PatientEase-Domain-Aware RAG for Rehabilitation Instruction Simplification.

BACKGROUND: Rehabilitation depends on using instructional materials, which many patients find difficult to understand; thus, their adherence to the safety and care may be affected. Text simplification systems used, in general, do not usually focus on procedure-oriented guidance or the degree of personalization required in rehabilitation settings. METHODS: We present PatientEase, a domain-aware retrieval-augmented generation framework that changes rehabilitation instructions to simple words without changing the clinical meaning. PatientEase incorporates two complementary retrievers that is a corpus retriever that is tuned for rehabilitation and a user-aligned retriever that is conditioned on patient profiles, together with a role-structured, multi-agent rewriting pipeline; outputs can be further refined by using reinforcement learning from human feedback with a composite reward for readability, factuality, and clinician-preferred structure. RESULTS: The latter was quite comprehensively compared in four benchmark tests against baselines, wherein SARI, FKGL, BERTScore, and MedEntail indices are employed, as well as clinician-patient assessments. PatientEase achieves 52.7 SARI and 92.1% factual entailment, and receives the highest fluency and simplicity ratings; ablations also underline each module's role. CONCLUSIONS: PatientEase paves the road for safer, patient-centered communication in rehabilitation and lays the groundwork for trustworthy clinical dialogue systems.