Research map / Large language models / Retrieval and knowledge
KnowGuard: Knowledge-Driven Abstention for Multi-Round Clinical Reasoning
Xilin Dang, Kexin Chen, Xiaorui Su, Ayush Noori, Iñaki Arango, Lucas Vittor, LONG XINYI, Yuyang Du, Marinka Zitnik, Pheng-Ann Heng
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 |
|---|---|---|---|
| replace LLM self-assessment abstention with knowledge-graph investigation | GPT-4 doctor agent · ioAFRIMEDQA/ioMEDQA/ioCRAFT-MD | accuracy: ioMEDQA 64.23 -> 70.98 vs Scale Rating | Improved |
| add evidence evaluation stage (ranking) to KG retrieval (ablation) | KnowGuard · ioAFRIMEDQA | accuracy: 66.22 -> 72.60 | Improved |
| add evidence evaluation stage (ablation) | KnowGuard · ioAFRIMEDQA | avg. turns: 2.69 -> 7.03 | Mixed |
| add multi-modal KG triplet evidence on top of text evidence (ablation) | KnowGuard · ioMEDQA | accuracy: 64.79 -> 70.66 | Improved |
| add text evidence retrieval (no KG) (ablation) | Scale Rating-style baseline · ioCRAFT-MD | accuracy: 65.40 -> 62.73 | No clear effect |
| add patient population reasoning (PPR) (ablation) | KnowGuard · ioAFRIMEDQA | accuracy: 72.60 -> 73.20 | Improved |
| add patient population reasoning (PPR) (ablation) | KnowGuard · ioMEDQA and ioCRAFT-MD | accuracy: 74.29 -> 74.12; 71.92 -> 71.96 | No clear effect |
| add rationale generation and self-consistency | KnowGuard · ioAFRIMEDQA | accuracy: 68.70 -> 73.20 | Improved |
| add rationale generation and self-consistency | Binary Decision · ioMEDQA | avg. turns: 7.69 -> 13.00 | Hurt |
| add rationale generation and self-consistency | Scale Rating · ioAFRIMEDQA | accuracy: 63.06 -> 61.36 | Hurt |
| use Long Context retrieval instead of KG exploration | Long Context · three datasets | accuracy: ioMEDQA 59.95 vs 70.98 | Hurt |
| vary evidence weights around optimum (e.g. LLM relevance 0.5/0.6/0.7) (ablation) | KnowGuard · ioMEDQA | accuracy: 70.59 / 74.12 / 71.64 | Mixed |
Closest papers by meaning
- Reasoning-Enhanced Healthcare Predictions with Knowledge Graph Community Retrieval (ICLR 2025 · similarity 0.89)
- Simulating Viva Voce Examinations to Evaluate Clinical Reasoning in Large Language Models (NeurIPS 2025 · similarity 0.88)
- ER-Reason: A Benchmark Dataset for LLM Clinical Reasoning in the Emergency Room (NeurIPS 2026 · similarity 0.87)
- KGARevion: An AI Agent for Knowledge-Intensive Biomedical QA (ICLR 2025 · similarity 0.86)
- Optimizing Inference-Time Compute for Medical Reasoning via Uncertainty Quantification (ICML 2026 · similarity 0.86)
- RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis (NeurIPS 2025 · similarity 0.86)
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.
