Foundation Agents Meet Agentic Deep Research: Evidence-Grounded Clinical Code Forecasting

基础智能体与智能体深度研究的结合：基于证据的临床代码预测

Abstract: Next-encounter ICD forecasting predicts which standardized diagnosis codes will be documented at a future visit from the longitudinal record available beforehand. The task is prospective and multi-label: the target note does not yet exist, and several codes may be correct. Structured EHR foundation models capture recurrence and temporal progression, whereas language foundation models generate flexible diagnostic hypotheses.

摘要： 下一次就诊 ICD（国际疾病分类）预测旨在根据现有的纵向记录，预测未来就诊时将记录哪些标准化的诊断代码。这是一项前瞻性的多标签任务：目标病历记录尚不存在，且可能有多个代码同时正确。结构化电子健康记录（EHR）基础模型能够捕捉疾病的复发和时间演变，而语言基础模型则能生成灵活的诊断假设。

We introduce ICD-Deepresearch, a DeepResearch workflow that composes these predictive foundation models with medical search and ICD dictionaries. Because no source reveals the future code set, research evaluates candidate transitions by linking patient evidence, external clinical relations, and exact code semantics under a fixed top-K budget.

我们引入了 ICD-Deepresearch，这是一种将上述预测性基础模型与医学搜索及 ICD 词典相结合的深度研究工作流。由于没有任何来源能直接揭示未来的代码集，该研究通过在固定的 Top-K 预算下，关联患者证据、外部临床关系和精确的代码语义来评估候选转换。

Candidate Generation uses SparseEHR to produce an EHR Prior that initializes two bounded Research Expansion rounds; an independent GPT-5 Direct Forecast supplies complementary candidates. Final Selection validates, deduplicates, and jointly ranks both paths, after which a separate module writes rationales without changing predictions.

候选生成阶段使用 SparseEHR 生成 EHR 先验，用于初始化两轮受限的研究扩展；独立的 GPT-5 直接预测则提供补充候选。最终选择阶段会对两条路径进行验证、去重和联合排序，随后由一个独立的模块在不改变预测结果的前提下撰写推理依据。

Finally ICD-Deepresearch achieves patient-averaged precision/recall of 24.60/35.09% on MIMIC-III and 25.14/48.32% on MIMIC-IV. Physicians rate 51% and 68% of its retrieved documents useful, compared with 22% and 39% for standalone GPT-5 web search and 32% and 41% for Medical Deep Research. ICD-Deepresearch therefore improves over the registered local comparators while retrieving evidence with higher physician-rated usefulness than the standalone research systems.

最终，ICD-Deepresearch 在 MIMIC-III 数据集上实现了 24.60%/35.09% 的患者平均精确率/召回率，在 MIMIC-IV 上则达到了 25.14%/48.32%。医生认为其检索到的文档中有 51% 和 68% 是有用的，相比之下，独立的 GPT-5 网络搜索仅为 22% 和 39%，而医学深度研究（Medical Deep Research）为 32% 和 41%。因此，ICD-Deepresearch 不仅优于现有的本地对比系统，而且在检索证据的医生评价实用性方面也高于独立的搜索系统。