Provenance and Attribute Groups

During processing, TORCH assigns each resource to one or more attribute groups based on the CRTDL definition and the reference resolution process.

After extraction, TORCH documents the relationship between attribute groups and the resulting resources using FHIR Provenance resources. Each attribute group produces one corresponding Provenance resource.

Purpose

The Provenance resources provide traceability and interpretability of the extraction result:

• They document which attribute group caused which resources to be included in the output.
• They allow consumers to understand the origin of extracted data within the CRTDL definition.
• They support auditing and reproducibility of the extraction process.

Structure

For each attribute group:

• A Provenance resource is created.
• The targets of the Provenance resource are the extracted FHIR resources that were assigned to that attribute group.
• The Provenance resource references the attribute group identifier.

This establishes a many-to-many relationship:

• A resource can be linked to multiple attribute groups.
• An attribute group can reference multiple resources.

Semantics

• Provenance reflects the assignment after extraction.
• It captures why a resource is present in the output, i.e. which attribute group selected it.

Limitations and Considerations

• Resource-level granularity
  Provenance is recorded at the resource level and does not indicate which specific elements within a resource were selected.

• Shared resources across groups
  If a resource is relevant to multiple attribute groups, it will appear in multiple Provenance resources.