Zonghai Yao, Huixue Zhou, Feiyun ouyang, Hong Yu
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai