Policy October 21, 2026

Medicare's fast lane for breakthrough devices asks for the evidence radiology AI rarely has

By Aldo Ruffolo, DO, MBA

Medicare has proposed a faster route to national coverage for certain breakthrough devices. The entry ticket is an outcomes study in Medicare patients, designed with FDA and CMS before it starts. Very little of the radiology AI in our reading rooms today was built that way, and I think that is the most useful thing about the proposal.

I read and sign reports every day, and I build Portal at ApertureAI, which routes AI-drafted reports into PACS for a radiologist to review and sign. I'll be upfront about that bias. This essay is about what the proposal says, what the American College of Radiology asked CMS to change, and what the criteria tell us about the evidence radiology AI should be carrying.

What CMS proposed

The pathway is called RAPID, short for Regulatory Alignment for Predictable and Immediate Device coverage. CMS and FDA announced it on April 23, 2026, and CMS published the procedural notice (CMS-3487-NC) in the Federal Register on Aug 11, 2026 [1][2]. The comment period closed Oct 13 [1]. CMS says it will respond to comments in a later final notice [2]. As of today I have not found a date for that.

The mechanics are simple to state. If a device in RAPID completes its study and shows improved health outcomes, CMS would post a tracking sheet and a proposed national coverage determination (NCD) on the same day FDA authorizes the device. There would be a 30-day comment period, and CMS's goal is a final NCD about 60 days after authorization for Class II devices and 90 days for Class III [2]. Depending on the risk of the device and the size of the remaining evidence gaps, the NCD could include coverage with evidence development [1][2].

RAPID builds on earlier attempts at the same problem, including the Transitional Coverage for Emerging Technologies pathway CMS finalized in August 2024 [1].

Who can get in

The eligibility list is where the policy shows its hand. A device has to meet all of the following [1][2]:

  1. FDA Breakthrough designation, plus either a planned De Novo request as a presumptive Class II device enrolled in FDA's Total Product Life Cycle Advisory Program (TAP), or a planned PMA as a Class III device. A footnote also admits 510(k) devices whose primary predicate was authorized through De Novo no more than 18 months before acceptance.
  2. The IDE pre-submission stage. The manufacturer has to plan an IDE study that enrolls Medicare beneficiaries and measures clinical outcomes that FDA considers appropriate and that CMS confirms would show improved health outcomes for that population.
  3. A plausible Medicare benefit category.
  4. No controlling NCD already in place.
  5. Separate payability, meaning the device can be billed to Medicare.
  6. No statutory or regulatory exclusion. In vitro diagnostics are excluded.

The notice is also explicit about who is out: devices that are already market authorized, or already being studied under an IDE, are not appropriate for RAPID [1]. The point is to get CMS into the room before the pivotal study is designed, so the study answers Medicare's question and not only FDA's.

Compare that with what we actually use

FDA keeps a public list of AI-enabled medical devices. I downloaded it on Oct 21. The page says its content is current as of Sep 4, 2026, and the most recent decision in the file is dated June 29, 2026 [3]. The list has 1,614 devices. Radiology is the lead review panel for 1,230 of them. Of those 1,230, 1,212 (98.5%) reached the market through 510(k). Nine came through De Novo and nine through PMA approvals or supplements [3].

510(k) is a reasonable pathway for what it does. It establishes substantial equivalence to a predicate. It was never designed to show that patients do better.

The outcomes evidence is thinner still. In August, a PLOS Digital Health analysis of 1,357 FDA-authorized AI devices through Dec 5, 2025 found that 34 (2.5%) were linked to a registered prospective trial and 3 (0.2%) had evaluated patient-centered outcomes such as mortality, morbidity or readmissions [4]. Radiology accounted for 1,059 of those devices, and fewer than 1% of the radiology tools had a prospective trial [4]. I wrote about that paper in August [5].

Put the two together and the picture is plain. RAPID is not a pathway for the radiology AI we already have. It is a statement about the evidence CMS wants from whatever comes next.

What ACR asked for

ACR filed comments on Oct 12 and published a member update on Oct 15 [6][7]. The letter supports faster access to technologies that improve outcomes for Medicare beneficiaries. It also asks CMS for several safeguards. These are the ones that matter most for imaging AI [6]:

  • Keep the evidence standard. ACR asks CMS to say in the final notice that RAPID does not allow coverage "based solely on FDA authorization, product novelty, or diagnostic performance measures that are not tied to patient outcomes." For radiology technology, the letter says evidence "should not rely solely on technical performance, workflow efficiency, or diagnostic accuracy" unless those measures are linked to meaningful patient outcomes.
  • Make Breakthrough designation less opaque. Breakthrough decisions are generally not public. ACR points out that the designation rests on a "reasonable expectation" from preliminary data and "is not a determination that the device will have clear practical utility, can be readily integrated into real-world clinical workflows, or is reasonable and necessary for Medicare beneficiaries." It asks CMS to publish, for each accepted device, the condition, the intended Medicare population, the outcomes the study will measure and the expected milestones.
  • Don't let RAPID jump the line. CMS proposed to prioritize RAPID NCDs over other NCD requests when resources are tight [1]. ACR objects, arguing that traditional NCDs and reconsiderations often involve technologies with more mature evidence.
  • Settle payment early. ACR asks CMS to explain how the separate-payability requirement applies to devices paid through packaged or bundled payment and to "integral software or digital components." That question matters a great deal for imaging AI, and I'd like to see CMS answer it directly in the final notice.

I agree with all four. The second one deserves more attention than it gets. Breakthrough designation already shows up in press releases as if it were a result. ACR's own description of it, as an early judgment that a device might be better than the current standard of care, is the accurate one.

What outcomes could radiology AI even measure?

This is where I am less certain, and I think the field is too.

Radiology AI mostly changes what a radiologist sees, how fast a study gets read, or what the report says. The patient outcome sits several steps downstream, after a referring clinician acts on the report. ACR's letter lists outcome types that could count for imaging technologies: earlier or more accurate diagnosis that changes management, fewer or less severe complications, reduced need for invasive procedures, better treatment selection, improved survival [6]. Every one of those is measurable. None of them is cheap, and all of them need a study designed for that purpose from the start, with patients who look like Medicare patients. CMS notes that about half of Medicare patients have two or more diseases and that pivotal device studies often exclude people like them [1].

For a report-drafting tool like the one I build, the questions are even less settled. Whether a given reporting tool is a regulated device at all is its own discussion. I don't think that changes the useful part of the RAPID criteria. Even for tools that will never enter the pathway, they are a good checklist. Who is the patient population, what outcome would show benefit, and was that decided before the study began or after the results came in?

What I'd do with this now

If you sit on an imaging AI committee, three things are worth doing before CMS finalizes anything.

  1. When a vendor cites Breakthrough designation, ask what outcome they plan to measure and in what population. A designation is a starting point. Treat it as one.
  2. Look up the tools you already run on FDA's AI-enabled device list [3] and note the pathway. That tells you what kind of evidence was required for marketing and what was not.
  3. If you are asked to evaluate a tool through an outcome lens, start from ACR's list of outcome types [6], not from the vendor's AUC.

Faster coverage for tools that prove they help patients is good policy. Coverage because a tool is new is not. RAPID, as proposed, mostly gets that distinction right. What CMS writes in the final notice will tell us whether it holds.

References

  1. Centers for Medicare & Medicaid Services. Medicare Program; Regulatory Alignment for Predictable and Immediate Device (RAPID) Coverage Pathway. Notice with comment period, CMS-3487-NC. Federal Register. Aug 11, 2026;91:51710-51718. https://www.federalregister.gov/documents/2026/08/11/2026-16368/medicare-program-regulatory-alignment-for-predictable-and-immediate-device-rapid-coverage-pathway
  2. Centers for Medicare & Medicaid Services. Regulatory Alignment for Predictable and Immediate Device Coverage Pathway (CMS-3487-NC) Procedural Notice Inviting Comment. Fact sheet, Aug 7, 2026. https://www.cms.gov/newsroom/fact-sheets/regulatory-alignment-predictable-immediate-device-coverage-pathway-cms-3487-nc-procedural-notice
  3. U.S. Food and Drug Administration. Artificial Intelligence-Enabled Medical Devices (list and CSV download; content current as of Sep 4, 2026; accessed Oct 21, 2026). https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices
  4. Abulibdeh R, Cajas Ordóñez SA, Celi LA, Gorijavolu R, Izath N, Markussen Lunde T. 1,357 AI medical devices cleared, 3 actually tested on patient outcomes. PLOS Digit Health. 2026;5(8):e0001597. https://doi.org/10.1371/journal.pdig.0001597
  5. Ruffolo A. 950 cleared devices, three tested on outcomes. radiologist.com, Aug 20, 2026. https://radiologist.com/insights/950-cleared-zero-outcomes.html
  6. American College of Radiology. Comment letter to CMS re: CMS-3487-NC, RAPID Coverage Pathway. Oct 12, 2026. https://edge.sitecorecloud.io/americancoldf5f-acrorgf92a-productioncb02-3650/media/ACR/Files/Medicare-Medicaid/Coverage/ACR-Comment-CMS-RAPID-Coverage-Pathway.pdf
  7. American College of Radiology. ACR Submits Comments About CMS RAPID Coverage Pathway. Oct 15, 2026. https://www.acr.org/News-and-Publications/2026/comments-on-cms-rapid-coverage-pathway

Related on radiologist.com: 950 cleared devices, three tested on outcomes · Why we chose audit-traceable over foundation model