How Multi-Agent AI Systems Are Improving Clinical Decision Support at BJC Healthcare
BJC Healthcare is deploying coordinated teams of AI agents that go beyond simple chatbots to pull data, triage patients, and nudge clinicians at the right time — including a learning reviewer that continuously adapts from 35,000+ patient records.
BJC Healthcare's medical director Nathan Moore presented at HIMSS26 on a new approach to clinical AI: deploying multiple coordinated AI agents that work together to enhance clinical workflows, moving well beyond the capabilities of standalone chatbots or single-purpose models.
The centerpiece is a 'learning reviewer agent' for mortality risk assessment and advance care planning. Trained on over 35,000 patient records, the system incorporates ongoing clinician feedback rather than remaining static after deployment. It uses supervised learning that continuously adapts to local practice patterns, making its recommendations more relevant over time.
The key architectural insight is that these agents don't just answer questions — they take action in complex workflows. They pull data from multiple sources, triage patients based on risk scores, and nudge clinicians at the right moment with context-appropriate recommendations. This represents a shift from AI as a passive information tool to AI as an active participant in care coordination.
Moore outlined a framework for designing AI agents with memory capabilities, applicable to advance care planning, chronic disease management, and other high-stakes clinical support domains. The approach addresses a persistent criticism of clinical AI: that static models trained on historical data fail to account for the evolving nature of medical practice and local institutional knowledge.