A multimodal multi-agent LLM framework for identifying key drivers of sleep disorders

Area of Science:

• Sleep Science
• Artificial Intelligence
• Computational Biology

Background:

• Existing sleep disorder research often analyzes lifestyle, behavioral, physiological, and occupational factors in isolation.
• Traditional statistical methods struggle with complex interactions, and many machine learning approaches lack clinical interpretability.

Purpose of the Study:

• To develop an interpretable large language model (LLM)-based multi-agent multimodal framework for analyzing sleep disorders.
• To address limitations in current methods for modeling complex interactions and ensuring interpretability in sleep research.

Main Methods:

• Developed a framework with three specialized agents: Data Analyst, Physiology and Health Analyst, and Validation Analyst.
• Utilized a multi-agent, multimodal approach for comprehensive sleep disorder analysis.
• Applied the framework to both public and synthetic sleep-health datasets.

Main Results:

• High caffeine intake was linked to increased sleep disorder risk.
• Different physical activities showed varied associations with insomnia and sleep apnea.
• Combined effects of occupational context, stress, stimulants, and physiology influence sleep disorder profiles.

Conclusions:

• The LLM framework enhances interpretability and evidence-based reasoning in multimodal sleep analysis.
• The framework aids in reducing unsupported claims in sleep disorder research.
• Cross-dataset analysis served as a robustness check, not external clinical validation.