Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ColaCare: Enhancing Electronic Health Record Modeling through Large Language Model-Driven Multi-Agent Collaboration

Zixiang Wang, Yinghao Zhu, Huiya Zhao, Xiaochen Zheng, Dehao Sui, Tianlong Wang, Wen Tang, Yasha Wang, Ewen Harrison, Chengwei Pan, Junyi Gao, Liantao Ma

TL;DR

ColaCare addresses the challenge of predicting clinically relevant outcomes from structured EHR data by fusing domain-specific expert models with Large Language Model–driven multi-agent collaboration, inspired by multidisciplinary clinical teams. It employs a retrieval-augmented generation module to inject up-to-date medical knowledge from MSD guidelines and orchestrates discussions among DoctorAgent and MetaAgent to produce transparent, evidence-based reports. Across three real-world EHR datasets, ColaCare demonstrates superior predictive performance, particularly in AUPRC, while providing case-study–level interpretability and a practical per-patient cost profile. The work advances clinical decision support by harmonizing structured data modeling, external knowledge, and human-aligned reasoning for personalized precision medicine.

Abstract

We introduce ColaCare, a framework that enhances Electronic Health Record (EHR) modeling through multi-agent collaboration driven by Large Language Models (LLMs). Our approach seamlessly integrates domain-specific expert models with LLMs to bridge the gap between structured EHR data and text-based reasoning. Inspired by the Multidisciplinary Team (MDT) approach used in clinical settings, ColaCare employs two types of agents: DoctorAgents and a MetaAgent, which collaboratively analyze patient data. Expert models process and generate predictions from numerical EHR data, while LLM agents produce reasoning references and decision-making reports within the MDT-driven collaborative consultation framework. The MetaAgent orchestrates the discussion, facilitating consultations and evidence-based debates among DoctorAgents, simulating diverse expertise in clinical decision-making. We additionally incorporate the Merck Manual of Diagnosis and Therapy (MSD) medical guideline within a retrieval-augmented generation (RAG) module for medical evidence support, addressing the challenge of knowledge currency. Extensive experiments conducted on three EHR datasets demonstrate ColaCare's superior performance in clinical mortality outcome and readmission prediction tasks, underscoring its potential to revolutionize clinical decision support systems and advance personalized precision medicine. All code, case studies and a questionnaire are available at the project website: https://colacare.netlify.app.

ColaCare: Enhancing Electronic Health Record Modeling through Large Language Model-Driven Multi-Agent Collaboration

TL;DR

ColaCare addresses the challenge of predicting clinically relevant outcomes from structured EHR data by fusing domain-specific expert models with Large Language Model–driven multi-agent collaboration, inspired by multidisciplinary clinical teams. It employs a retrieval-augmented generation module to inject up-to-date medical knowledge from MSD guidelines and orchestrates discussions among DoctorAgent and MetaAgent to produce transparent, evidence-based reports. Across three real-world EHR datasets, ColaCare demonstrates superior predictive performance, particularly in AUPRC, while providing case-study–level interpretability and a practical per-patient cost profile. The work advances clinical decision support by harmonizing structured data modeling, external knowledge, and human-aligned reasoning for personalized precision medicine.

Abstract

We introduce ColaCare, a framework that enhances Electronic Health Record (EHR) modeling through multi-agent collaboration driven by Large Language Models (LLMs). Our approach seamlessly integrates domain-specific expert models with LLMs to bridge the gap between structured EHR data and text-based reasoning. Inspired by the Multidisciplinary Team (MDT) approach used in clinical settings, ColaCare employs two types of agents: DoctorAgents and a MetaAgent, which collaboratively analyze patient data. Expert models process and generate predictions from numerical EHR data, while LLM agents produce reasoning references and decision-making reports within the MDT-driven collaborative consultation framework. The MetaAgent orchestrates the discussion, facilitating consultations and evidence-based debates among DoctorAgents, simulating diverse expertise in clinical decision-making. We additionally incorporate the Merck Manual of Diagnosis and Therapy (MSD) medical guideline within a retrieval-augmented generation (RAG) module for medical evidence support, addressing the challenge of knowledge currency. Extensive experiments conducted on three EHR datasets demonstrate ColaCare's superior performance in clinical mortality outcome and readmission prediction tasks, underscoring its potential to revolutionize clinical decision support systems and advance personalized precision medicine. All code, case studies and a questionnaire are available at the project website: https://colacare.netlify.app.
Paper Structure (42 sections, 13 equations, 3 figures, 7 tables)

This paper contains 42 sections, 13 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs in Medical Tasks
  4. LLM-Driven Multi-Agent Collaboration in Medical Field
  5. Problem Definition
  6. EHR Datasets Formulation
  7. Predictive Objective Formulation
  8. Methodology
  9. Structured EHR Information Extraction
  10. Multi-Agent Collaborative Consultation
  11. Generation of Initial Reviews
  12. Synthesized Preliminary Report
  13. Collaborative Consultation Process
  14. Multimodal Fusion Network
  15. Experimental Setups
  16. ...and 27 more sections

Figures (3)

  • Figure 1: A three-stage workflow of ColaCare's multi-agent collaborative consultation. (1) multiple DoctorAgents generate initial reviews based on EHR outputs; (2) a MetaAgent synthesizes these reviews into a preliminary report; and (3) an iterative consultation process where DoctorAgents and the MetaAgent collaborate through multiple rounds until reaching consensus, resulting in a final report.
  • Figure 2: Overall architecture of our proposed ColaCare framework. It consists of three main components: (1) a structured EHR information extraction module that processes patient EHR data and generates embeddings through EHR models, (2) a RAG module that processes medical guidelines and retrieves relevant knowledge, and (3) a multi-agent collaborative consultation module where multiple DoctorAgents interact with a MetaAgent to produce a final report. The framework concludes with multimodal fusion of report and EHR embeddings to generate prediction logits.
  • Figure 3: The case of a patient from ESRD dataset, who deceased within one year after the last follow-up visit. Important indicators are shown in dark red color. Diseases are shown in red color. Healthy indicators are shown in green color.