0.1.0 - ci-build
TrustAITransparencyIG - Local Development build (v0.1.0) built by the FHIR (HL7® FHIR® Standard) Build Tools. See the Directory of published versions
The Implementation Guide defines profiles covering the complete lifecycle of AI-supported clinical decision making. They are organized into three architectural layers, plus one resource-independent profile for marking AI-generated content. Each profile page lists the regulatory requirements it implements.
| Layer | Profile | Base Resource |
|---|---|---|
| Static System Context | Trust_AIDevice | Device |
| Static System Context | Trust_AIOrganization | Organization |
| Static System Context | Trust_AIModelCard | DocumentReference |
| AI Output Context | Trust_AIObservation | Observation |
| AI Output Context | Trust_AIProvenance | Provenance |
| AI Output Context | Trust_AIAuditEvent | AuditEvent |
| Clinical Decision Context | Trust_AIHumanOversightAssessment | ArtifactAssessment |
| Clinical Decision Context | Trust_AIPractitionerRole | PractitionerRole |
| Clinical Decision Context | Trust_AIPatientExplanation | Communication |
| AI Involvement Marking | TrustAIData | any Resource |
These profiles describe the AI system, responsible organizations, and technical documentation independently of a specific clinical execution.
Represents the AI system as an identifiable and versioned system component.
It includes metadata such as:
Represents organizations involved in the AI lifecycle, including manufacturers, deployers, and healthcare providers.
It contains the official contact of the legal entity and may also contain:
Represents model-card documentation and technical documentation.
It references the full technical documentation and instructions for use, and contains structured metadata regarding:
These profiles document the AI-generated output, its execution, and its legal traceability.
Represents an AI-generated clinical finding. It references the patient and the AI system that generated it, records the time of generation, and carries a mandatory flag indicating AI origin.
AI outputs that are not Observations can be represented by any FHIR resource that is marked according to TrustAIData, as shown in Use Case 2.
Documents the data lineage and the legal processing context of an AI output:
Documents the technical execution of the AI system: the execution period, the AI system as source and agent, the generated output, any reference databases used, and an optional integrity signature of the log entry.
These profiles document human oversight and patient communication.
Represents the human review of an AI-generated output. It documents the reviewer, the oversight action (validation, override, or correction), the rationale, and any evidence shown to the reviewer.
Represents the reviewing healthcare professional, the organization in which they perform the oversight role, their specialty, and whether they have completed training specific to the AI system.
Documents the patient-facing explanation of an AI-supported clinical decision (AI Act Art. 86): the decision that is explained, the explanation content, when it was provided, and whether the patient was informed about the use of AI.
A resource-independent profile that marks any FHIR resource as AI-involved through a meta.security label from the Trust AI Involvement Code System (ai-generated, ai-assisted, ai-reported, ai-asserted). AI outputs without a dedicated profile, such as a DiagnosticReport, can thereby be documented with the same transparency pattern.