Research map / Large language models / Retrieval and knowledge
Expert-guided Clinical Text Augmentation via Query-Based Model Collaboration
Dongkyu Cho, Miao Zhang, Gregory D Lyng, Rumi Chunara
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 weak-expert (biomedical NER) guidance to LLM augmentation | Qwen-3 (Naive augmentation) · MIMIC-III readmission | accuracy: 0.520 -> 0.599 | Improved |
| add weak-expert guidance to LLM augmentation | BERT (Naive augmentation) · MIMIC-III readmission | accuracy: 0.736 -> 0.757 | Improved |
| add weak-expert guidance to LLM augmentation | Qwen-3 (Naive augmentation) · MIMIC-III length-of-stay | RMSE: 16.357 -> 15.563 | Improved |
| apply naive LLM augmentation instead of none | Qwen-3 · MIMIC-III readmission | accuracy: 0.526 -> 0.520 | Hurt |
| apply CATO augmentation instead of none | Qwen-3 · MIMIC-III length-of-stay | RMSE: 17.835 -> 18.677 | Hurt |
| add weak-expert guidance to naive augmentation | Llama-3.2-3B generalist · synthetic notes token level | PR: 0.51 -> 0.79 | Improved |
| add weak-expert guidance to naive augmentation | Llama-3.2-3B generalist · synthetic notes token level | HR: 0.59 -> 0.33 | Improved |
| use biomedical-ner-all weak expert rather than general-expert (ablation) | augmentation pipeline · synthetic notes token level | PR: 0.53 -> 0.79 | Improved |
| use general-expert NER instead of no expert (ablation) | augmentation pipeline · synthetic notes token level | HR: 0.59 -> 0.50 | Improved |
| use a larger strong generalist (Llama-3.2-3B instead of 1B) (ablation) | augmentation pipeline · synthetic notes | PR: 0.66 -> 0.79 | Improved |
| replace expert collaboration with DPO-trained single Strong Expert (ablation) | Llama-3.2-3B · MIMIC-III readmission | accuracy: 0.583 -> 0.560 | Hurt |
| replace expert collaboration with DPO-trained single Strong Expert (ablation) | Qwen-3 · MIMIC-III readmission | accuracy: 0.599 -> 0.582 | Mixed |
Closest papers by meaning
- MedAlign: A Clinician-Generated Dataset for Instruction Following with Electronic Medical Records (AAAI 2024 · similarity 0.86)
- Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrieval (ICLR 2025 · similarity 0.85)
- MedJourney: Benchmark and Evaluation of Large Language Models over Patient Clinical Journey (NeurIPS 2024 · similarity 0.85)
- Instruction Tuning Large Language Models to Understand Electronic Health Records (NeurIPS 2024 · similarity 0.85)
- MedCalc-Bench: Evaluating Large Language Models for Medical Calculations (NeurIPS 2024 · similarity 0.85)
- Toward Better EHR Reasoning in LLMs: Reinforcement Learning with Expert Attention Guidance (AAAI 2026 · similarity 0.84)
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.
