Tag line: Can the long-maligned CMMI redeem itself under Trump? Warshawsky argues that only mandatory models and smart use of AI can finally deliver real savings.
###
The Center for Innovation (CMMI) at CMS has been kicked around ever since it was created by the Affordable Care Act in 2010. It has the authority to waive any section of Medicare law for the purpose of running a demonstration or model - and there is no explicit limit on how big or how long such a demonstration can be.
The current head of CMMI is Abe Sutton, who worked at McKinsey, had some accomplishments in the Trump I administration, and snuck in a law degree at Harvard during the four Biden years.
I track articles on how to improve CMMI, and I may be a little jaded on the topic. But this week, I found one that turned out to be quite interesting and insightful. The author is Mark Warshawsky of the American Enterprise Institute.
Find the whole AEI article here:
I disagree with one general point of his - "eliminate reimbursement of AI." His point is, he wants hospitals directed to use of AI to reduce health system costs, rather than focus on AI they would get paid for. But creation of some products and I.P. in AI will surely require investment and funding. He quotes and tracks back to a 2022 article on paying for AI, by Parikh and Heimchen, which has been cited 90 times.
Basically, I think he could retitle CMMI as CMPE, "The Center for Medicare Productivity and Efficiency."
Below, a short AI summary.
###
AI CORNER
##
Summary (Chat GP5)
In this American Enterprise article, economist Mark J. Warshawsky argues that the Center for Medicare and Medicaid Innovation (CMMI)—created under the Affordable Care Act to test cost-saving health-care payment models—has failed to deliver on its promise. Despite $10 billion per decade in funding and 49 pilot models launched since 2011, only a handful showed positive results, and net savings have turned negative.
Warshawsky traces the failure to voluntary participation, conflicting incentives, and weak evaluation methods.
He then asks why the second Trump administration has not dismantled CMMI, given its ACA origins and association with Biden-era equity initiatives. Instead, the administration has repurposed CMMI, terminating some models and launching new mandatory pilots—notably, a specialty-care model with two-sided financial risk and a prior-authorization model for Medicare fee-for-service patients.
A new CMMI strategic plan emphasizes prevention, patient empowerment, competition, and fiscal discipline. Warshawsky welcomes this reset but says it will not transform the sector without a broader push for productivity and efficiency, particularly through artificial intelligence. He describes how AI could streamline diagnostics, documentation, and treatment planning, lowering costs while improving outcomes.
The piece concludes with six proposed payment reforms to integrate AI responsibly—ranging from outcome-based rewards to “negative incentives” for providers who ignore proven AI tools. His central claim: CMMI could yet redeem itself if it becomes the vehicle for technology-driven efficiency, rather than more bureaucratic experimentation.
##
Sidebar - Parikh 2022 Paper - AI Summary
Writing in npj Digital Medicine (2022), Ravi B. Parikh and Lorens A. Helmchen warned that the emerging per-use reimbursement model for AI tools (e.g., CMS’s CPT codes and NTAP add-ons for Viz LVO and IDX-DR) risks overuse, waste, and fraud, much as fee-for-service imaging once did.
They reviewed the rapid FDA clearance of more than 200 AI medical devices and the first wave of CMS reimbursement codes, then proposed five alternative frameworks for paying for AI that could promote value rather than volume:
-
Forgo separate reimbursement when AI savings or revenue gains already accrue to providers.
-
Incentivize outcomes, tying payment to measurable improvements in quality or efficiency (e.g., stroke outcomes for imaging AI).
-
Advance market commitments—“X Prize”-style competitions awarding lump-sum payments for AI that solves major care challenges and releases its code.
-
Time-limited add-on payments (2–3 years) for novel AI tools, after which costs are folded into bundled or episode-based rates.
-
Reward interoperability and bias mitigation, paying more for AI validated across diverse populations and settings.
Their central argument: AI’s scalability and automation require new payment logic. Per-use billing overvalues machine work and can inflate costs; value-based and time-limited models are better aligned with efficiency, equity, and patient benefit.
Warchawsky adds his #6, proactively penalize providers who ignore the use of efficiency-increasing AI.
