Federal Register · STAT News · Sep 11, 2026
If you flag likely cardiovascular disease from non-invasive inputs, this is your precedent now. FDA's final order creates a new Class II category — cardiovascular machine learning-based notification software — off a De Novo request from Viz.ai's Viz HCM, which means the next tool in this space follows a path instead of arguing for one. Read the special controls, not the classification: they attach conditions on training-data diversity, subgroup performance reporting, and overreliance risk. That last one is the interesting word — FDA is now naming the clinician's behavior around the alert as something the sponsor has to manage, a labeling and human-factors obligation, not a model-performance one. Same week, the agency cleared the "Queen of Hearts" EKG algorithm for occlusion MI the standard read misses. Both products are notification software. Neither diagnoses. The regulatory center of gravity in cardiac AI has moved to who gets told, how fast, and what happens when they stop checking.
Lauren Curtis, Hey Health Tech · Sep 11, 2026
ChatGPT for Healthcare can now connect directly to Epic, letting an authorized clinician synthesize notes, labs, meds and specialist documents inside the workflow — across a base north of 325 million patient records. The consolidation nobody is tracking here isn't M&A, it's platforms absorbing features. Epic says more than 85% of its customers now use some form of Epic AI and still doesn't grow by acquisition — it doesn't have to. Meanwhile health IT and digital health is the largest healthcare M&A segment by disclosed value in the first half of 2026, so the money is moving too. The question for anything sitting between a clinician and a chart is which of three moats you actually hold: genuine uniqueness, independent distribution, or integration deep enough that ripping you out costs more than the feature is worth. Most point solutions have been quietly relying on a fourth — that nobody big had gotten around to it yet.
80/20Summarization was never the product. It was the demo. The durable version of your company is whatever you do with the write-back, the accountability, or the workflow the platform doesn't want to own — and you have roughly one funding cycle to find out which one that is.
Ruth Reader, Second Opinion · Sep 10, 2026
Unresolved, and it just got more unresolved. ONC's HTI-5 final rule is landing without the proposed information-blocking fixes — including the provisions that would have narrowed the exemptions for withholding records and extended the framework to autonomous and agentic systems requesting them. Epic argued ONC exceeded its statutory authority, and the AHA opposed the changes too. Word is the rule finalizes in the next couple of weeks. Design as if your agent has no standing of its own: every request rides on a named human's authorization, logged, with an audit trail that proves it. HTI-6 now carries all the pressure, which means this question gets answered on somebody else's timeline.
HatersFive rules into the Cures Act and the agentic-access question gets punted to rule six. The vendors who complained about ambiguity are the same ones who just won by preserving it. Don't build a roadmap assumption on a regulation that's been "a couple weeks out" before.
Hospitalogy · Sep 10, 2026
Because it's the only line item where the waste is measured in physician-hours and everybody already agrees it's broken — physicians report roughly 13 hours a week on prior auth, and smaller and rural practices absorb the worst of it. Delaware's plan is a $50M bet, the only Rural Health Transformation application that explicitly targets prior auth, and the design choice is the part worth stealing: N-by-1 instead of N-by-N. Providers keep clinical records, plans keep coverage decisions, and the shared layer carries only the request, the evidence, the status and the response — deliberately not a data repository. Their stated fear is the right one: federal prior-auth API rules hit Medicare Advantage, Medicaid, CHIP and Marketplace plans in January 2027, and if every plan connects separately to every provider, the industry rebuilds the fax tangle in JSON and calls it interoperability.
BuilderThe thin-waist design is doing the work: identity, routing, status, receipts, nothing else. If your prior-auth product's value proposition is the connections you've painstakingly built one at a time, a neutral shared layer isn't a partner — it's the thing that makes your moat a commodity. Better to be an endpoint worth routing to.
Health API Guy, Brendan Keeler · Sep 11, 2026
His premise is that ubiquity beats elegance every time — a technically superior network reaching 70% of counterparties loses to an ugly one reaching 99%. AI has made the old analog rails cheaper to use (voice agents sitting on hold, agents reading faxes) without producing a single new ubiquitous digital one. The failure mode he names is cacophony: when both sides automate, outbound agents meet inbound voice AI, volume explodes, and humans get crowded out of their own network. His proposal is a deliberately dumb layer — identity, permissions, routing, trust, receipts — with all the intelligence at the endpoints, and he thinks only the federal government has the leverage to force adoption fast enough, because that's the only thing that's ever worked in this industry.
BuilderIf you're building an agent that talks to other organizations, you're currently paying the integration tax and calling it a moat. Read this as a warning about what your cost structure looks like the day a neutral routing layer exists — and design so your value lives at the endpoint, not in the plumbing.
