Developing provider digital twins for personalized provider-patient communication via a RAG-based conversational...

Area of Science:

• Artificial Intelligence in Medicine
• Digital Health
• Clinical Communication Systems

Background:

• Digital twins are advancing personalized medicine through data-driven individual modeling.
• Existing research primarily focuses on patient digital twins, neglecting the provider's role in clinical communication.
• Effective clinical communication is crucial for tailored patient interventions and healthcare outcomes.

Purpose of the Study:

• To develop and evaluate GRACE (Generalized RAG-Enhanced Conversation Framework) for constructing provider digital twins (ProDTs).
• To emulate clinicians' communicative and cognitive behaviors for improved provider-patient interactions.
• To explore the application of ProDTs in proactive and context-aware healthcare communication.

Main Methods:

• GRACE integrates three modules: physician-informed dialog script generation, a Retrieval-Augmented Generation (RAG) pipeline for knowledge updating, and an LLM-based conversational interface.
• The framework was tested using HPV vaccination counseling as a use case.
• Evaluation involved the HealthBench benchmark and a user study with clinician feedback.

Main Results:

• GRACE demonstrated feasibility, trustworthiness, and adaptability in emulating provider communication patterns.
• The framework supports proactive and context-aware provider-patient communication.
• Results indicate the potential for cognitively grounded digital twins in scalable healthcare applications.

Conclusions:

• GRACE represents a conceptual advancement in creating provider digital twins for healthcare.
• The framework enhances clinical communication by modeling provider behavior.
• This technology paves the way for safer, more scalable, and personalized digital twin applications in medicine.