arXiv cs.AI· Yuxin Xiao, Sheng Zhang, Chandan Singh, Tristan Naumann, Hoifung Poon, Jianfeng Gao, Xiaodong Liu·· 4 小时前AI 评分42
强化学习微调 EHR 基础模型提升临床推理能力
Reinforcement Learning over Patient Trajectories for Clinical Reasoning in EHR Foundation Models
AI 导读
一项研究提出强化学习微调框架，将 EHR 基础模型视为患者轨迹上的生成式策略，把住院再入院等临床预测问题构造为事件条件、时间窗口化的推理任务，并设计时间感知、对 rollout 敏感的奖励。RL 微调在多个任务上持续优于预训练骨干与强基线，使小模型在数据受限时媲美更大的预训练模型，并带来跨任务正向迁移。该论文已被 ML4H 2026 接收。
正文 · 原文
Abstract:Electronic health record (EHR) foundation models trained on longitudinal patient trajectories have demonstrated strong performance across diverse clinical prediction tasks. However, their clinical reasoning capabilities remain constrained by next-token prediction on limited and incomplete EHR data. To address this, we propose a reinforcement learning (RL) fine-tuning framework that treats EHR foundation models as generative policies over patient trajectories. We formulate common clinical prediction problems (e.g., hospital readmission) as event-conditioned, time-windowed reasoning tasks. We then design time-aware, rollout-sensitive rewards to account for finite rollout lengths and temporally inconclusive outcomes. We find that RL fine-tuning consistently improves over pre-trained backbones and strong baselines. Notably, it enables smaller models to match larger pre-trained models in data-limited regimes and induces positive transfer across tasks. Further analysis shows that RL fine-tuned models generate trajectories with stronger structural and semantic alignment to ground truth and greater downstream utility.
| Comments: | Accepted by ML4H 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computers and Society (cs.CY) |
| Cite as: | arXiv:2609.12277 [cs.LG] |
| (or arXiv:2609.12277v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.12277 arXiv-issued DOI via DataCite |
Submission history
From: Yuxin Xiao [view email]
[v1]
Thu, 10 Sep 2026 23:10:38 UTC (202 KB)
[v2]
Tue, 10 Nov 2026 00:24:02 UTC (199 KB)
来源：arXiv cs.AI · arxiv.org