ProActIS: Boosting LLMs' Proactive Information Seeking in Dynamic Clinical Workflows with a Lightweight Exam Proposer

ACL ARR 2026 May Submission16388 Authors

Keywords: Medical AI; Proactive Information Seeking; Dynamic Evidence Acquisition; Examination Recommendation

Abstract:

Large language models (LLMs) have shown strong promise as clinical reasoning agents, yet most benchmarks assume complete evidence availability before reasoning begins.
However, real clinical practice unfolds through dynamic clinical workflows with progressive disclosure of patient evidence, requiring clinicians to decide what evidence to acquire next from partial and evolving patient states.
This mismatch poses a key barrier to reliable and wide deployment of LLM-based clinical agents.
To address this gap, we introduce **ProActIS**, a lightweight state-conditioned examination proposer for proactive evidence acquisition. Given an evolving patient state, ProActIS ranks standardized laboratory, microbiology, and radiology examinations, together with *STOP*, through structured patient-state decomposition and action-card encoding.
We construct a leakage-controlled benchmark from MIMIC-IV, comprising 81,960 hospital admissions and 860,099 turn-level patient-state/examination pairs over a 129-examination action bank.
On full-bank next-examination ranking, ProActIS achieves 0.602 Recall@1, 0.886 Recall@5, and 0.725 MRR, substantially outperforming frequency and LLM baselines.
In LLM-assisted dynamic workflows, ProActIS improves evidence coverage and trajectory alignment, highlighting proactive evidence acquisition as a distinct bottleneck and a critical step toward reliable clinical language agents in real world.

Paper Type: Long

Research Area: Clinical and Biomedical Applications

Research Area Keywords: Clinical and Biomedical Applications; Information Retrieval and Text Mining;

Contribution Types: Model analysis & interpretability

Languages Studied: English

EMNLP 2026 AI Reviewing Experiment: yes

Submission Number: 16388