Change proposals in FHIR: human-in-the-loop review for AI-driven writes (www.osbytes.io)

A recent proposal has been made to enhance the FHIR (Fast Healthcare Interoperability Resources) protocol by integrating a human-in-the-loop review process for AI-driven changes to clinical data. This approach recognizes that while AI can efficiently generate updates, such as modifying medication lists, it is crucial to have a qualified human review these changes before they are executed. The proposed system allows AI agents to draft change proposals in the form of transaction Bundles, capturing the intended alterations without immediately applying them. This facilitates a layer of oversight similar to the plan/apply split used in infrastructure engineering, where a plan is validated and approved before execution. The significance of this proposal lies in its potential to improve patient safety and data integrity within health informatics by ensuring that human expertise is leveraged in data alterations. The technical framework employs FHIRPath Patch operations to express updates, enabling a more precise and reviewable format compared to full resource replacements. Enhancements also include the ability to defer reviews through Task objects, streamlined metadata for ease of review, and built-in mechanisms to ensure that changes take current data into account, thereby reducing errors related to concurrent modifications. Overall, this approach represents a meaningful step toward integrating AI in healthcare while maintaining critical human oversight.