Research map / Large language models / Language, safety and interpretability
Critic–Adviser–Reviser Cyclic Refinement: Towards High-Quality EMR Corpus Generation with LLMs
Chen Ning, Xien Liu, Chenwei Yan, Xiao Zhang, Xinxin You, Yuxuan Zhou, Xiangling Fu, Ji Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Critic-Adviser-Reviser staged cyclic refinement to direct generation | Qwen2.5-7B-Instruct EMR generator · EMR quality | content completeness (%): 77.1 -> 91.2 | Improved |
| add stage-wise refinement instead of generic all-in-one Self-Refine | EMR generator · EMR quality | knowledge coverage (%): 78.0 -> 94.1 | Improved |
| use LLM-CARe synthetic EMRs instead of MedSyn (real EMR text in context) | downstream classifier · examination recommendation | micro accuracy (%): 82.9 -> 85.3 | Improved |
| add refinement to LLM generation | Qwen2.5-7B-Instruct · diagnosis prediction | micro accuracy (%): 81.8 -> 82.6 | No clear effect |
| add LLM-CARe on other backbones (Llama3.1, Meditron3, R1-Distill) | LLM Direct · EMR quality (avg) | average quality (%): 49.3 -> 77.5 (Llama3.1); 53.9 -> 76.4; 55.5 -> 80.5 | Improved |
| add each of Critic, Adviser and Reviser (ablation) | LLM-CARe · EMR quality and downstream tasks | quality and accuracy | Improved |
| give initial draft generator a real EMR as example (ablation) | LLM-CARe · EMR quality and downstream tasks | quality and accuracy | No clear effect |
| add more than two refinement cycles per stage (ablation) | LLM-CARe · EMR quality and downstream tasks | quality and accuracy | No clear effect |
| use LLM-CARe on MIMIC-IV-Note (different institution) | LLM-CARe vs LLM Direct · diagnosis prediction (MIMIC-IV-Note) | macro accuracy: 0.532 -> 0.590 | Improved |
| use real EMR in-context examples (MedSyn) instead of quality-guided generation | MedSyn · hospital course completeness | completeness | Hurt |
Closest papers by meaning
- Collaborative Synthesis of Patient Records through Multi-Visit Health State Inference (AAAI 2024 · similarity 0.86)
- EHRNoteQA: An LLM Benchmark for Real-World Clinical Practice Using Discharge Summaries (NeurIPS 2024 · similarity 0.85)
- MedAlign: A Clinician-Generated Dataset for Instruction Following with Electronic Medical Records (AAAI 2024 · similarity 0.85)
- Instruction Tuning Large Language Models to Understand Electronic Health Records (NeurIPS 2024 · similarity 0.84)
- Toward Better EHR Reasoning in LLMs: Reinforcement Learning with Expert Attention Guidance (AAAI 2026 · similarity 0.83)
- IGAMT: Privacy-Preserving Electronic Health Record Synthesization with Heterogeneity and Irregularity (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
