PatientEase-Domain-Aware RAG for Rehabilitation Instruction Simplification

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Patient Education

Background:

• Rehabilitation relies on instructional materials often difficult for patients to comprehend, impacting adherence to care.
• Existing text simplification systems lack focus on procedure-oriented guidance and personalization crucial for rehabilitation.

Purpose of the Study:

• To develop PatientEase, a novel framework for simplifying rehabilitation instructions while preserving clinical meaning.
• To enhance patient comprehension and adherence through personalized, easy-to-understand guidance.

Main Methods:

• PatientEase utilizes a domain-aware retrieval-augmented generation framework with specialized corpus and user-aligned retrievers.
• A role-structured, multi-agent rewriting pipeline refines instructions, with optional reinforcement learning from human feedback for improved quality.

Main Results:

• PatientEase demonstrated high performance in benchmark tests, achieving 52.7 SARI and 92.1% factual entailment.
• The system received top ratings for fluency and simplicity, with ablations confirming the importance of each module.

Conclusions:

• PatientEase promotes safer, patient-centered communication in rehabilitation settings.
• The framework establishes a foundation for developing trustworthy clinical dialogue systems.