From Passive to Proactive: A Multi-Agent System with Dynamic Task Orchestration for Intelligent Medical Pre-Consultation
🤖 AI Summary
To address the limitations of passive interaction, weak contextual management, and insufficient support for high-concurrency scenarios in AI-powered pre-consultation systems, this paper proposes a centralized-control-based hierarchical architecture comprising eight autonomous agents, enabling proactive task orchestration and cross-model collaboration. The system dynamically schedules 13 fine-grained subtasks across four core clinical phases: triage, current illness history collection, past medical history acquisition, and chief complaint generation—fully supporting on-premises deployment and end-to-end privacy preservation. By integrating heterogeneous large language models—including GPT-OSS 20B, Qwen3-8B, and Phi4-14B—it establishes the first “task-driven” pre-consultation paradigm. Evaluated on 1,372 real-world electronic health records, the system achieves 87.0% primary specialty triage accuracy, 80.5% secondary specialty classification accuracy, 98.2% task completion rate, an average physician rating of 4.42/5.0, and ≤17 dialogue turns per session—demonstrating substantial improvements in operational efficiency and clinical applicability.
Application Category
📝 Abstract
Global healthcare systems face critical challenges from increasing patient volumes and limited consultation times, with primary care visits averaging under 5 minutes in many countries. While pre-consultation processes encompassing triage and structured history-taking offer potential solutions, they remain limited by passive interaction paradigms and context management challenges in existing AI systems. This study introduces a hierarchical multi-agent framework that transforms passive medical AI systems into proactive inquiry agents through autonomous task orchestration. We developed an eight-agent architecture with centralized control mechanisms that decomposes pre-consultation into four primary tasks: Triage ($T_1$), History of Present Illness collection ($T_2$), Past History collection ($T_3$), and Chief Complaint generation ($T_4$), with $T_1$--$T_3$ further divided into 13 domain-specific subtasks. Evaluated on 1,372 validated electronic health records from a Chinese medical platform across multiple foundation models (GPT-OSS 20B, Qwen3-8B, Phi4-14B), the framework achieved 87.0% accuracy for primary department triage and 80.5% for secondary department classification, with task completion rates reaching 98.2% using agent-driven scheduling versus 93.1% with sequential processing. Clinical quality scores from 18 physicians averaged 4.56 for Chief Complaints, 4.48 for History of Present Illness, and 4.69 for Past History on a 5-point scale, with consultations completed within 12.7 rounds for $T_2$ and 16.9 rounds for $T_3$. The model-agnostic architecture maintained high performance across different foundation models while preserving data privacy through local deployment, demonstrating the potential for autonomous AI systems to enhance pre-consultation efficiency and quality in clinical settings.
Problem
Research questions and friction points this paper is trying to address.
Transforming passive medical AI into proactive inquiry systems
Addressing limited consultation times through dynamic task orchestration
Enhancing pre-consultation efficiency with multi-agent clinical framework
Innovation
Methods, ideas, or system contributions that make the work stand out.
Hierarchical multi-agent framework for proactive medical inquiry
Centralized control orchestrates four primary pre-consultation tasks
Model-agnostic architecture maintains performance across foundation models
🔎 Similar Papers
No similar papers found.
