John Moehrke Consent About AI
0.1.0 - ci-build International flag

John Moehrke Consent About AI, published by John Moehrke (Moehrke Research LLC). This guide is not an authorized publication; it is the continuous build for version 0.1.0 built by the FHIR (HL7® FHIR® Standard) CI Build. This version is based on the current content of https://github.com/JohnMoehrke/ConsentAboutAI/ and changes regularly. See the Directory of published versions

Artifacts Summary

This page provides a list of the FHIR artifacts defined as part of this implementation guide.

Structures: Resource Profiles

These define constraints on FHIR resources for systems conforming to this implementation guide.

Consent with use of the Limits extension

Consent profile that includes the use of the PermissionRuleLimit extension to express limits on the use of data for ML training.

Structures: Extension Definitions

These define constraints on FHIR data types for systems conforming to this implementation guide.

Permission Rule Limit

Extension to express the limit on the use of data for ML training to only de-identified data.

This extension should be used as a ModifierExtension as it is critical to the proper understanding of the consent. If the rule limit is not observed, then the consent is effectively not valid and should not be applied.

Example: Example Instances

These are example instances that show what data produced and consumed by systems conforming with this implementation guide might look like.

AI Device

A Device that indicates an AI system or model. This is not using the AI-Transparency profile in order to keep this IG simple.

Consent for Clinical Decision Support

Consent for using data for clinical decision support purposes.

Consent for Machine Learning Training

Consent for using data for machine learning training purposes.

Consent for Machine Learning Training on De-Identified Data

Consent for using de-identified data for machine learning training purposes.

Consent to allow specific AI for a specific purpose

Consent to allow the use of a specific AI device for a specific purpose. In this case a Device is used to represent the specific AI system or model. In this case the purpose is for CDS use.

Consent to disallow Clinical Decision Support

Consent to disallow using data for clinical decision support purposes.

Consent to disallow Machine Learning Training

Consent to disallow using data for machine learning training purposes.

Consent with all kinds of Provisions

Consent that includes provisions using the various AI mechanisms along with normal TPO.

This consent

  1. permits the patient's care team to have full access
  2. permits TPO purposes to have access to Normal data, but not sensitive dta
  3. permits ML training purposes to have access to Normal de-identified data, but not sensitive data

Given that FHIR R4 Consent has only ONE root level provision, we need to extra deep nesting:

Provisions

  • Permit all the purposes
    • deny ML training for sensitive data
      • permit ML training for de-identified data
    • deny TPO – just so we can skip to a permit
      • permit TPO for normal confidentiality
      • permit TPO for the care team for sensitive data
Dummy Patient example

Dummy patient example for completeness sake. No actual use of this resource other than an example target