Research map / Large language models / Agents, code and math
Language Agents for Hypothesis-driven Clinical Decision Making with Reinforcement Learning
David Bani-Harouni, Chantal Pellegrini, Ege Özsoy, Nassir Navab, Matthias Keicher
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 |
|---|---|---|---|
| train LA-CDM with SFT hypothesis loss and GRPO instead of zero-shot | LA-CDM (Qwen2.5-7B-Instruct) · MIMIC-CDM | mean accuracy (%): 64.5 -> 81.3 | Improved |
| train LA-CDM instead of untrained ReAct | LA-CDM · MIMIC-CDM | average test cost ($): 1480.32 -> 1295.61 | Improved |
| add test-cost reward (ablation) | LA-CDM · MIMIC-CDM | average test cost ($): 1427.85 -> 1295.61 | Improved |
| add test-cost reward (ablation) | LA-CDM · MIMIC-CDM | mean accuracy (%): 82.3 -> 81.3 | No clear effect |
| add hypothesis agent (ablation) | LA-CDM decision agent · MIMIC-CDM | mean accuracy (%): 78.5 -> 82.3 | Improved |
| add hypothesis agent (ablation) | LA-CDM decision agent · MIMIC-CDM | average test cost ($): 1410.01 -> 1427.85 | Hurt |
| train hypothesis generation with supervised fine-tuning | hypothesis agent · MIMIC-CDM | hypothesis accuracy (%): 75.7 -> 81.9 | Improved |
| train confidence calibration with GRPO betting reward | hypothesis agent · MIMIC-CDM | ECE: 0.069 -> 0.037 | Improved |
| use all patient data with SFT instead of sequential testing | Qwen2.5-7B-Instruct · MIMIC-CDM | mean accuracy (%): 81.3 -> 92.8 (cost 1295.61 -> 3792.79) | Improved |
| process textual notes and imaging reports instead of tabular-only | LA-CDM vs SM-DDPO · MIMIC-CDM | mean accuracy (%): 37.0 -> 81.3 | Improved |
Closest papers by meaning
- Simulating Viva Voce Examinations to Evaluate Clinical Reasoning in Large Language Models (NeurIPS 2025 · similarity 0.89)
- Large Language Models Are Clinical Reasoners: Reasoning-Aware Diagnosis Framework with Prompt-Generated Rationales (AAAI 2024 · similarity 0.88)
- MedExAgent: Training LLM Agents to Ask, Examine, and Diagnose in Noisy Clinical Environments (NeurIPS 2026 · similarity 0.88)
- Timely Clinical Diagnosis through Active Test Selection (NeurIPS 2025 · similarity 0.87)
- MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology (NeurIPS 2025 · similarity 0.87)
- MDAgents: An Adaptive Collaboration of LLMs for Medical Decision-Making (NeurIPS 2024 · similarity 0.87)
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.
