< + > Secondary AI Land Grab in Healthcare to Commercialize “Governance”

The following is a guest article by Andrew Hines, CTO at Canvas Medical. (Originally posted on his LinkedIn)

There’s a secondary AI land grab happening in health care to commercialize “governance” of AI software. You might be surprised who’s staking out positions in the nascent Health AI Governance market:

1. CHAI (Coalition for Health AI) was founded very recently, is very active, and has a ton of high impact academic and industry membership developing guidelines and certification frameworks. Some of the output is borderline pedantic, but they’re iterating fast and trying to get practical with an “Assurance Labs” program akin to ONC-ATLs.

2. TJC (The Joint Commission) is the hospital accreditation juggernaut and in late 2023 they announced a new certification in relation to AI: Responsible Use of Health Data Certification. Doesn’t seem to get into model measurement, but overlaps upstream with data use.

3. Avanade SAIGE (“Smart AI Governance Engine” from Duke, Microsoft, Accenture) was announced in the Sphere at HLTH this week and so far seems, like the Sphere itself, to be hollow. The promo video is set to an anthemic score and b-roll footage is so mid. But never underestimate Duke or Microsoft, so we’ll see. The focus seems to be registration and control of AI rather than measurement, for now. Makes sense: measurement is way harder.

4. DiME Seal (Digital Medicine Society Seal) is meant to be an attestation of quality rather than a governance model, product, or service. It was announced this month and appears to have 15 products that have gone through it.

5. Aidoc BRIDGE with NVIDIA (Blueprint for Resilient Integration and Deployment of Guided Excellence). I don’t know what to say about this one, still trying to make sense of it after their HLTH announcement.

6. Epic Siesmometer (their one and only open source project) is focused on measurement rather than governance, which makes a lot of sense since the EMR is the cosmic microwave background of health care workflows. Instrumentation must happen in the EMR.

7. HIMSS AMAM (Analytics Maturity Assessment Model) is… I don’t know what. The EMRAM certainly had traction for EMR adoption, but it’s unclear if HIMSS can use their existing distribution to drive adoption of AMAM when it seems a bit off the AI mark.

Valid AI was launched out of UC Davis earlier this year, but seems to be inactive (CORRECTION: Valid AI is active). Surprisingly to some and perhaps expected by others, the NCQA and AMIA don’t seem to have staked out clear positions or products or services just yet.

ASTP/ONC and the FDA are the obvious federal forces governing this territory. It strikes me as tricky, let’s say, to have so much overlap in quality assurance products and services right now when the sheriffs in town don’t quite have things figured out. It’s a sign we’re living in the Wild West of health care AI.

Time will tell where all these “governance” products and services go. My hope is we do not end up in another CQM-like boondoggle where the numbers often mean more to CFOs than patients.