A Two-Stage RAG-Based System for Personalized Medical Plan Generation
Despite recent success in applying large language models (LLMs) to electronic health records (EHR), most systems focus primarily on assessment rather than treatment planning. We identify three critical limitations in current approaches: they generate treatment plans in a single pass rather than following the sequential reasoning process used by clinicians; they rarely incorporate patientspecific historical context; and they fail to effectively distinguish between subjective and objective clinical information.
Motivated by the SOAP methodology (Subjective, Objective, Assessment, Plan), we introduce MedPlan, a novel framework that structures LLM reasoning to align with real-life clinician workflows. Our approach employs a two-stage architecture that first generates a clinical assessment based on patient symptoms and objective data, then formulates a structured treatment plan informed by this assessment and enriched with patient-specific information through retrievalaugmented generation.
Comprehensive evaluation demonstrates that our method significantly outperforms baseline approaches in both assessment accuracy and treatment plan quality.
Our two-stage framework mirrors the clinical process used by healthcare professionals
Generates clinical assessment (A) based on patient symptoms (S) and objective data (O), incorporating both self-history and cross-patient references through RAG.
Formulates structured treatment plan (P) informed by the generated assessment and enriched with patient-specific context and similar case references.
Leverages patient history and cross-patient similarities using hybrid retrieval with BM25 and semantic search, refined by cross-encoder re-ranking.
MedPlan significantly outperforms baseline approaches across all evaluation metrics
Interactive system for real-world clinical deployment
The system provides physicians with an intuitive interface for generating structured, evidence-based treatment plans
@misc{hsu2025medplanatwostageragbasedpersonalized,
title={MedPlan: A Two-Stage RAG-Based System for Personalized Medical Plan Generation},
author={Hsin-Ling Hsu and Cong-Tinh Dao and Luning Wang and Zitao Shuai and Thao Nguyen Minh Phan and Jun-En Ding and Chun-Chieh Liao and Pengfei Hu and Xiaoxue Han and Chih-Ho Hsu and Dongsheng Luo and Wen-Chih Peng and Feng Liu and Fang-Ming Hung and Chenwei Wu},
year={2025},
eprint={2503.17900},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2503.17900}
}