Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TeamMedAgents: Enhancing Medical Decision-Making of LLMs Through Structured Teamwork

Pranav Pushkar Mishra, Mohammad Arvan, Mohan Zalake

TL;DR

TeamMedAgents introduces a modular, teamwork-informed multi-agent framework that translates organizational psychology principles (Big Five) into LLM-based medical decision-making. It operationalizes five components—leadership, mutual monitoring, team orientation, shared mental models, and mutual trust—as independently configurable mechanisms within a four-phase deliberation process, supplemented by adaptive component activation. Across eight medical benchmarks (11,545 questions), it achieves 77.63% overall accuracy, with selective activation yielding 2–10 point improvements over strong baselines and a 96.2% convergence rate. The work demonstrates that evidence-based collaboration patterns can enhance safety-critical AI performance, guiding principled design of coordinated AI systems for complex reasoning tasks in medicine and beyond.

Abstract

We present TeamMedAgents, a modular multi-agent framework that systematically translates evidence-based teamwork principles from organizational psychology into large language model collaboration for medical decision-making. Building upon Salas et al.'s "Big Five" teamwork model, we operationalize five core components as independently configurable mechanisms: shared mental models, team leadership, team orientation, trust networks, and mutual monitoring. Our architecture dynamically recruits 2-4 specialist agents and employs structured four-phase deliberation with adaptive component selection. Evaluation across eight medical benchmarks encompassing 11,545 questions demonstrates TeamMedAgents achieves 77.63% overall accuracy (text-based: 81.30%, vision-language: 66.60%). Systematic ablation studies comparing three single-agent baselines (Zero-Shot, Few-Shot, CoT) against individual teamwork components reveal task-specific optimization patterns: shared mental models excel on knowledge tasks, trust mechanisms improve differential diagnosis, while comprehensive integration degrades performance. Adaptive component selection yields 2-10 percentage point improvements over strongest baselines, with 96.2% agent convergence validating structured coordination effectiveness. TeamMedAgents establishes principled methodology for translating human teamwork theory into multi-agent systems, demonstrating that evidence-based collaboration patterns enhance AI performance in safety-critical domains through modular component design and selective activation strategies.

TeamMedAgents: Enhancing Medical Decision-Making of LLMs Through Structured Teamwork

TL;DR

TeamMedAgents introduces a modular, teamwork-informed multi-agent framework that translates organizational psychology principles (Big Five) into LLM-based medical decision-making. It operationalizes five components—leadership, mutual monitoring, team orientation, shared mental models, and mutual trust—as independently configurable mechanisms within a four-phase deliberation process, supplemented by adaptive component activation. Across eight medical benchmarks (11,545 questions), it achieves 77.63% overall accuracy, with selective activation yielding 2–10 point improvements over strong baselines and a 96.2% convergence rate. The work demonstrates that evidence-based collaboration patterns can enhance safety-critical AI performance, guiding principled design of coordinated AI systems for complex reasoning tasks in medicine and beyond.

Abstract

We present TeamMedAgents, a modular multi-agent framework that systematically translates evidence-based teamwork principles from organizational psychology into large language model collaboration for medical decision-making. Building upon Salas et al.'s "Big Five" teamwork model, we operationalize five core components as independently configurable mechanisms: shared mental models, team leadership, team orientation, trust networks, and mutual monitoring. Our architecture dynamically recruits 2-4 specialist agents and employs structured four-phase deliberation with adaptive component selection. Evaluation across eight medical benchmarks encompassing 11,545 questions demonstrates TeamMedAgents achieves 77.63% overall accuracy (text-based: 81.30%, vision-language: 66.60%). Systematic ablation studies comparing three single-agent baselines (Zero-Shot, Few-Shot, CoT) against individual teamwork components reveal task-specific optimization patterns: shared mental models excel on knowledge tasks, trust mechanisms improve differential diagnosis, while comprehensive integration degrades performance. Adaptive component selection yields 2-10 percentage point improvements over strongest baselines, with 96.2% agent convergence validating structured coordination effectiveness. TeamMedAgents establishes principled methodology for translating human teamwork theory into multi-agent systems, demonstrating that evidence-based collaboration patterns enhance AI performance in safety-critical domains through modular component design and selective activation strategies.

Paper Structure

This paper contains 30 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-Agent LLM Systems
  4. Multi-Agent Medical AI
  5. Teamwork Theory and Computational Implementation
  6. Method
  7. Teamwork Component Implementation
  8. Team Leadership
  9. Mutual Performance Monitoring
  10. Team Orientation
  11. Shared Mental Models
  12. Mutual Trust
  13. Implementation Details
  14. Architecture and Model
  15. Parameters and Reproducibility
  16. ...and 15 more sections

Figures (2)

  • Figure 1: TeamMedAgents framework showing four-phase collaborative decision-making: dynamic recruitment assigns task-specific specialists with hierarchical weights, adaptive component selection and Round 2 independent assessment, Round 3 structured discussion with leadership mediation and mutual monitoring, and trust-weighted Borda count aggregation
  • Figure 2: TeamMedAgents algorithm with four-phase architecture. Config $\mathcal{C}$: SMM (Shared Mental Model), L (Leadership), TO (Team Orientation), T (Trust), MM (Mutual Monitoring).