NoteChat: A Dataset of Synthetic Patient-Physician Conversations Conditioned on Clinical Notes.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Clinical Informatics

Background:

• Clinical documentation is a major contributor to physician burnout.
• Generating accurate and efficient patient-physician dialogues is crucial for medical records.
• Existing methods for synthetic dialogue generation have limitations.

Purpose of the Study:

• To introduce NoteChat, a novel cooperative multi-agent framework using Large Language Models (LLMs) for generating patient-physician dialogues.
• To evaluate the effectiveness of NoteChat in producing synthetic dialogues that improve clinical note generation.
• To demonstrate NoteChat's superiority over state-of-the-art models in dialogue quality.

Main Methods:

• Developed a cooperative multi-agent framework (NoteChat) with role-specific LLMs.
• Utilized structured role-play and strategic prompting for LLM interaction.
• Augmented the MTS-dialogue dataset with synthetic dialogues generated by NoteChat.
• Conducted comprehensive automatic and human evaluations involving domain experts.

Main Results:

• NoteChat-augmented training data improved clinical note generation performance.
• NoteChat significantly outperformed state-of-the-art models, including ChatGPT and GPT-4.
• Expert evaluation showed NoteChat generated superior synthetic patient-physician dialogues, exceeding current models by up to 22.78%.

Conclusions:

• NoteChat offers a powerful new approach for generating high-quality synthetic patient-physician dialogues.
• The framework has the potential to enhance clinical documentation processes.
• NoteChat can contribute to reducing physician burnout by streamlining documentation tasks.