You put what, where? What FHIR should be doing about AI/ML
summary
Proposes a concrete draft: a white paper and an Algorithmics-on-FHIR implementation guide that define new resources to capture ML outputs, notably an inferredEntity resource and inferredRelationship, plus using a meta tag for provenance and confidence so systems carry confidence scores when translating to Condition or DiagnosticReport. Recommends model and data cards for transparency, points out an R4/R4B profile break that needs fixing now, and warns of patient-facing harms (eg pregnancy or cancer inferences) if machine vs human data aren’t separated.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 2:44Overlapping Resources
- 3:44What could happen
- 4:46Explainability Transparency
- 6:51Data and Model Cards
- 8:33Algorithms on Fire Implementation Guide
- 10:04Leverage AI
- 10:19New bundle types
- 10:46Universally useful data
- 11:21New resource types
- 12:22Confidence to condition resource
- 12:50Start the conversation
- 13:13Why this is important
- 13:48Who will be responsible
- 14:36QA
- 20:57Conditions
related talks
39mAI, FHIR & the future of healthcare data
37mAn update on AI and FHIR
20mResponsible AI in Health
36mAvoiding 5 missteps that undermine your AI readiness and success
19mAnalyzing FHIR Data Using ML and Analytical Tools
48mAnalytics and AI on FHIR
18mMedical AI Beyond Prediction: From Normative Learning to Agentic Healthcare AI
45mGenerative AI in medicine: Improving patient & provider experiences