The Program Built to Cut Health Spending Raised It Instead

In 2010, the Affordable Care Act created the Center for Medicare and Medicaid Innovation -- a unit inside CMS, the federal agency that runs Medicare and Medicaid -- and gave it one statutory job: test new ways of paying for care that reduce federal health spending as well as improve quality of care for beneficiaries⧉.▸ Fifteen years and $11.4 billion later, GAO⧉ -- Congress's own independent auditor, whose findings agencies are expected to answer -- reports that the center has tested 70 of these payment models and certified just four of them, a 5.7% hit rate, for permanent, nationwide use. The Congressional Budget Office's own check on the entire program, cited in GAO's report⧉, found it did the opposite of its job: net federal spending rose by $5.4 billion from 2011 through 2020, not down.

The center's process runs in four phases: identify a payment idea, design a model and estimate its likely effect on federal spending, build the infrastructure and recruit participants, then operate and evaluate it for a testing period -- typically under 10 years⧉.▸ To go from test to permanent policy, a model has to clear a bar CMS administers itself: its Office of the Actuary has to certify that expanding the model won't increase net program spending, and CMS has to determine the expansion won't cut beneficiary coverage. In practice, that means officials look for a 90-to-95% likelihood⧉ the model would reduce federal spending before certifying it -- a demanding threshold only four of the 70 models tested since 2011 have cleared.

The Pioneer Accountable Care Organization Model⧉ gave experienced provider groups more responsibility for costs and savings; it produced $384 million in Medicare savings in its first two years (2012-2016), and CMS folded its elements into the Medicare Shared Savings Program in 2015.▸ Prior-authorization requirements on repeated non-emergency ambulance rides -- the kind billed for routine dialysis trips -- saved $1 billion over five years of testing (2014-2020) and were rolled into other CMS compliance programs in 2020. A Home Health Value-Based Purchasing Model that rewarded home health agencies for better outcomes saved $949 million from 2016 to 2020 and was expanded in 2022. And a diabetes-prevention coaching program for Medicare beneficiaries with prediabetes saved $2,650 per participant in its first 15 months, expanding in 2018. Three more tested models -- the Accountable Care Organization Investment Model, the Maryland All-Payer Model, and the Medicare Care Choices Model -- also produced net savings during testing but were never expanded, CMS officials said, for reasons including limited participation.

The four successes are real, but they are a small fraction of what the Innovation Center has cost to run.▸ The Congressional Budget Office -- Congress's own official scorekeeper for the budgetary effects of legislation -- estimated that the center spent $7.9 billion operating models from 2011 through 2020, and that those models reduced federal health spending by $2.6 billion over the same period. Net that out and the program didn't save money at all: it added $5.4 billion to federal spending, the opposite of what the 2010 law built it to do. CBO has since projected a further $1.3 billion increase from 2021 through 2030. Even if every dollar the four expanded models saved during their own testing periods is added up -- $2.3 billion combined, from the three that report a program-total figure -- that sum covers less than half of what CBO says the whole 70-model effort cost in net spending over roughly the same stretch.

The Innovation Center's own analysis names three reasons its models struggle to produce savings, and all three trace back to one design choice: 22 of its 24 active models let participants choose whether to join.▸ Voluntary models need generous incentive payments to attract participants, which eats into any savings. They invite selection bias -- providers tend to join only when they already expect to come out ahead. And they suffer from attrition: participants facing possible losses often quit before a model gets the chance to prove it works. Among the 18 completed models with final evaluations, attrition among participating health care organizations ranged from zero all the way up to 78%. CMS has responded by planning to test more mandatory models: only two of the 24 models active in January 2025 required participation, but the mandatory Increasing Organ Transplant Access Model began testing in July 2025, the mandatory Transforming Episode Accountability Model followed in January 2026, and a mandatory Ambulatory Specialty Model is planned for January 2027.

The money being spent is also shifting away from the models themselves.▸ In fiscal year 2015, 82% of the center's $1.3 billion in annual obligations went to model-specific work -- contracts to implement, monitor, or evaluate a given model. By fiscal year 2024, that share had fallen to 54% of a much smaller $789 million budget, while the combined share going to shared administrative overhead and cross-model support functions grew from about 18% to nearly 46%. CMS attributes the shift to consolidating IT and other common functions across models for efficiency; whatever the cause, close to half of what the program now spends each year isn't going toward testing a specific idea.

CMS has also said, in its own 2024 report to Congress, that it has never expanded a model primarily because it improved care quality -- only because it didn't raise spending -- and that it's developing a new strategy to change that, with no date attached to when that strategy takes effect.▸ Its own newly adopted 2026 performance target for the share of measures across its models that are outcome-based, rather than process-based, is 53%, just one point above the 52% baseline it started from in 2025. Meanwhile, $8.0 billion of the $20 billion Congress dedicated to this program remains unobligated, available through the second decade's appropriation, which runs through fiscal year 2029.