AI-Driven Discovery of Temporal-Demographic Interactions in Emergency Department Care Delivery: AMulti-Agent Collaborative Analysis of Healthcare Equity Patterns

Keywords: A.I agent for Healthcare, LLMs

Abstract: Emergency departments serve as critical healthcare access points, yet persistent disparities in care delivery remain poorly understood, particularly regarding the complex interactions between temporal factors and patient demographics. This study demonstrates the capability of artificial intelligence agents to autonomously conduct comprehensive scientific research investigating these interactions. We employed a novel multi-agent collaborative framework utilizing eight distinct AI models across 58 meticulously documented interactions, analyzing 91,359 patient encounters from four emergency department sites collected between December 31, 2023, and December 30, 2024. The AI-driven analysis revealed significant baseline disparities, with Hispanic/Latino patients experiencing 10.9 minutes longer door-to-provider times and Other/Unknown patients facing 13.0-minute delays compared to White, Non-Hispanic patients. Surprisingly, our models detected a protective effect during high-census periods, where disparities decreased rather than increased, challenging conventional hypotheses about crowding-induced inequities. The interaction coefficients indicated that as ED census increased from the 25th to 85th percentile, length-of-stay disparities decreased by 2.3 minutes for Other/Unknown patients and 6.8 minutes for Hispanic/Latino patients. System-wide 95th percentile wait times reached 93.5 minutes for door-to-provider time and 562 minutes for total length of stay. This study represents a watershed moment in AI-driven scientific discovery, demonstrating that artificial intelligence agents can successfully conduct end-to-end scientific research with minimal human intervention. The discovery of protective effects during high-census periods showcases AI's capability to identify counterintuitive patterns that challenge conventional wisdom. While AI successfully revealed these complex patterns, the persistent baseline disparities underscore the continued need for human action in implementing equitable healthcare solutions.

Submission Number: 246