Medicare Paid Add-Ons for Five AI Devices Without Testing Their Benefit. From 2028, Evidence Sets the Price.

TLDR: Medicare paid add-ons to five AI devices without testing clinical benefit, then closed that route from 2028. Medtech teams should plan evidence as the input to payment, not proof after launch.

Medicare paid add-ons to five AI devices for 2027 and kept clinical benefit out of scope

The Centers for Medicare & Medicaid Services (CMS) published its fiscal year (FY) 2027 inpatient prospective payment system (IPPS) final rule¹ (CMS-1849-F) in the Federal Register (FR) on 4 August 2026. It approves new technology add-on payments (NTAP) for 19 technologies, and it is the clearest statement yet of how Medicare payment for artificial intelligence (AI) devices will work from 2028. Sixteen came through the Breakthrough Device “alternative pathway”, and five of those sixteen are AI software. All five appear on the Food and Drug Administration (FDA) list of AI-enabled medical devices², and the rule’s own descriptions of four of them name AI or machine learning.

The rule spells out what CMS examined. On BriefCase-Triage, it wrote that “substantial similarity and substantial clinical improvement are not within the scope of CMS’s evaluation¹ for new technology add-on payment eligibility under the alternative pathway”. On the Ceribell monitor it said the same of clinical performance, and on the Bayesian device the same of performance concerns raised by a commenter. A Breakthrough designation, FDA marketing authorisation, a newness window and a cost test were enough.

The FDA data point the same way. All five devices reached the market through 510(k) clearance, which establishes substantial equivalence to a device already on the market. Kainjoo’s recount of the FDA list, current as of 4 September 2026², finds 1,614 AI-enabled devices, 1,553 of them (96.2%) cleared through 510(k). The clearance route and the add-on both admitted these five products on equivalence, Breakthrough status and cost evidence, before any comparison of patient outcomes against existing care. Medicare is paying for the possibility of benefit, at a price it sets before the benefit is measured.

The alternative pathway closes with 2028 applications, after a two-year grandfather window

The same rule ends the route that paid them. CMS finalised its proposal “to repeal the alternative pathway for new technology add-on payment¹ beginning with applications received for new technology add-on payments for FY 2028″. From then on, every applicant, Breakthrough devices included, must show the technology is new, costly and “an advance that substantially improves” diagnosis or treatment for Medicare beneficiaries.

The repeal comes with a limited grandfather clause. A technology designated as a Breakthrough Device by 30 September 2026, and authorised for that indication by 1 May 2028, may still apply under the alternative pathway for FY2028 and FY2029. A parallel window keeps outpatient device pass-through payment open on the same terms for calendar years (CY) 2028 and 2029. For any AI product designated after 30 September 2026, the substantial clinical improvement test applies from its first application.

The traditional pathway shows what the five were spared. Glytec’s Command Center, an electronic medical record (EMR) integrated glycemic management system, applied under the traditional route for FY2027. CMS judged it “substantially similar to existing software-based EMR-integrated glycemic management systems”¹, including the 2012 Glucommander System, so it failed the newness test and CMS left clinical improvement undecided. Under the alternative pathway, that similarity question was out of scope. From FY2028 every AI applicant faces it, followed by the substantial clinical improvement test.

CMS gave its reasons in the same rule. Citing a commenter, it wrote that it may be relevant to consider whether the alternative pathway “may reduce incentives to generate evidence of improved outcomes¹ for Medicare beneficiaries”, and added that FDA marketing authorisation alone “is often insufficient to support Medicare decision making”. CMS estimates that the 16 alternative-pathway approvals will cost about $418 million in FY2027. The five AI devices are the software slice of a route whose evidence standard CMS has now judged too low.

Outpatient and physician rules hold software payment level while CMS tightens price control

The two calendar-year 2027 rules apply the same discipline to software paid outside the hospital stay. Both were still proposals when this analysis closed on 14 October 2026, with final rules due around 1 November.

The CY2027 outpatient prospective payment system (OPPS) proposed rule³ (CMS-1850-P, 91 FR 41918) renames “Software as a Service” as Software as a Medical Service (SaMS). It would designate 36 HCPCS (Healthcare Common Procedure Coding System) codes as SaMS and move 21 of them from clinical ambulatory payment classifications (APCs) into New Technology APCs, “maintaining approximate payment rate continuity with CY 2026 payment”, under a new status indicator, “O1”. CMS calls this an interim policy. Its stated concern is that the current cost-based system “may not effectively constrain pricing” for technologies whose value sits in proprietary algorithms with scalable, non-material costs.

The CY2027 physician fee schedule proposed rule⁴ (CMS-1848-P) adopts the same SaMS term. It proposes moving ten algorithmic analyses of laboratory data off the Clinical Laboratory Fee Schedule (CLFS) and onto contractor pricing under the physician fee schedule (PFS). CMS warns that paying them through CLFS methods “may create significant vulnerabilities for the Medicare program”. The same rule proposes a Merit-based Incentive Payment System (MIPS) improvement activity, “Clinician Use of Artificial Intelligence (AI) to Improve Patient Care”, which rewards governed, transparent use of AI in clinical workflows.

For an AI company, the reading is level payment today and a tighter price model tomorrow. The 21 outpatient codes keep roughly their 2026 rates under the interim New Technology APC policy, and the ten laboratory algorithms move to contractor pricing, while CMS gathers the cost and use data to set SaMS rates on its own terms.

Fast-track coverage now starts with an outcome study agreed before launch

Coverage now asks for outcome evidence up front. On 23 April 2026, CMS and the FDA announced the RAPID coverage pathway⁵ (Regulatory Alignment for Predictable and Immediate Device). It is open to Class II and Class III Breakthrough Devices. Eligible devices must be the subject of an investigational device exemption (IDE) study that enrols Medicare beneficiaries and measures clinical health outcomes agreed by the FDA and CMS. In return, CMS issues a proposed national coverage determination (NCD) on the day of FDA authorisation, which could bring national coverage as soon as two months after market authorisation.

The RAPID procedural notice⁶ (CMS-3487-NC, 91 FR 51710) was published on 11 August 2026; its comment period closed on 13 October. The pathway takes effect when CMS publishes the final notice. RAPID is written for Breakthrough Devices of every kind, AI included. Its logic applies directly to AI makers: the fast route to Medicare coverage starts with an outcome study designed together with the payer, before launch.

For an AI sepsis or triage tool, the outcome study differs from the accuracy study that supports a 510(k). Clearance rests on how well the algorithm flags a finding; an outcome study measures what happens to the patient after the flag, in a Medicare-age population and against usual care. That calls for a comparator arm, an endpoint a payer recognises and a sample drawn from beneficiaries. Teams that agree those three elements with the FDA and CMS before the pivotal study reduce the risk of running a second study after launch.

Switzerland reached a similar sequence by another road. The Future of Health Grant, backed by the insurer CSS and the EPFL Innovation Park, launched three hospital pilots⁷ in September 2026 that measure impact on one shared protocol for hospitals, insurers and the canton. Kainjoo Life covered how that design puts the insurer inside each pilot from the first protocol. Medicare’s RAPID pathway applies the same logic at national scale: the payer helps define the evidence before the product launches.

Uptake bought by an add-on is a payment signal, and launch models should book it as one

An add-on payment is reimbursement on top of the standard hospital payment, so it lowers what a hospital pays to try a technology. Hospitals that adopt under that subsidy show that the price works. That is a separate finding from showing that the clinical value works. A launch plan that books add-on-era volume as product-market fit carries a demand line that ends when the add-on period does.

The final rule’s own numbers show the size of that step. CMS puts the average cost of the Ceribell monitor at $3,340 per case, with an add-on of up to $2,171¹. While the add-on runs, a hospital’s net outlay on a qualifying case can fall to about $1,169. When the newness period ends, the hospital carries the full $3,340 inside the standard diagnosis-related group payment, close to three times the subsidised figure. A demand model calibrated on the subsidised years overstates the volume a hospital will buy at full price, unless the product has by then shown an outcome benefit that justifies the cost. The same arithmetic applies, at smaller amounts, to every device in Exhibit 1.

The evidence base gives that risk a number. A cross-sectional study in JAMA Health Forum found 60 of 950 AI-enabled devices⁸ (6.3%) linked to 182 recall events, 43.4% of them within a year of clearance. Devices without reported clinical validation were recalled more often, and lack of clinical validation was independently associated with recall (odds ratio 2.8). For a product whose uptake rests on an add-on, an early recall can end the subsidised period before any outcome evidence exists to carry the price.

A five-question payment-evidence check turns this into planning inputs for each AI product:

1. Which payment route carries it today? NTAP, outpatient pass-through, a New Technology APC, PFS contractor pricing, or the standard payment alone.
2. Does that route test clinical improvement? The traditional NTAP pathway tests substantial clinical improvement and RAPID asks for agreed outcome data; the alternative pathway rests on Breakthrough status and cost alone.
3. When does it sunset? The newness period, the FY2029 grandfather limit, and the interim SaMS policy each carry a date.
4. What outcome study would RAPID or a traditional NTAP application ask for? Name the endpoint, the Medicare population and the comparator now.
5. What does uptake look like without the add-on? That figure is the demand estimate the business case should rest on.

Three moves for medtech teams on Monday

First, list every AI product’s Medicare payment route and the date it ends, including whether the product qualifies for the FY2028–FY2029 grandfather window. Second, rebuild the demand model in two versions, with and without the add-on, and plan commercial targets on the second. Third, start the outcome-study design the 2028 rules will require. Medical affairs owns the endpoint and the Medicare-age population; market access owns the payer conversation that RAPID now makes possible before launch.

Under the 2026 rules, evidence becomes the input that sets Medicare payment. Teams that design the outcome study before launch arrive at the 2028 test with the data it asks for. For help mapping a portfolio against these routes, talk to Kainjoo Life.

This article has been reviewed by Haider Alleg, Managing Partner at Kainjoo SA.

References

1. Centers for Medicare & Medicaid Services. Medicare Program; Hospital Inpatient Prospective Payment Systems for Acute Care Hospitals and the Long-Term Care Hospital Prospective Payment System and Policy Changes and Fiscal Year 2027 Rates (CMS-1849-F). Federal Register 91(148), 4 August 2026. https://www.govinfo.gov/content/pkg/FR-2026-08-04/pdf/2026-15833.pdf
2. US Food and Drug Administration. Artificial Intelligence-Enabled Medical Devices (list, content current as of 4 September 2026). https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices
3. Centers for Medicare & Medicaid Services. Medicare Program: Hospital Outpatient Prospective Payment and Ambulatory Surgical Center Payment Systems, CY 2027 proposed rule (CMS-1850-P). Federal Register 91 FR 41918, 7 July 2026. https://www.federalregister.gov/documents/2026/07/07/2026-13656/
4. Centers for Medicare & Medicaid Services. CY 2027 Payment Policies under the Physician Fee Schedule and Other Changes to Part B Payment and Coverage Policies, proposed rule (CMS-1848-P), 16 July 2026. https://public-inspection.federalregister.gov/2026-14327.pdf
5. US Food and Drug Administration. CMS and FDA Announce RAPID Coverage Pathway to Accelerate Patient Access to Life-Changing Medical Devices, 23 April 2026. https://www.fda.gov/news-events/press-announcements/cms-and-fda-announce-rapid-coverage-pathway-accelerate-patient-access-life-changing-medical-devices
6. Centers for Medicare & Medicaid Services. Medicare Program; Regulatory Alignment for Predictable and Immediate Device (RAPID) Coverage Pathway (CMS-3487-NC). Federal Register 91 FR 51710, 11 August 2026. https://www.federalregister.gov/documents/2026/08/11/2026-16368/
7. Future of Health Grant. The Future of Health Grant unveils its Pilot Factory, 3 hospital pilots and 6 new start-ups selected from 606 applicants, September 2026. https://future-of-health.org/fohg-pilot-factory-3-hospital-pilots-6-new-startups/
8. Lee B, Kramer P, Sandri S, et al. Early Recalls and Clinical Validation Gaps in Artificial Intelligence–Enabled Medical Devices. JAMA Health Forum, 2025;6(8):e253172. doi:10.1001/jamahealthforum.2025.3172. https://jamanetwork.com/journals/jama-health-forum/fullarticle/2837802