Building Next-Generation Clinical Workflows with FHIR and AI

Details

This article is a technical deep dive into the 'FHIR Agent Studio' proposed by Sean Connelly, detailing a methodology for building next-generation medical workflows using AI agents. The studio allows users to integrate AI agents into FHIR workflows using reusable building blocks. Twelve specific agents are demonstrated, covering tasks such as 'follow-up tracking for abnormal lab results' and 'drug safety review due to kidney function decline.' The technical architecture defines each agent as a pipeline: 'Trigger → Evidence (FHIR · SQL · Vector Search) → Agent (LLM) → Action (draft FHIR).' This process runs on InterSystems IRIS, allowing developers to inspect and test every step individually. Clinical use cases include surfacing agent proposals as decision support within an EHR (like a patient's chart or medication list), and presenting them in plain language via a dedicated patient app. Crucially, the platform ensures transparency by tracing all execution steps (prompts, model responses, SQL queries). This full traceability is designed to build clinical trust. The system emphasizes safety: every proposed action is a draft requiring human approval, ensuring that AI output remains grounded in retrieved evidence and adheres to strict contracts, thereby addressing the 'black box' problem in healthcare IT.