AI Transparency on FHIR, published by HL7 International / Electronic Health Records. This guide is not an authorized publication; it is the continuous build for version 1.0.0-current built by the FHIR (HL7® FHIR® Standard) CI Build. This version is based on the current content of https://github.com/HL7/aitransparency-ig/ and changes regularly. See the Directory of published versions

Changes

This page tracks the signficant changes. All changes are tracked in Jira and on GitHub.

1.0.0-alpha

• Rewrite guidance page (Sam, need ticket numbers here)
• FHIR-54421, FHIR-54264, FHIR-54577, FHIR-54259 - Add a profile of DocumentReference for AI Input Prompts as distinct from Model-Cards
• FHIR-54149 - AIinputsCS - CodeSystem title inconsistent with computable name

1.0.0-ballot

Initial Public-Comment

1.0.0-current

• FHIR-57918 Remove experimental provenance codes
• FHIR-54801 Remove code AIRPT
• FHIR-54141 Remove experimental codes
• FHIR-54368 Revise codes in AIdeviceTypeVS - different AI technologies and fleshes out a bit the scenarios and data types.
• FHIR-54244 Include an extension to capture the certainty of an AI inference
• FHIR-57756 More narrative explanation of element level Provenance explaining the extensions
• FHIR-54267 Require DocumentReference, and put futures on independent informative page.
• FHIR-54180 Describe plans for R5/R6
• FHIR-53975 Order of Relationship between Device and Model-Card
• FHIR-54527 Remove use of Device.note to reduce variability on R4 AI device modeling
• FHIR-54734 For better readability, switch section 2 and 3
• FHIR-54371 AI Device representation and its elements
• FHIR-54391 Provide a position on the preferred model card - indifferent
• FHIR-54528 Describe how to exchange AI provenance via RESTful create
• FHIR-54279 and FHIR-54281 Explain how to discover that AI was used, and how to find the provenance or label that indicates this.
• FHIR-54885 Add some clarifying text to the Scope section that detection is not in scope
• FHIR-54893 Clarify AI vs non-AI scope to avoid inconsistent observability implementations.
• FHIR-54726 Scope needs some refinement
• FHIR-54233 Elaborate the Operation section
• FHIR-53973 Clarification on input data
• FHIR-54723 AI device conflicting type binding and property fixed values and extension
• FHIR-54725 Remove unusable AI-Data Profile on Resource as not usable.
• FHIR-54707 Refrain from using Tag, use Label

todo

• more use-cases, and cleanup of what is there
• process use-case is expressing how a human-in-the-loop works.
• FHIR-54280, FHIR-54580 address if labeling is mandatory or optional
• FHIR-54894 refine use of "observability" and "transparency" in the document
• FHIR-54757 clarify combining AIdevice profile with other device profiles such as a smart watch.
• FHIR-54583 clarify use of AI Transparency in post market surveillance
• clarify is AIAST the only code, so a valueSet is not needed. (add counter example using CDS for non-AI algorithm)
• define transparency creation actor and transparency use actor
• assure use of normative words, possibly in requirements resource
• overall readability and flow of the document
• need page 1 graphic?
• plain language summary
• FMM increase? – Malte Sussdorff (implemented it in 2 weeks)
• didn't address cascading AI models used.