CMS signals intent to revamp how it pays for clinical software and AI

STAT News – Thursday, July 16, 2026
By Katie Palmer

For years, Medicare has been wrestling with how to pay for artificial intelligence and other software-based clinical tools. The Centers for Medicare and Medicaid Services is good at calculating the costs of physical items, from a cotton swab to the wear and tear on a CT scanner. But an algorithm to predict cardiac risk from a CT scan, or an AI-based map to visualize prostate cancer’s spread? Less so.

This month, in its proposed rules for hospital outpatient payments and physician fees for 2027, CMS has signaled that it’s ready to build a more consistent payment structure for clinical software and AI that factors in their impact on patient outcomes. It’s starting — as an interim step, just for 2027 — by proposing a practical change to the way it labels and pays for several clinical software and AI services.

After requesting feedback several times on AI and software payment structures, “it’s really the first time that we’ve heard the agency say that they are planning a different policy moving forward,” said Cybil Roehrenbeck, executive director of industry group the AI Healthcare Coalition. Medicare payment — and its influence on private insurers’ coverage of emerging technologies — has a significant impact on what clinical software and AI gets commercialized and reaches patients.

It’s inherently difficult to put a price on software. There’s the cost of developing the tool itself, the value it holds for patients, and the time it takes for doctors to interpret its results and put them into action. Medicare and regional administrative contractors have sometimes put rates on software based on what practices pay for it — but that can vary widely. Some hospitals pay vendors for every click, while others buy licenses or subscriptions for bulk use, making it hard to nail down a consistent per-use cost and raising concerns about runaway spending.

“Given what they have paid for the few software devices that they have analyzed, and potential utilization, they are questioning what is the correct payment,” said former CMS chief medical officer Lee Fleisher, CEO of Rubrum Advising.

Recategorizing AI and software codes

As it works toward a long-term solution, CMS’ first proposed change is to the monikers for these technologies. Instead of referring to them as “software as a service,” CMS wants to use the more precise term “software as a medical service,” or SaMS.

In its hospital outpatient payment rule, CMS is proposing that 21 of those SaMS codes — including those for algorithms that interpret electrocardiograms for cardiac issues such as pulmonary hypertension and hypertrophic cardiomyopathy, software to predict fracture risk from a CT scan, and retinal imaging AI to detect diabetic retinopathy — move to a new payment track in 2027. It called that change “the first step towards standardization and applying a consistent payment methodology across these services.”

Instead of falling under existing clinical Ambulatory Payment Classification groups, the agency wants to move separately-paid software codes to New Technology APCs. That’s the track CMS uses for — like the name says — new technologies that don’t have much claims data available.

Experts see those interim recategorizations as a practical tool to help CMS more consistently examine software and AI claims without shoehorning them into existing payment models. “New Technology APCs let CMS pay for SaMS without pretending it fits a resource-based model it clearly doesn’t, and it buys them claims data,” said Aditi Joshi, who helps organizations implement digital health tools as CEO of Ardexia.

CMS also wants to create a new status indicator, O1, to identify those software-based services from other new technologies.

Reimbursement codes for software that has previously been bundled into other payments, like computer-aided detection for mammograms, wouldn’t be changed.

The shift should maintain approximate rates from 2026, the agency wrote, with payments for software ranging anywhere from $21 to $900 per use. That includes six software services that have already been assigned a New Technology APC — including one with a rare Category 1 CPT code that analyzes plaque in coronary CT angiograms.

Separate payment, or bundled discounting

“This proposal doesn’t make meaningful changes and definitely doesn’t expand, for example, payment for AI,” said Maya Sandalow, director of health policy at the Bipartisan Policy Center.

Still, the industry is closely watching to see how — and therefore how much — CMS will pay for technology tools with the newly-proposed O1 label.

The agency proposes to pay for that software separately, with no rate changes when it’s billed alongside another service. That approach is supported by many AI vendors and cautioned against by MedPAC, the independent agency that advises Congress on Medicare, because of its potential to drive overuse.

But it’s rare for clinical algorithms to work in isolation, said Joshi: “They’re layered onto an imaging read, a path[ology] workup, something else billable.” And in the proposed outpatient rule, CMS asked directly for comment on whether it should consider reducing payment for AI services when they’re billed alongside something else.

Roehrenbeck, also a partner at law firm Hogan Lovells Cadwalader, said AI Healthcare Coalition supports separate payment for AI, but it would “strongly oppose” an approach that could significantly cut reimbursement for AI when grouped with other procedures.

While the coding changes are a move in the right direction, said Paul Rudolf, an attorney and physician who represents health care clients as they navigate coding, coverage, and payment, “it’s the first step, I think, of several before we find out what’s really going on.”

Stakeholders will want to know what reasoning and criteria were used to develop the proposal, he said, and how CMS will handle new codes as additional algorithmic services enter clinical care.

For now, it’s unclear how the proposed changes would impact individual AI services, or how they might influence CMS’ long-term approaches to AI payment. “CMS has indicated pretty clearly over the last year that they’re very interested in tying payment for these types of services to outcomes,” Sandalow said, “and they’re interested in payment for the services being deflationary.”

To that end, CMS has also proposed to eliminate a reimbursement pathway for devices designated as breakthroughs by the Food and Drug Administration, a growing number of which are AI-enabled.

Breakthrough devices have been able to qualify for supplemental payments for new technologies without demonstrating a substantial clinical improvement over alternatives. Both inpatient and outpatient hospital payment rules now propose to roll back that simplified track.

“CMS is saying, Show us the evidence,” Fleisher said. “Show us that the benefit is not directly accrued to the doctor using it, but really is accrued by having less downstream costs.”

Paying for AI in diagnostic lab tests

In another proposal, mirrored in both the hospital outpatient rule and the physician fee schedule, CMS wants to separate algorithmic analysis from fees for clinical lab services. That would bring payment strategy for secondary analyses like genetic risk algorithms in line with other interpretive AI.

“We believe that whether the SaMS performs algorithmic analyses of an imaging test (e.g. CT scan) or whether it performs an algorithmic analysis on data generated from a laboratory test, all algorithmic analyses should be treated consistently,” the proposed rules read.

The AI payment challenges for CMS in the lab diagnostic arena are similar: There isn’t a lot of transparent data, the agency wrote, to allow it to understand what an algorithmic add-on truly costs the lab that implements it. The physician fee schedule has to remain budget-neutral — if Medicare starts paying more for one service, that money has to come out of something else — but as long as AI tests remain on the clinical laboratory fee schedule, they aren’t subject to that mandate.

“A lot of these AI-only technologies that bigger industry players have put forward have gotten rather handsome reimbursement rates,” said Annie Scrimenti, principal at Shorehaven Consulting and former policy director for the Association for Molecular Pathology. “I think there is a bit of anger in that certain AI-only technologies might have been priced a little too high.”

The hospital outpatient proposal would reassign 10 codes that currently are paid as part of a diagnostic laboratory test to New Technology APCs, with payments ranging from $401 to analyze expression of cancer genes all the way up to $15,000 for an exome sequence analysis to identify disease-causing mutations. Under the proposed physician fee schedule, those same 10 codes would be priced by contractors at the regional, not federal, level.

The proposal raises a number of questions, said Scrimenti, including how AI analyses will be validated if they aren’t required to be run by CLIA-certified labs like the diagnostic tests they are applied to. “Under this proposal, AI-only tests could fall into a no-man’s-land — not clearly CLIA, not clearly FDA, not clearly a medical service.”

Removing the tests from the clinical lab fee schedule would also “dramatically change how they are being priced and then also how they’re going to be coded by the AMA, which in parallel is developing their own AI-only code set,” she said.

Open questions

The American Medical Association maintains the CPT codes used by providers for billing, and its Digital Medicine Coding Committee has discussed a new framework, tentatively titled Clinically Meaningful Algorithmic Analyses, to cover AI. The AMA also recently updated its language to describe different categories of AI — ones that work to assist or augment physicians, and some that work completely autonomously.

But it remains unclear how the AMA’s framework will ultimately code for AI tools, and how that will dovetail with CMS’ goals for outcome-based payment. Neither of this month’s proposed rules includes a concrete vision of how Medicare will pay for software as a medical service going forward, and how those fees may differ depending on where they’re applied.

“This just underscores how complicated it is to try to figure out how to fit AI into our broader payments system,” said Sandalow. “It’s interesting that CMS has noted the challenges with doing so for the last few years, but still isn’t ready to finalize exactly what that should look like.”

What is clear, though, is that “the agency wants health systems to integrate AI into their practice, both for clinical and administrative activities,” she said. In another signal, the proposed physician fee schedule would create a new incentive for doctors to use AI. They could earn credit toward their performance-based Medicare payments score, known as MIPS, by establishing AI governance and monitoring for a wide range of AI tools, including those that summarize medical literature to support clinical decision-making, generate notes, and identify care gaps.

And CMS once again put out a call for thoughts on how it should pay for that whole wide world of AI. It pointed to the growing use of generative AI-based scribes and chatbots, and asked how it might pay for technology-enabled primary care despite the challenges in valuing those tools. It asked pointed questions about how it might capture outcomes in the short and long term to inform payment models.

CMS has made it clear in statements “that we can’t afford to pay for each of these separately,” Fleisher said. In the proposed rule, he sees the agency signaling that doctors could have an opportunity to create — and then share in — savings by implementing new software and AI tools.

“This is the way to both control health care costs, but reward those individuals who come up with novel technology,” said Fleisher, “rather than the traditional way of just paying more for every new widget.”

Access this article at its original source.