How to Reform the CMS Innovation Center with a Choice and Competition Approach

Key Takeaways

Introduction

In May 2025, the Centers for Medicare and Medicaid Services (CMS) Innovation Center released a new strategy consisting of three main pillars: promoting evidence-based prevention, empowering individuals to achieve their health goals through data transparency, and enabling and growing choice and competition in the American health care system. These pillars are all in service to the core principle of the new strategy: protecting the taxpayer. This new strategy provides a promising roadmap for the Innovation Center to get back on track and produce savings for taxpayers.

This paper builds on these three pillars and explores past failures at the Innovation Center, particularly those issues arising from the reliance on voluntary models, including benchmarks, quality improvements, and a lack of focus on savings. To address these issues, new models should:

• be true demonstrations or experiments that can verifiably test innovative policies;
• prioritize mandatory demonstrations;
• generally be limited in size, scope, and time;
• incorporate market-based principles, meaning expanding choice and competition in the market; and
• focus on producing tangible savings.

Additionally, this paper makes recommendations to policymakers about statutory and regulatory actions to ensure threshold accountability at the Innovation Center going forward.

Background

The CMS Innovation Center was created by the Affordable Care Act (ACA) to “test innovative payment and service delivery models to reduce program expenditures … while preserving or enhancing the quality of care.”¹ The models are essentially experiments on how payment policy can be changed to improve the efficiency of health care delivery in Medicare and Medicaid, with potential participants including all stakeholders who receive payment from those two programs. The Innovation Center received an initial $10 billion in funding for its first decade, followed by another $10 billion for the second decade and every decade thereafter.²

The Innovation Center has failed to live up to its establishing principles. Although it was initially expected to reduce net spending by $2.8 billion between 2011 and 2020, the Congressional Budget Office (CBO) further estimated in 2016 that the Innovation Center would save $34 billion between 2017 and 2026.³ The Innovation Center has in fact increased net direct spending by $5.4 billion between 2011 and 2020, and CBO projects that it will increase spending a further $1.3 billion between 2021 and 2030.⁴ According to a 2023 CBO report, the Innovation Center initiated 49 models between 2011 and 2020 with published evaluations. Of those, six generated “significant savings,” and only four were certified for expansion beyond their original parameters.⁵ Perhaps particularly damning, in the 14 years since the Innovation Center has been active and out of 90 current and former models, only four models have been authorized for nationwide expansion.⁶

Causes of Failure

The Innovation Center had several long-standing causes for its failure to reduce spending. These issues, highlighted in a 2021 New England Journal of Medicine article by the then-Director Brad Smith, include a reliance on voluntary models, concerns with benchmark design, and quality measurement problems.⁷ Within each of those issues, there were consistent problems with favorable selection—where participants were choosing only models that were most likely to be profitable for them with little downside risk—and a lack of focus on savings.

The Problem with Voluntary Models

Ideally, health care providers want to participate in innovative financing models because they also have a shared interest in reducing cost and improving quality. However, because these models often require providers to take on some financial risk for an uncertain gain, providers are often reluctant to join them. Forcing providers into models—particularly when the Innovation Center was new and political backlash around the ACA was building—is politically difficult. Thus, the Innovation Center generally incentivized providers to voluntarily participate. Many models are centered around a shared-savings approach: Participants take on the risk, and if they fail to reach a benchmark of savings established by the Innovation Center, they are responsible for the loss, but if they exceed the benchmark, they get to keep some or all of the savings. Participants are also incentivized through increased payments for meeting certain standards, particularly in models focused on improving care quality.

With voluntary models, providers generally participate only if they can reasonably expect financial benefit, meaning there is either little savings left over for the government or very little risk to the providers if they fail to produce savings. Thus, favorable selection abounds in voluntary models. Some providers who find themselves losing money will drop out of voluntary models or not even sign up to begin with. To prevent this, the Innovation Center occasionally alters voluntary models in order to keep participants, further reducing savings as concessions are given to the participating providers.⁸

For example, the Oncology Care Model (OCM) had to add additional payments to providers to ensure robust participation, which ultimately prevented net savings from occurring.⁹ The OCM had a relative reduction in spending of 2.1 percent compared to non-OCM episodes of care—meaning all services rendered to a patient for a specific condition in a given length of time. Ultimately, this was not enough to offset both upfront and performance-based payments meant to encourage participation that led to a net loss of $639 million for Medicare.¹⁰ Other models had similar issues.

Benchmarks in Innovation Center Models

Benchmark Design

A central issue with benchmarks in voluntary models is that the Innovation Center typically sets them prospectively rather than retrospectively (see text box). In short, a prospective benchmark is used to develop a hypothetical of spending in the market without the demonstration.

Because a prospective benchmark is locked in before the model starts, it does not capture what would have happened anyway without the model—normal changes in prices, volumes, coding, or competition. That makes it hard to know whether any reported “savings” came from the model or from normal market changes. In the Comprehensive Care for Joint Replacement (CJR) model, these preset targets did not move with the market, so participants could earn bonuses even when prices or volumes outside the model shifted. Independent evaluations using retrospective benchmarks found that the prospective model benchmarks showed savings to be 235 percent higher than the retrospective benchmarks in the control groups of the evaluations showed—meaning models that appeared to produce gross savings of $1.9 billion actually produced net losses of more than $583 million.¹¹

The ideal prospective benchmark would accurately predict hypothetical spending in the absence of the model, but this is impossible to do perfectly—costs change over time for reasons of supply and demand as well as innovation in techniques and technology. For example, the CJR model set the benchmark for each procedure as the average cost of the procedure across the previous three years. When the model was active (2016-2022), the CJR model generally showed savings, but the three-year benchmark created an unforeseen problem: As a result of a combination of changes in patient preferences, non-CJR Medicare payment policies, and increased clinical evidence for best recovery practices, there were significant yearly decreases in the use of skilled nursing facilities after hip and knee surgeries, which the CJR’s three-year average benchmark design could not account for as quickly as the general market could. Episodes of service in the CJR model (after accounting for these decreases) were significantly more expensive than their comparison episodes. In its fifth year of operation, the CJR model lost $95.4 million. It was only after major payment reductions were made in the sixth year that there were estimated savings—$30.8 million across all six years after payment reductions resulted in $54.2 million in estimated savings in the final year.¹²

Retrospective benchmarks avoid some of these issues in that they naturally incorporate market trends over the same period (because the general costs or control group reflects those changes). However, such benchmarks also have some drawbacks. First, they make it difficult to direct behavior change as they are set only after participants have made decisions and completed changes to how they provide services. This inherently limits the scope and extent of interventions or changes in behavior providers can consider, as participants do not know what they are being measured against. If the providers knew they needed a small amount of savings, they would likely make small changes to their practices, whereas knowing they need a large amount of savings would result in greater changes to their practices. Put another way: It is difficult for participants to set a course, including investing in changes, without a map. Additionally, retrospective benchmarks tend to make voluntary models more difficult to execute, because participants do not have clarity regarding the financial targets and are thus even less willing to participate given that they do not know the potential risk.¹³

However, this challenge can be mitigated to some extent: Retrospective benchmarks that are strictly formulaic and leave no discretion at all for the Innovation Center when they are calculated help participants better understand how they are being measured and ensure that success in a model is not dependent on arbitraging or predicting government decisions. Retrospective benchmarks can also be improved if they are based not on specific numbers (e.g., what would have been spent without the model) but instead on who achieved the most savings (e.g., the top 10 percent of participants in terms of total savings get to keep 90 percent of their savings, the second best decile get to keep 75 percent, and so on). This may allow for participants to adjust for actual market conditions and incentivizes participants to find ways to save the most amount possible—not to merely settle for an objective the Innovation Center has set. CMS’ proposed mandatory Ambulatory Surgery Model takes an approach like this by tying payment to participants’ performance relative to their peers.¹⁴

Additionally, the use of averages in benchmarking (necessary to account for regional or national spending on a given service or population) leads to favorable selection in voluntary models. Some providers will have costs below the benchmark (meaning they are likely to profit from participation), and others will have costs above it (meaning they are likely to lose money from participation). Providers on the losing end are unlikely to participate in the model, meaning a model may show savings that reflect only the participation of providers who knew they were likely to make a profit before starting. Voluntary models inherently create this favorable selection problem: Providers will participate only if they believe it will be profitable. Therefore, in general, providers who are more efficient and have the resources and know-how to make a model profitable will participate.

Quality Measurement Issues

Quality improvement in Innovation Center models—a core initial aim—remains elusive. Although the Innovation Center can collect quality metrics on its models through a wide variety of means—including surveys, registries, and claims, only claims data is normally available for the control-group.¹⁵ In 2021, the Innovation Center’s models had control group data for only 55 percent of the quality metrics used in models. Meanwhile, participants are still paid based on quality metrics that cannot be compared to a control. Smith explained that

across a representative sample of nine models, 71 quality metrics were used to determine performance-based quality payments, but only 39 of those metrics were included in the evaluations…. 99 of the 138 quality metrics that were used in the Center’s evaluations were not included in the models’ performance-based quality payments.¹⁶

This means the Innovation Center was frequently paying providers to meet quality metrics for which it had no comparison to determine if these participants were actually improving quality over standard practice.

A more fundamental problem is that measures of quality are also notoriously difficult to create: The rule of thumb is often “what can be measured will be measured.” This leaves out less tangible measures of quality—the difference between a physician practice checking a patient’s chart about his last vaccine versus having a conversation about his health history. Measures of quality determined by government bureaucrats may not align with the measures that patients actually value. As described in a previous Paragon report, “Patients exercising their preferences about where to receive care can be a better indicator of provider quality than measures designed through bureaucratic processes.”¹⁷ The difficult-to-define nature of what “quality” care means, combined with the voluntary nature of most models, creates favorable selection problems with quality metrics: Providers prefer to join models with quality metrics that are viewed as easier to meet. Thus, in order to attract needed participants, the Innovation Center is motivated to put relatively less emphasis on more meaningful quality metrics.

402 Demonstration Waiver Authorities in CMS

Lack of Focus on Savings

Across all three areas—voluntary models, benchmark design, and quality improvement measurements—the root of the problem is the lack of focus on savings. Instead, during its first decade, the Innovation Center focused on ensuring participation in models. Voluntary models were more politically palatable than mandatory ones, but they require strong incentives to get providers to participate. These incentives result in increased payments to participants and benchmarks being adjusted to make goals easier to meet, functionally eliminating much of the savings potential for many models. Paying providers for reaching “quality metrics” that could not ensure that quality improved indicates that even this core purpose of the Innovation Center was subject to the demands of maintaining participation. To quote one CMS evaluation of 21 models: “Generous financial incentive payments, which helped ensure robust participation in models, made it difficult for many models to demonstrate net savings.”²⁰ While participation in models is important, the statutory purposes of the Innovation Center should not be sacrificed to that end.

First Principles for Fixing the Innovation Center

Despite these problems, the Innovation Center holds promise as a tool of policymaking to produce genuine improvements in spending and quality in federal health programs. But to match its lofty mission, the Innovation Center needs to reorient around the concept that models are demonstrations. Importantly, all of these demonstrations should be true and limited demonstrations, and mandatory demonstrations should be prioritized. If a model is shown to work, then Congress can make its feature permanent, or CMS can pursue broader rulemaking.

True Demonstrations

Statute allows the Innovation Center to waive existing laws and regulations governing Medicare and Medicaid in order to implement its models.²¹ The core purpose of the Innovation Center models is to experiment with policy changes with the goal of producing savings and improving quality. “True demonstrations” are experiments that seek to test if a policy saves taxpayer dollars and improves quality. Experimentation—and the ability to learn from the demonstration—should be at the heart of every model.

The Innovation Center has not always abided by this core principle. For example, the Biden administration attempted to use the Innovation Center to implement a Two Dollar Drug List Model, which the Trump administration ended before it could be implemented.²² The proposed model—flat $2 copays for select CMS-listed generic drugs across all benefit phases—was not an experiment: Roughly 20 percent of beneficiaries (mostly in Medicare Advantage) already have $2 generic copays. In 2024, nine of 14 national stand-alone prescription drug plans offered $0 copays for preferred generics, and the median copay for non-preferred generics was $5.²³ This “model” was simply an attempt to standardize and potentially expand this benefit offering (even to non-generics) and make it mandatory. This would have pushed up premiums and, because of the IRA’s 6 percent cap on premium increases, federal subsidies. Real experiments test novel ideas against credible counterfactuals. This proposal did neither and risked misuse of Innovation Center authority.

The Innovation Center should make true experimentation a core standard of all models. Models, like all other scientific experiments, should be designed, tested, and evaluated. They need to demonstrate measurable results—not simply continue operating because they are politically popular or administratively easier than pursuing legislative change.

Time Limits

For similar reasons, demonstrations need to have clear time limitations, with extensions generally avoided. Given the broad waiver authority, models without time limits could be continued indefinitely and after the model has stopped demonstrating anything, particularly if a demonstration is politically popular. Once a demonstration reaches the end of its prescribed time period, the Innovation Center should conduct its standard end-of-demonstration evaluation and make a recommendation to Congress and CMS on whether to implement the model on a nationwide scale.

Limited in Size and Scope

There are important reasons to consider limits on the size and scope of models. The purpose of the Innovation Center is to test innovative models, not to implement sweeping policy under the guise of a model. As such, for example, demonstrations that include the entire (or nearly the entire) Medicare or Medicaid population allow for untestable policy to functionally become the law of the land without the appropriate congressional or regulatory actions. Potential limits on models could be to implement the policy change only in a distinct set of regions, states, or metropolitan statistical areas (MSAs) or a subset of beneficiaries across the nation, including through randomization. Under current statute, a model may be expanded beyond its initial scope if (1) the Secretary of Health and Human Services (HHS) determines that the model is likely to save money without reducing quality or improve quality without increasing spending, (2) the chief actuary of CMS determines that the expansion would reduce net program expenditures or not result in an increase, and (3) the Secretary of HHS determines that such expansion would not deny or limit coverage or the provision of benefits.²⁴ Going forward, the Innovation Center should adopt a policy of doing expansions only through full notice-and-comment rulemaking. For demonstrations the Innovation Center would like to be made permanent or span the entire (or nearly the entire) Medicare or Medicaid population, the administration should instead request that Congress expand such models through statute. On top of the Innovation Center adopting these policies on its own, Congress should consider putting such guardrails into statute.

Furthermore, there is merit to limiting the number of models that are active at any given point in time. For example, in 2023, 39 percent of all Medicare fee-for-service beneficiaries were in Medicare accountable care organizations (groups of providers that aim to coordinate care and share in the cost or savings of episodes of care), including those in the Medicare Shared Savings Program (which is not an Innovation Center model). This large proliferation makes it difficult to test models without other programs interfering with the results. CBO has noted that the proliferation of multiple models within single health care systems can create conflicting incentives for providers.²⁵ These overlapping models and programs may also increase administrative burden and complexity, which can create a disincentive for small providers to participate, as discussed previously. More transparency regarding the overlap of models and attribution of savings would help mitigate this particular concern. It would also provide public insight into the impact of models on certain subsets, such as smaller providers and beneficiaries who rely on them.

When smaller providers are more likely to eschew models due to complex administrative requirements, it means larger providers make up the bulk of participants in models, and thus the results and lessons reflect only the experience of those larger providers. One way to mitigate this is to create separate tracks for smaller providers, assuming that can be done without undermining the utility of any benchmark as discussed above. Without smaller providers participating one way or another, the model cannot show if there were any potentially negative effects on these smaller providers. As such, any program-wide changes made to Medicare payments may unintentionally bias changes in Medicare payments against smaller providers.

The effect of administrative complexity extends to the internal operations of CMS generally as well, which has a history of being hamstrung by technological and operating deficiencies that have resulted in the loss of millions of dollars in waived or improper payments.²⁶ Fewer models would help mitigate this issue and allow more resources to be focused on ensuring effective model implementation.

Ensuring that models are limited in size, scope, and length of time and requiring transparency along the lines outlined above also mitigates concerns about a potential “medallion effect”—in which early model participants obtain advantages. But most importantly, such limitations collectively safeguard against future abuses that attempt to change policy through demonstrations on shaky bases, including with respect to lower or same cost or quality, which are effectively subject only to administrative self-restraint.

Prioritize Mandatory Demonstrations

The Innovation Center should prioritize mandatory demonstrations. Concerns about the potential for mandatory models to become de facto changes to law can be ameliorated by ensuring that all models are limited and fulfill the definition of a true demonstration, as described above. Mandatory demonstrations eliminate the favorable selection problem whereby participants choose only models they are likely to profit from. Ultimately, the true potential of a model to achieve program-wide savings is unclear if the participants are functionally limited to only those who are best equipped to achieve those savings.

Voluntary models can succeed if the benchmarks (and incentives) are crafted correctly. However, this has proven very difficult to accomplish, particularly when benchmarks are made prospectively. The downsides of voluntary models typically outweigh the greater difficulty in implementing mandatory models.

Mandatory models require the Innovation Center to go through the rulemaking process, which significantly delays a demonstration, and comments from stakeholders and entrenched interests can also impact the program’s design. When mandatory models are not practical for political or bureaucratic reasons, the alternative options presented to providers should be limited and not simply allow the target providers to maintain the status quo (with exceptions for any necessary control groups). The recent WISeR model released by the Innovation Center has narrowed the choices that providers have in cooperating with participating companies without subjecting them to mandates (see text box).²⁷ This approach to voluntary models is a clever way to reduce the problem of favorable selection.

The WISeR Model

Focus on the Savings

Models have often reflected an either/or emphasis when it comes to focusing on savings or quality. But the statute establishing the Innovation Center demands both: “test innovative payment and service deliver models to reduce program expenditures … while preserving or enhancing the quality of care”²⁸ However, in health care, quality and efficiency often go hand in hand. For example, one analysis found that hospitals with lower costs also had lower rates of patients experiencing harm and that hospitals with better margins frequently deliver better care.²⁹ Models do not need to choose between lower costs and better quality.

The new strategy for the Innovation Center refocuses on making “protecting the federal taxpayer a foundational principle” of the agency.³⁰ The strategy includes requiring downside risk for participants and providers, refining and simplifying benchmarking methodology, prioritizing high-value care and services, and ensuring that all model tests are fiscally sound with a pathway to certification, among other items.³¹

There are other actions that can ensure that the Innovation Center remains fiscally prudent. Some demonstrations may not work despite best efforts, but the Innovation Center should pursue (and Congress should follow up in statute with) a policy requiring savings throughout the lifespan of a demonstration. Specifically, a given model should be initiated only if CMS’s Office of the Actuary’s midpoint projections for that model show positive net savings, which would serve as a guardrail against launching models based on overly optimistic assumptions. Similarly, Innovation Center policy (and eventually, statute) should require that those demonstrations whose costs exceed a certain predetermined threshold in the middle of their duration should have periodic assessments if future savings are likely to be produced. Such predetermined thresholds can be model-specific to account for different sized models. If savings are unlikely after an initial grace period, the Innovation Center should terminate the model.

Additionally, notwithstanding the value of limiting the number of models as outlined above, while all models should be savings-focused, it should be emphasized that policymakers should consider judging the overall success of the Innovation Center based on the aggregate amount of savings produced by models and not based on the number of models that produce savings. Models are experiments, and the potential for experiments to fail is a crucial part of learning. For example, if the agency launches ten models, nine of which lose money with aggregate losses of $15 billion, but the tenth model saved $5 billion, the total portfolio cost would be $10 billion, and the Innovation Center should be considered a failure.

Defaulting to a Market-Based Approach

The Innovation Center is tasked with designing new payment models to save taxpayer dollars and improve quality. This is difficult in the convoluted U.S. health care system, where the primary recipients of services (patients) are not the primary payers (third parties such as insurance companies or the federal government), and so the true value of any given service to the patient—measured in both cost and quality—is functionally unknowable, as the services are oriented toward the requirements of payers and not the choices of patients. Private insurers are more responsive to patient preferences, because patients provide revenue through premiums, whereas government payers (Medicare, Medicaid, etc.) have no such incentives.

Therefore, the Innovation Center should default to a market-based approach, meaning that models look to market solutions instead of government intervention. Models that seek to have payments reflect market conditions—and specifically empower patients to be the primary determinants of value—should be prioritized. Helpfully, the Innovation Center has already stated that promoting choice and competition and empowering individuals by aligning financial incentives with health priorities are two of the three pillars of its new strategic vision. The Innovation Center would do well to follow through on this promise by publishing models that deploy a market-based approach and/or seek to discover aspects of the market, such as prices based on what patients actually value. While it is still very early, the latest models, including the WISeR model and the latest updates to the Achieving Healthcare Efficiency through Accountable Design (AHEAD) model (see text box), indicate that the Innovation Center intends to follow through on the promise of its strategic vision.³² Other topics of exploration that would benefit from market-based demonstrations include market-informed pricing, shoppable services, and competitive bidding.³³

The AHEAD Model Update

Recommendations for Statutory Changes

One primary criticism of the Innovation Center is that its waiver authority is too broad and delegates too much congressional authority and power to an executive branch agency. Congress should ensure that the Innovation Center remains true to its experimental purpose and does not become a de facto way for sweeping changes to Medicare law across the country. As shown above with the Biden Medicare Two Dollar Drug List Model demonstration, the waiver authority granted to the Innovation Center is broad and could be abused by an administration looking to circumvent Congress to use the Innovation Center to make nationwide, sweeping policy changes.

Congress should amend the statute to include boundaries on the length of time demonstrations may last. The ideal length of time is subject to reasonable debate, but around five to seven years is probably close to an optimal time limit for most models. Additionally, Congress should limit the number of demonstrations in effect at any given point in time. Flooding the program with demonstrations, even if they are time-limited, is another potential way for the executive branch to bypass congressional authority. The Innovation Center currently has 25 active models. It should work with Congress to develop a reasonable limit on the number of models, at the same time not allowing models in their final stages to impede new models starting up. Alternatively, Congress may want to consider not allowing more than a set number of models to be introduced annually.

Furthermore, Congress should place limits on the geographic spread of demonstrations. This can be done by limiting the number of MSAs, regions, or states included in models at any given point in time. Population limits should also be considered. For example, limiting a model to areas with a population equivalent to the 10 largest MSAs would cover a total population of over 90 million people. Keeping demonstrations limited requires keeping their size limited. Finally, any new demonstrations that are successful and might qualify to be scaled up nationwide should be approved by Congress before that happens.

Finally, Congress should critically evaluate the Innovation Center after a reasonable period of time to determine whether the reforms initiated in the Trump administration have succeeded in reversing the disappointing trajectory of the Innovation Center. If, at that time, an evaluation finds that the Innovation Center itself has cost more than it saves, it is reasonable for policymakers to terminate it.

Conclusion

The Innovation Center has largely failed to design models that lower costs and improve quality. But under new leadership and direction, the Innovation Center has a chance to improve Medicare and Medicaid policies, and its new strategy is a promising start to rectify past failures. To prove that its promise is achievable and earn its continuation, the Innovation Center should refocus its idea of a model toward true, limited, and primarily mandatory demonstrations that are market-based and focus on creating transparent savings. This new approach would ultimately ensure that scarce taxpayer funds are well used and, most importantly, that patient care is re-centered away from a bureaucratic process.

Footnotes