Another Obamacare Failure: Will Trump Fix the Center for Medicare and Medicaid Innovation?

Although the Trump administration is eager to make cuts to excess spending, the wasteful Center for Medicare and Medicaid Innovation (CMMI) remains standing. CMMI was created by the Affordable Care Act (ACA) in 2010, and it aims to “bend the curve” of rapid health care cost growth, in part by reducing the significant spending waste—estimated to be as much as 25 percent—that it is widely believed to exist in the health care sector.

To confront these costs, CMMI conducts time-limited pilot programs, called models, that test innovative ways of paying for or delivering services to beneficiaries and reducing program expenditures while preserving or enhancing the quality of health care. In 2011, CMMI received $10 billion in mandatory funding for model testing. It is scheduled to collect another $10 billion in 2020 and will automatically receive an additional $10 billion in each subsequent decade, although modest reductions occurred through sequestration for 2020 and are scheduled for 2030.

In a considerable expansion of the prior demonstration authority of the Centers for Medicare and Medicaid Services (CMS), which oversees CMMI, the latter agency can require mandatory participation—such as by providers—in its model testing. Furthermore, if the CMS actuary certifies that a model reduces spending without impairing quality—or improves quality without increasing spending—the secretary of health and human services may extend its duration and implement it nationwide.

CMMI’s History of Failure

The authors of the ACA put much stock in the ability of CMMI to reduce program spending. The Congressional Budget Office (CBO) projected that CMMI would reduce federal spending, on net, by $2.8 billion over 2011–20 and by $77.5 billion over 2021–30 as positive model results accumulated and new ones were implemented. Yet despite initiating and evaluating 49 models from 2011 to 2020—nearly all focused on Medicare—the CMS actuary has certified only four for expansion, one of which has subsequently been estimated to produce no savings. As a result, the projected savings from CMMI have entirely disappeared and turned negative. CBO now estimates that CMMI costs, on net, $5.4 billion in its first decade and projects it will cost $1.3 billion in its second decade.

CBO attributes the failure of CMMI to produce savings to the decision to make models voluntary, despite the law allowing mandatory participation. This has led to adverse selection by providers who benefit rather than lose from various bonus payments within the models. CBO and others also point to the proliferation of models, sometimes operating within the same health care systems, which has created conflicting incentives for providers, especially when combined with external changes in payment mechanisms. Others have recommended that CMMI adopt consistent and robust methodologies for measuring quality and spending in its evaluations.

Surprise at Department of Government Efficiency Inaction

Given this long and uninspiring record, it is surprising that the Trump administration and its Department of Government Efficiency (DOGE) did not close CMMI. After all, the center was created as a prominent part of the ACA, a law that the first Trump administration and Republicans in Congress tried to repeal. Moreover, during the Biden administration, CMMI’s remit was expanded to emphasize the improvement of health equity, a type of diversity, equity, and inclusion initiative that the second Trump administration has been quite aggressive in eliminating as discriminatory. The fact that CMMI is embedded in the law would not have been thought to be a deterrent, given that other apparently legally required agencies, such as the US Agency for International Development, Consumer Financial Protection Bureau, Social Security Advisory Board, and US Institute of Peace, have effectively been slated for closure. Instead, the Trump administration has made several recent changes and announced a new strategic plan, suggesting that it plans to use CMMI to advance its goals.

How Donald Trump is Changing CMMI

In March 2025, CMMI announced the termination of four models, modified others, and claimed significant net savings. It also announced the launch of two new models. One is an ambulatory specialty model, designed to improve prevention and upstream management of chronic diseases in order to reduce avoidable hospitalizations and unnecessary procedures. Participation in this five-year test would be mandatory for specialists who commonly treat “traditional” fee-for-service Medicare beneficiaries with heart failure or lower back pain in an outpatient setting in selected regions. These specialists would be subject to a two-sided risk arrangement—causing financial gains and losses of up to 9 percent, with no glide path—based on their performance relative to peers in improving patient health outcomes and coordinating with primary care providers, including through the use of interoperable health record technology.

The second model is a six-year test of more aggressive prior authorization in traditional Medicare. Prior authorization is the practice of conducting preservice reviews of planned medical interventions, including medications and procedures. It is common in commercial health insurance and Medicare Advantage plans, but not in traditional Medicare. There is reason to expect that this model will reduce spending without compromising quality. The model will be tested in six states, using private-sector companies that have successfully implemented technology-enhanced prior authorization, and will focus on 17 nonemergency services believed to be especially susceptible to excessive use. These companies will be compensated based on avoided costs.

The Trump administration also recently released a new strategic plan for CMMI. It justified the agency’s continued existence by stating that CMMI “has made investments in the necessary infrastructure to support broad system reform.” The plan has three pillars:

1. “Promote evidence-based prevention.”
2. “Empower people to achieve their health goals.”
3. “Drive choice and competition.”

The pillars are underpinned by a foundational principle: protecting federal taxpayers. This requires that all alternative payment models for Medicare and Medicaid include downside risk and that providers share in that risk.

The Importance of Increasing Health Care Sector Productivity

Although these changes and plans by CMMI seem broadly sensible, they alone will not be enough to ensure transformational change to the system of macroeconomic significance. There is a pressing need for a more sustained and comprehensive focus on raising productivity in the health care sector as an overriding goal, particularly through the introduction of new technologies—including artificial intelligence, which is in line with directions from the Trump administration in its AI Action Plan. As my coauthors and I have shown elsewhere, the very low rate of productivity improvement in the health care sector, compared with the rest of the economy, raises relative health care prices, lowers effective disposable incomes, worsens inequality, increases budget deficits and government debt, and drives up interest rates.

Department of Health and Human Services Secretary Robert F. Kennedy has stated bluntly that massive and often ineffective health care spending on widespread chronic diseases will bankrupt the US; it is an existential problem. By contrast, an improvement in health sector productivity growth—even to half the rate of the general economy—that is passed on to payers would lower health prices and substantially improve consumer welfare. When combined with increased investment to support new applications of technology, higher health sector productivity would eventually more than halve projected federal deficits and debt and more than triple growth in consumer welfare.

An emphasis on AI in health care to raise productivity growth can also align with the new CMMI goals of promoting prevention, empowering agency and competition, and, above all, protecting federal taxpayers. The current approach legislated in the ACA of simply assuming large generic productivity gains in the sector and paying accordingly does nothing to incentivize their actual realization. It will ultimately lead to rationing of care or lower quality if the pay cuts are not overridden, as eventually occurs through the political process.

Because of the poor incentive structure in health care, owing to extensive government provision and insurance coverage, providers do not compete on price, and the pressure to lower costs through the deployment of capital and technology is weak. Indeed, past experience with the introduction of new health care technologies—such as MRIs and CT scans—that are paid for separately led to overutilization and cost increases as providers purchased new systems and passed the costs to health care plans and consumers. Such overuse may also lead to overdiagnosis and trigger spending on health services of little value.

Despite the many AI innovations that are demonstrably improving administration and quality of care while lowering labor costs, the health care industry has been slower than others to adopt AI. Although algorithmic limitations, data access problems from lack of interoperability, and regulatory barriers contribute to this lag, misaligned incentives are central. CMMI must therefore strategically and efficiently incentivize the introduction of effective AI in health care through its authority to establish and expand payment reform, using reimbursement schedules that build in an expectation of adopting cost-saving techniques, applying pressure for their use, and capturing most of the gains for payers. Although some are touting the potential for autonomous self-service applications of AI to reduce costs through expanded consumer choice and greater market competition, the current incomplete state of AI’s development means these may be controversial—for now.

AI in Health Care

Drug discovery, medical imaging, diagnostics and prognosis, patient monitoring, robotic surgery, personalized treatment design, and administrative operations are among the areas where AI has already been applied in health care. These applications use large language models that interpret oral and written speech, machine learning software that produces decisions or predictions through algorithms and statistical inference, artificial neural networks that rely on deep learning (in some cases beyond human capacity), and generative AI systems like chatbots.

Many of these tools have been shown to reduce costs while also improving the speed of and accuracy of diagnosis. For instance, AI can reduce the time needed to analyze dozens of kidney images for diagnosing polycystic disease from 45 minutes of expert review to mere seconds of inexpensive computer processing. More broadly, AI-enabled interpretation of MRIs and X-rays is being used to prompt timely interventions in strokes, pneumonia, breast cancer, heart attacks, and skin lesions, with fewer false positives and negatives. Another example is ambient listening technology that generates physician’s notes, as well as chatbots that handle scheduling and billing questions.

A Suggestion for CMMI Payment Models

CMS has created a code and some new technology add-on payments for AI devices. But this fixed approach, if expanded, would likely lead to increased spending, overutilization, and even fraud. It also fails to account for the low marginal cost of AI, the likelihood of declining costs of producing medical AI over time, and the downstream effect of generating more billed treatments, for example, through more “upcoding” by Medicare Advantage plans and accountable care organizations.

As a related matter, there has been a push within traditional Medicare to move away from fee-for-service payments toward value-based payment approaches, which tie compensation to outcomes. To incentivize the use of specific AI devices that have been demonstrated to be effective and are already in use, CMMI should establish and test different payment models to evaluate cost savings, diffusion, improvements in patient outcomes, and other key metrics. Professors Ravi Parikh and Lorens Helmchen have proposed five payment approaches, summarized in the first five items of the list below, and I suggest a sixth mechanism.

1. Eliminate Separate Reimbursement of AI. Providers may accrue gains through an increase in downstream services, manufacturer incentives in gainsharing models, or better outcomes. This approach, however, may discourage the initial utilization of efficient technology that could eventually lower costs and improve outcomes in the long run.
2. Incentivize Outcomes. Reward providers for patient or process outcomes that are tied to AI. For example, reimbursements could be higher for stroke-detection AI devices if they are demonstrated to improve care.
3. Advance Market Commitments for New Solutions. Conduct a competition with a significant prize for the AI device that achieves certain health care goals. The winner would collect the prize and then make the code publicly available for general use and subsequent development.
4. Provide Time-Limited Add-On Reimbursement. Provide an additional payment for two or three years to help cover development costs, similar to Medicare payments for certain drugs not yet incorporated into bundled or episode-based compensation. Afterward, bundled value-based payments would be adjusted to cover marginal costs or reward performance.
5. Reward Interoperability and Bias Mitigation. To encourage broad and accurate adoption beyond the narrow existing use patterns that will likely be employed by AI training, payers should offer incentives for AI that are generalizable and interoperable across populations, geographies, and systems. A problem with this approach, however, is that it does not capture efficiency gains for payers—an issue similar to when the federal government essentially paid for electronic health record systems, presumably to improve efficiency, but never explicitly captured the savings.
6. Introduce Negative Incentives. To avoid permanent add-on payments and overuse, and to fairly embed cost-reducing and outcome-improving AI in provider workflows, we should penalize through lower payments providers (such as clinicians, hospitals, accountable care organizations, and Medicare Advantage plans, depending on the particular AI technology) that fail to use AI devices that have demonstrated good results. This mandatory model, if successful in maintaining quality of care while reducing costs, would capture efficiency gains for taxpayers and consumers while rewarding the investment made by manufacturers.

The critical importance of improving efficiency in the health sector is evident across many metrics. The promise of AI to make a major contribution must be fully realized but strategically incentivized, and CMMI is just the government agency to do so.

Mark J. Warshawsky is a senior fellow and the Wilson H. Taylor Chair in Health Care and Retirement Policy at the American Enterprise Institute, where he focuses on Social Security and retirement issues, pensions, long-term care, disability insurance, and the federal budget.