The Great CMMI Reset
What Was Cut and Why
On March 12, 2025, in one of the new administration’s first concrete actions in federal health policy, the Center for Medicare and Medicaid Innovation announced that it would end four alternative payment models before their originally scheduled termination dates. The announcement was terse. CMS had conducted a “data-driven review” of its model portfolio. Some models would conclude as scheduled. Others would stop by December 31, 2025. The agency estimated the changes would produce $750 million in savings, without specifying its methodology.
The four terminated models were the Primary Care First Model, the Making Care Primary Model, the mandatory ESRD Treatment Choices Model, and the Maryland Total Cost of Care Model. Two previously announced models, the Medicare $2 Drug List and the Accelerating Clinical Evidence initiative, would not move forward at all. A fifth active model, the Integrated Care for Kids program, was flagged for potential reduction or restructuring.
The announcement came less than two months after the Biden administration had ended the Medicare Advantage Value-Based Insurance Design Model in December 2024. That VBID termination, which CMS acknowledged would not be replaced, had already removed the only active CMMI model operating inside Medicare Advantage. The March 12 action cleared four more models from the FFS innovation portfolio in a single announcement.
What makes the terminations analytically significant is not their scale, though eliminating four models simultaneously was atypical. It is what they signal about the theory of value that now governs CMMI’s operations.
The Models and Why Each Fell#
The four terminated models had different histories, participant counts, and performance profiles. They were not cut for identical reasons, and understanding why each was ended clarifies what the administration is actually looking for.
Primary Care First ran from 2021 through its terminated end date of December 31, 2025, one year earlier than planned. It was a voluntary model offered in 26 states and regions designed to test whether advanced primary care delivery could reduce total cost of care. Participating practices received prospective population-based payments adjusted for attributed beneficiary acuity, flat visit fees for face-to-face encounters, and performance-based bonuses tied to quality metrics. The model had 2,175 practices as of December 2023. An interim evaluation published in 2025 found it had not reduced hospitalizations and had increased Medicare expenditures by 1.3 percent. Twenty-seven percent of initial participants had left the model by its third year. PCF became a case study in the voluntary participation problem: the practices that stayed were not generating savings, and the ones that left had self-selected out when outcomes trended unfavorable.
Making Care Primary was the most consequential termination in terms of abruptness. It had launched only in July 2024, as a 10.5-year model intended to give primary care clinicians with varying levels of value-based care experience a pathway toward prospective population-based payments. It enrolled 772 practices across eight states: Colorado, Massachusetts, Minnesota, New Jersey, New York, North Carolina, and Washington. It offered three tracks with graduated risk and upside-only performance incentives. No interim evaluation data existed at the time of termination. MCP ended nine years before its scheduled conclusion, with no results to evaluate, because the administration concluded that a voluntary, upside-only model structured around gradual payment migration could not be expected to generate certifiable savings.
ESRD Treatment Choices had a different character from the other three. It was mandatory. Created through rulemaking in 2020 under President Trump’s executive order on Advancing American Kidney Health, ETC ran from 2021 through its terminated date of December 31, 2025, two years before its originally planned 2027 end date. It required ESRD facilities and managing clinicians in randomly selected geographic areas to modify their payment structures to incentivize home dialysis and kidney transplantation over in-center dialysis. The model’s mandatory nature made it theoretically better positioned to generate certifiable savings than its voluntary counterparts. Its early termination reflects a more specific political dynamic: the executive order that created it was a Trump administration initiative, but the model’s design and operation through 2024 was a Biden era execution, and the current administration’s review concluded it was not performing against its cost-reduction targets at a level that justified continuation.
Maryland Total Cost of Care was the most orderly termination of the four, in part because a successor model was already waiting. Maryland had operated under all-payer rate-setting mechanisms since the early 1970s, with CMS demonstrations providing the federal waiver authority to maintain that system. The TCOC model, which began in January 2019, imposed a per-capita limit on Medicare total cost of care in Maryland and built on the Maryland All-Payer Model that preceded it. It was scheduled to end in December 2026, but CMS and the state were already planning the transition to the Advancing All-Payer Health Equity Approaches and Development Model, which Maryland was set to join in January 2026. The TCOC termination was essentially an administrative step-down ahead of a planned transition rather than a performance-based cut.
The Models Not Pursued#
Two models that had been announced but not launched were formally abandoned in the March 12 announcement.
The Medicare $2 Drug List had been introduced as one of the Biden administration’s responses to drug affordability concerns, proposing to cap beneficiary out-of-pocket costs for generic drugs at $2 per month through a CMMI demonstration. It had been partially positioned as a bridge offering while the VBID termination was being processed. The Trump administration did not pursue it. The decision reflected a broader shift in pharmaceutical policy orientation: the administration’s preferred levers for drug costs are international reference pricing through GLOBE and GUARD and direct price negotiation under the IRA framework, not consumer cost-sharing caps operated through CMMI demonstrations.
The Accelerating Clinical Evidence Model had been designed to expand the use of evidence-based approaches to clinical decision-making in Medicare. It was withdrawn following an executive order rescission. Its termination was a policy alignment decision rather than a performance or cost-savings judgment.
The $750 Million Question#
CMS’s claim that the early terminations would produce $750 million in savings was unaccompanied by any published methodology. The figure has been widely cited and largely accepted in media coverage without examination of how it was derived.
The question is not trivial. Models that are losing money, in the sense of increasing net federal expenditures above what they cost to operate, can generate positive savings when terminated if their continuation costs exceed their benefits. But the $750 million estimate cannot be verified against CMS’s own actuarial projections because those projections have not been made public. It is not clear whether the estimate accounts for the disruption costs to participating providers transitioning out of models, the administrative wind-down costs CMS itself will incur, or the potential reversion of attributed beneficiaries to higher-cost patterns of care in the absence of model infrastructure.
The skepticism embedded in CBO’s prior analysis of CMMI suggests this matters. When the Congressional Budget Office reviewed CMMI’s first decade in September 2023, it found that the center had spent $7.9 billion to operate models and that those models had reduced spending on health care benefits by $2.6 billion, producing a net increase in direct spending of $5.4 billion, or about 0.1 percent of net Medicare spending between 2011 and 2020. CBO had originally projected at the time of the Affordable Care Act’s enactment that CMMI would generate $2.8 billion in net savings over that same period. The actual outcome was nearly $8 billion worse than projected.
Of the 49 models CBO reviewed, six generated statistically significant savings and four were certified for expansion. The certification rate had declined over time rather than accelerating as the center learned from earlier model generations. CBO’s updated forward projections for 2021 through 2030 estimated that CMMI would continue to increase net federal spending by approximately $1.3 billion over that decade, though with considerable uncertainty.
The CBO finding provided the political foundation for the administration’s model review. It established that voluntary CMMI models, taken as a class, had not delivered what the program’s designers projected and that the primary structural explanation was voluntary participation. When providers can select into models they expect to benefit from and exit when circumstances change, the selection effects undermine the counterfactual logic that makes model savings credible. Models that look like they are reducing spending may simply be retaining the providers and patients most likely to generate favorable outcomes regardless of model incentives.
The BPCI-A Precedent#
The prior administration had already confronted this problem through the Bundled Payments for Care Improvement Advanced model, which offers a cautionary parallel. BPCI-A is a voluntary model that allows hospitals, physician groups, and post-acute care facilities to take financial risk for care episodes. MedPAC’s analyses found that BPCI-A participants generated positive gross savings to Medicare but that a significant portion of those savings reflected episode composition effects: participants had been reducing their involvement in complex, expensive episodes while maintaining participation in less complex, lower-risk ones. The net Medicare savings were real but smaller than the gross figures implied, and the model’s design did not prevent the portfolio management behavior that reduced its policy value.
BPCI-A has continued under the current administration, but its experience shapes how CMMI now thinks about voluntary model design. A model that allows participants to manage which patients fall under its scope is not generating the kind of savings that can be certified for expansion without adjusting for the selection effects baked into participation patterns.
What Survived and Why#
The March 12 announcement was explicit that the model terminations did not represent a retreat from value-based care. CMS reaffirmed primary care as a foundational component of the center’s strategy. What the announcement actually ended were models that could not demonstrate savings certifiable enough to justify their continuation under the administration’s framework.
The models that were not terminated on March 12 share a distinct profile. The Medicare Shared Savings Program is not a CMMI model but a permanent statutory program, and its survival is not a CMMI decision. Among active CMMI models, the ACO PC Flex Model, launched in 2025 with 24 participants and approximately 349,000 attributed beneficiaries, was explicitly preserved by the administration as consistent with its strategy, with CMS going so far as to direct PCF participants transitioning out of that model toward the MSSP 2026 application cycle. ACO REACH, the global and professional risk ACO model covering 103 organizations and approximately 2.5 million beneficiaries in 2025, continued without modification. The Kidney Care Choices Model, itself a Trump first-term creation, was left intact through its 2026 end date.
The Home Health Value-Based Purchasing expanded model is the one CMMI demonstration that CMS’s Office of the Actuary certified for expansion, a distinction that makes it uniquely insulated from the current review framework because certification is the standard the administration has established as the threshold for continuation and expansion.
What these surviving models have in common is either a payment design that creates genuine downside risk for participants, a direct lineage to Trump administration policy priorities, or actuarial certification that the model has reduced net federal spending. The three criteria overlap but are not identical, and they collectively define the administration’s theory of what CMMI models should do.
The Savings-Certifiability Threshold#
The Congressional mandate underlying CMMI requires that the Secretary of Health and Human Services determine that any model expanded beyond its initial test phase will reduce net program spending or will improve quality of care without increasing spending. That certification standard, established by section 1115A of the Social Security Act, has been met by a small number of models over the center’s history: the Pioneer ACO model, the Medicare Diabetes Prevention Program, the Maryland All-Payer Model, HHVBP, and a limited number of others.
The new administration’s application of this standard is broader and earlier than prior practice. Rather than certifying successful models for expansion after evaluation, the current framework applies the certification standard as a de facto screen for continuation. Models that cannot plausibly be projected to reach savings certification are being terminated before they consume additional resources to generate that negative conclusion. This is a coherent application of fiscal discipline to the model portfolio, but it forecloses the learning cycle that the original CMMI statute contemplated: test, evaluate, certify, expand, and repeat. Under the current framework, models that do not quickly show savings trajectories are terminated before the evaluation cycle produces definitive results.
The MCP termination is the clearest example of this logic. A model nine months into a ten-year design period cannot have produced evaluation results. It was terminated not because it had been shown to increase spending but because its design characteristics, voluntary participation, upside-only risk, gradual payment migration, parallel the characteristics that have historically predicted against savings certification. The administration effectively concluded that waiting for MCP to demonstrate what PCF had already demonstrated was not worth the cost.
The Mandatory Signal#
The March 12 announcement established the end of a model portfolio era defined by voluntary, upside-only participation structures. It did not announce mandatory replacements. But the May 2025 CMMI Strategic Refresh and the subsequent model announcements through the rest of 2025 made the direction explicit. Every major new CMMI model announced in 2025, including the WISeR prior authorization program, the GLOBE and GUARD mandatory rebate programs, and Geo AHEAD’s geographic risk framework, incorporated mandatory or near-mandatory participation structures at their core.
The policy logic is consistent. If voluntary models cannot be expected to generate certifiable savings because participants self-select against unfavorable financial outcomes, then the only way to produce savings that can survive CBO scrutiny is through mandatory participation that eliminates the selection bias. Mandatory models are not guaranteed to generate savings, but they are the only design architecture under which the counterfactual comparison needed for certification can be credibly constructed.
The ESRD Treatment Choices termination is the one data point that complicates this narrative. ETC was mandatory and was terminated anyway. The administration’s apparent conclusion was not that mandatory designs are insufficient but that ETC specifically was not achieving its kidney health objectives on a timeline and cost trajectory that warranted its continuation through 2027. The model had not been certified for expansion, and its two remaining years were not expected to generate results that would change that assessment.
The VBID Vacuum#
One dimension of the March 12 announcement that deserves more attention than it has received is what was left conspicuously absent: any CMMI activity inside Medicare Advantage.
The VBID termination in December 2024 ended the only active CMMI model that had operated within MA. VBID tested whether MA plans could use value-based benefit designs, waiving cost-sharing for high-value services and adding hospice coverage, to improve quality and reduce overall spending. Its termination removed the one formal experimental channel through which CMMI had been able to test MA plan payment and benefit design innovations.
The March 12 announcement did not include any replacement for VBID or any new MA-specific model. The May 2025 Strategic Refresh confirmed that the current CMMI portfolio is entirely FFS-focused. Every model launched in 2025 operates in Original Medicare fee-for-service: WISeR in FFS prior authorization, ACCESS in FFS chronic condition management, BALANCE in Part D and Medicaid drug coverage, MAHA ELEVATE in FFS lifestyle medicine, LEAD and ASM in FFS ACO and specialty payment.
This is a structural choice with strategic implications. At a moment when MA enrollment has crossed 54 percent of Medicare beneficiaries, CMMI’s innovation portfolio is testing exclusively in the 46 percent of the program that operates under fee-for-service. The absence of any MA-focused CMMI activity means that the most rapidly growing segment of Medicare is outside the center’s current experimental reach.
The practical consequence is that MA plan payment design, risk adjustment policy, and benefit structure are being shaped entirely by the rulemaking process through the annual Advance Notice and Rate Announcement cycle, not by model testing. That produces faster regulatory change but without the evaluation infrastructure that model testing provides.
What the Terminations Signal#
The March 12, 2025 announcement is best understood as the opening move in a larger portfolio reorientation rather than an isolated budget action.
It established that the administration would apply a savings-certifiability screen to the existing model portfolio and remove models that could not meet it. It communicated that voluntary, upside-only participation structures are no longer the default model design framework. It preserved models with mandatory features, actuarial certification, or direct policy lineage to administration priorities. And it signaled that the $750 million in claimed savings from the terminations, whatever its methodology, was less important as a budget number than as a signal of the administration’s willingness to prioritize program fiscal integrity over incumbent model relationships.
The CMMI that emerges from this reset is a narrower, higher-stakes center operating fewer models with larger mandatory footprints, clearer savings requirements, and a design philosophy oriented toward the trust fund clock rather than the innovation learning cycle. Whether this produces the certifiable savings the administration expects depends on whether mandatory model design, combined with the prevention and competition pillars of the May 2025 Strategic Refresh, can accomplish what voluntary model design has not.
The question is not rhetorical. The trust fund implications of getting CMMI right are real. If the program that cost $5.4 billion more than it saved in its first decade can be restructured to certify $10 billion in net savings over its next, that is actuarially meaningful in the context of a 2033 depletion date. If it cannot, the administration will have traded a set of imperfect learning models for a set of mandatory models that produce less savings than their architects projected and generate more provider and beneficiary disruption in the process.
The rest of this series examines each of the 2025 model announcements in turn to assess whether that question can now be answered.
Related Reading#
MCR-00_01 The Trust Fund Clock MCR-05_03 ACOs at Scale: The 2025-2026 Participation Surge and What It Signals MCR-03_05 CMS Under Pressure: Implementation Capacity, Workforce, and the Risk of Regulatory Overload