HL7 Security workgroup Addresses AI Provenance with AIAST Tag

Updated May 31, ⁣2025
​

The HL7 Workgroup plus meeting this week highlights the ⁤growing​ importance of artificial intelligence in healthcare.Recognizing the ⁢need for clarity, the Security Workgroup is emphasizing existing ⁤measures designed​ to support​ AI ‍data handling, especially concerning AI provenance.

A key concern is the ability to identify content created or assisted by AI,‍ including diagnoses, notes, and observations.Tagging this data ensures that downstream users are aware of its origin, informing ⁢their decisions. The ⁢Security Workgroup aims to provide a mechanism for ‍appropriate data tagging,⁣ neither promoting nor discrediting AI, but ensuring openness in AI-assisted healthcare.

To address⁣ this,the HL7 Security Workgroup ‌is promoting the ⁢use of the AIAST tag and​ the Provenance resource to ensure proper AI data management.

Provenance⁤ Tag: Identifying⁢ AI-Generated Data

The AIAST (Artificial Intelligence ​Asserted) tag is a‍ data tag applicable to any data,​ indicating its AI‌ origin. This ‌tag, defined within the‍ HL7 terminology, serves as security provenance metadata. It ‌identifies ⁤data or information objects ‌asserted by AI, ‌such ​as clinical ⁢decision support systems or machine learning⁢ algorithms.

The AIAST code is available for ⁣use ⁣in various standards, including HL7 v2, ⁤CDA, DICOM, ⁢and IHE-XDS. Its portability makes it⁤ a ⁣versatile tool for‌ carrying security tags across⁣ different platforms.

This tag can appear at the top of a FHIR‍ Resource in the .meta.security section,​ like this:

           "resourceType" : "Condition",
           "id" : "1",
           "meta" : {
              "security" : [{
                "system" : "http://terminology.hl7.org/CodeSystem/v3-ObservationValue",
                "code" : "AIAST" }
                ]
              },
           ... other content etc.....}
 

It can also be used with element-level tagging defined⁣ in DS4P inline security labels. For example,a DiagnosticReport might use​ an inline label to tag​ a specific .note element resulting from AI analysis.

Provenance ⁤Resource: providing ⁣Context

The‌ Provenance resource offers ⁤a more detailed ⁣approach when a‍ simple tag is​ insufficient. It leverages the AIAST ‍tag to provide complete information about the AI assertion, detailing the ‌background and context of the AI’s involvement.

this resource can use the target element extension or target path extension to pinpoint specific elements within the target ⁣resource⁤ that originated from AI‍ assertions. ​It can also identify the specific AI algorithm used via a device resource, enabling traceability and accountability.

The Provenance resource can also indicate the data from the patient chart that were ⁣considered by the AI algorithm.

Furthermore, the⁢ Provenance resource can indicate the specific AI algorithm using a Device resource.⁤ this allows users to understand the revision⁢ of the AI that ⁢was used. If a problem⁣ (bias) is later discovered with that version of ​the AI model, all decisions recorded from ‍it can​ be identified.

As with any Provenance, the other elements can be filled out to provide details ‌on when, why, where.

AI’s​ Role in ​Using ⁤Provenance

AI algorithms often analyze patient ⁢records to⁤ generate new⁤ diagnoses or notes. These algorithms ⁣should recognize the AIAST⁤ tag to differentiate between newly entered data and data⁢ derived from previous AI use. This awareness⁤ helps prevent “model collapse” or “feedback loops,” where AI inadvertently reinforces its own outputs. One approach is for AI to disregard data or elements previously authored by AI.

What’s next

As AI continues to‌ evolve⁣ in healthcare, ⁤the HL7 security ⁢Workgroup’s efforts ​to​ establish clear​ provenance standards will become increasingly ⁣vital. The use ⁢of the AIAST tag and Provenance‍ resource will promote ⁢transparency, accountability, and trust⁣ in AI-driven healthcare decisions.