Leveraging Large Language Models for Generating Responses to Patient Messages.

Area of Science:

• Artificial Intelligence in Healthcare
• Natural Language Processing for Clinical Communication

Background:

• Patient portal messages are crucial for patient-provider communication.
• Efficient and empathetic responses are essential for patient satisfaction and care.

Approach:

• Developed and fine-tuned two large language models (LLMs): CLAIR-Short and CLAIR-Long, using a large dataset of patient messages and physician responses.
• CLAIR-Long incorporated patient education and empathetic language, enhanced via OpenAI API.
• Evaluated model performance using primary care physician feedback on generated responses compared to ChatGPT.

Key Points:

• CLAIR-Short generated concise responses, mimicking provider style.
• CLAIR-Long provided enhanced patient education and empathetic communication.
• Physicians rated CLAIR-Long responses similarly to ChatGPT for empathy, responsiveness, and accuracy.

Conclusions:

• Fine-tuned LLMs can significantly improve patient-provider communication via electronic health record portals.
• AI-generated responses have the potential to increase efficiency and patient engagement in primary care settings.