HYPERNOVELTYINSTITUTE

Technical intelligence
for accelerating change.

Search the record

Home/Analysis

Daily Hypernovelty Lead · October 28, 2026

The Model Card Left the Health IT Checklist

The Department of Health and Human Services published a final rule in the Federal Register on October 28 that removes federal certification requirements for some transparency disclosures tied to predictive clinical decision-support tools. The change takes effect December 28, with specified provisions following on January 1, 2027.

Editorial image of an evidence card moving from a checklist to a buyer
AI-generated conceptual editorial illustration. Not documentary evidence.

The scope matters. HTI-5 applies to the Office of the National Coordinator for Health Information Technology’s certification program, which determines whether health IT products can carry ONC certification. It does not govern every clinical AI system or replace other oversight that may apply.

Within the certification program, ONC revised the Decision Support Interventions criterion. The final rule removes requirements to support, access, and modify source attributes for predictive interventions, along with intervention risk-management requirements. Those source attributes were meant to surface information similar to a model card: how an intervention was developed, its intended use and limitations, and how it performed across patient groups.

Vendors can still provide that information. The change is that ONC certification will no longer guarantee a standardized place to find it.

Across 60 certification criteria, ONC removed 14, revised seven, and retained 39 without change. The agency says it found no public evidence that users relied on the predictive-intervention requirements or that the requirements improved patient care. ONC also says clinicians may lack the time or technical background to use the disclosures at the point of care. The larger rule is framed as burden reduction and a reset of the certification program around future FHIR-based application programming interfaces.

Most commenters addressing the removal opposed it. They warned that health systems could inherit more work, transparency could decline, and smaller or rural organizations might struggle to reproduce evidence checks that had been built into certification. ONC disagreed that a significant burden shift was likely. It acknowledged that procurement, governance, and patient-safety teams may value the information, while arguing that buyers can seek it through vendor discussions and other channels.

That disagreement identifies the practical issue. Certification offered a common evidence location. Procurement is a negotiation. Large systems may be able to write disclosure duties into contracts, run their own validation, and monitor model changes. Smaller organizations may receive whatever a vendor chooses to package unless buyers, accreditors, insurers, or another regulator create a replacement norm.

ONC left room for another turn. The agency said it may consider future rulemaking that restores similar source-attribute transparency for evidence-based interventions. Strong voluntary disclosure practices could also preserve much of the information without a federal certification requirement. The next phase will show whether that evidence remains portable or becomes fragmented across contracts, portals, and sales conversations.

Verification bottleneck

A buyer now needs to verify two different things: what a predictive tool does, and whether the evidence supplied before purchase still describes the version running months later. A polished model card is useful only if it names the model version, intended population, known limits, validation date, and material changes. Without a shared update trail, procurement can collect a credible packet that quietly expires while the software continues to change.

The clearest test of ONC’s view will be observable practice. Do vendors keep publishing the removed attributes? Do contracts require notice when a model or training process changes? Can smaller providers obtain the same evidence as major health systems? If the answers are mostly yes, burden reduction may not cost much transparency. If the answers are no, the missing checklist will reappear as duplicated local work.

Opportunities

A compact procurement evidence packet could carry forward the useful parts of the removed requirements. It would request training-data provenance, intended-use boundaries, performance across relevant patient groups, known failure modes, risk controls, and a dated model-change receipt. The packet should travel with the tool after purchase rather than remaining in a sales portal or certification file.

This is also an opening for shared infrastructure. Hospital associations, purchasing groups, accreditors, and health IT cooperatives could maintain a common evidence template and update protocol. That would give smaller organizations a stronger starting point without forcing each buyer to invent its own model-governance system.

Sources

Back to the archive