A community-codesigned LLM-powered chatbot for primary care: a randomized controlled trial

Area of Science:

• Primary Healthcare
• Artificial Intelligence in Medicine
• Public Health

Background:

• Global shortage of primary healthcare physicians, especially in resource-limited settings.
• Large language models (LLMs) show potential for enhancing patient health awareness.
• Need for innovative digital health solutions to support primary care.

Purpose of the Study:

• To develop and evaluate an LLM-powered primary care chatbot, P&P Care.
• To assess the impact of P&P Care integrated with e-learning on objective health awareness.
• To explore the utility of a dual-track role-play codesign framework for LLM deployment.

Main Methods:

• Developed P&P Care using a dual-track role-play codesign framework with community stakeholders and researchers.
• Integrated the chatbot with e-learning modules.
• Conducted a randomized controlled trial with 2,113 participants across 11 Chinese provinces, comparing P&P Care with e-learning to a consultation-only group.

Main Results:

• The e-learning group using P&P Care demonstrated significantly higher objective health awareness (mean score 2.95 ± 1.22) compared to the consultation-only group (mean score 2.34 ± 1.02; P < 0.001).
• The study met its primary endpoint, indicating the effectiveness of the LLM-powered intervention.
• The codesign approach facilitated the development and testing of the chatbot.

Conclusions:

• LLM-powered chatbots, like P&P Care, can effectively enhance patient health awareness.
• The dual-track role-play codesign framework is a scalable solution for deploying LLMs in resource-limited primary healthcare settings.
• P&P Care shows promise for improving health education and outcomes in underserved populations.