Medicare and Medicaid cover 12.1 million of the same people — 3 percent of them are in a plan accountable for both

What is happening?

About 12.1 million people were enrolled in both Medicare and Medicaid in July 2024, roughly 18 percent of all Medicare beneficiaries — 8.7 million with full Medicaid benefits and 3.4 million with partial benefits. Of the 11.7 million counted in the MedPAC coverage table for that month, 358,000, or 3 percent, were in a health plan that assumes clinical and financial responsibility for both programs at once. Everyone else holds two coverages that qualify separately, pay separately and are appealed separately.

The label has grown much faster than the thing. Dual eligible special needs plans, known as D-SNPs, held 14 percent of this population in 2014 and 46 percent in July 2024. But 27 of those 46 points sit in the coordination-only tier, which MedPAC describes as having the lowest level of integration because such plans need not provide any Medicaid service at all. Highly integrated plans held 15 percent and fully integrated plans 3 percent, and MedPAC reports the relative shares across tiers have been fairly stable.

The one arrangement that bound the two payers together by contract closed on 2025-12-31. The Financial Alignment Initiative, whose first demonstrations began in July 2013, ran a capitated model under a three-way contract among CMS, a state and a plan. Seven states ended those demonstrations on that date — Illinois, Massachusetts, Michigan, Ohio, Rhode Island, South Carolina and Texas. Of the people enrolled in Medicare-Medicaid Plans as of July 2025, about 66 percent moved into fully integrated D-SNPs, about 18 percent into highly integrated D-SNPs, and about 16 percent were in plans that closed without converting.

Whose problem is this?

No single body on that list can finish this alone. The commission that recommends has no authority, the agency that regulates cannot make a state sign or a person enroll, the state has no role in administering Medicare, and the person at the center is choosing a plan for reasons that have little to do with integration.

Where does this problem end?

The boundary matters here because both programs already exist and neither one is missing.

What is absent is a party who owns the whole person, and a rule that makes anyone become one.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a share of this population that should be in an integrated arrangement, a date by which it should be reached, or a definition of success. The MedPAC chapter exists because Congress mandated a periodic assessment of how these plans perform relative to each other, and it sets no destination. The MACPAC recommendation of June 2022 asks Congress to require each state to write a strategy within two years, with no floor, no share and no consequence for not writing one. The statutory report where a federal target would most plausibly appear could not be opened this round.

How big is it?

Between 6.0 million and 11.0 million people, as of July 2024. The denominator is the MedPAC coverage table universe of 11,691,000, which excludes people lacking both Part A and Part B and residents of the territories. What is subtracted from it depends entirely on which published definition of integration is used, and the sources do not agree.

Under the narrowest bar — the CMS standard for plans that assume clinical and financial responsibility for both programs, which is highly or fully integrated status combined with exclusively aligned enrollment — 358,000 qualify, plus 256,000 in Medicare-Medicaid Plans and 60,000 in PACE. That leaves 11.0 million, or 94 percent, outside. Under the widest bar, counting every D-SNP tier including coordination-only, 5,694,000 qualify and 6.0 million remain, or 51 percent.

No source opened here chooses one of those definitions over the other.

An independent check points the same way. The KFF landscape analysis found that just 5 percent of this population received both sets of benefits through a single coverage arrangement in March 2021, counting only PACE and Medicare-Medicaid Plans, which implies roughly 11.6 million outside — slightly above the high bound, which is what the narrowest definition applied to an earlier year should produce.

The last published measurement is July 2024, and the figure is not restated for 2026 in anything opened here, so no post-closure number is computed.

Under what conditions does it arise?

1. No ledger holds the whole person. Eligibility is established separately in each program. Medicare covers hospital, post-acute, physician and drug benefits and is the primary payer where both programs overlap; Medicaid covers long term services and supports, wraparound benefits such as dental and transport, and pays Medicare premiums and cost sharing. MedPAC states that people in this position may receive care that is fragmented or poorly coordinated because of the difficulty of navigating two distinct and complex programs.

2. The payer that invests is not the payer that saves. MACPAC reported in June 2023 that states must spend resources to implement an integrated program but see the savings accrue to Medicare through reduced hospital and emergency department use, and that a shared savings mechanism comparable to the one carried by the three-way demonstration contract is not currently available for D-SNPs.

3. State capacity was never built, because there was never a reason to build it. MACPAC reported in June 2020 that because the two programs operate largely independently there has been little incentive for state staff to develop Medicare expertise, and that states have no role in administering Medicare at all. Arizona and Virginia are named as the states that built dedicated Medicare units.

4. Enrollment is voluntary and the demand driver is orthogonal to integration. MedPAC focus groups find these plans appealing largely for their extra benefits rather than for integrating the two coverages. Those extras are financed by Medicare Advantage rebates, with no Medicaid funding involved, so a plan holding no Medicaid contract can match them.

5. Every restriction on the regulated market opens an unregulated one. States control D-SNP entry through the State Medicaid Agency Contract and are not obliged to sign with every insurer. Chronic condition special needs plans used as D-SNP look-alikes rose from 5 in 2021 to 92 in 2026, adding roughly 100,000 enrollees between 2021 and 2025 — plans marketed to this population that carry no Medicaid contract whatsoever.

What has been tried?

Two directions were tried across two decades — build a contract that makes one entity accountable, and grade a plan type that already existed. The first ran for twelve and a half years and closed. The second grew to nearly half of this population, mostly in the tier that owes no Medicaid service.

What was found?

Why is it still unsolved?

Coordination failure — the fix requires Congress, CMS, fifty state Medicaid agencies, insurers and the beneficiary to move in the same direction at the same time, and each of them can decline without anyone being at fault.

The first movement is that every actor holds one lever and none holds enough. Congress could require integration and was asked to in 2013 and again in June 2022; neither request was enacted in anything opened here. CMS sets the plan standards but cannot compel a state to sign a contract or a person to enroll. States decide who may sell a D-SNP but have no role in administering Medicare, and MACPAC found they had little incentive to learn it. Insurers respond to state selectivity by selling a plan type that carries no Medicaid contract at all. And the beneficiary, whose enrollment is voluntary, picks on extra benefits. A gap that no single actor can close alone is not the same as a gap nobody noticed.

The second movement is that the one device that aligned the incentives was retired. The three-way capitated contract put CMS, the state and the plan on the same paper and carried a shared savings mechanism precisely because states pay for integration and Medicare collects the return. The managed fee-for-service variant paid a state retrospectively only if Medicare hit a savings target. MACPAC records that no comparable mechanism exists for D-SNPs. The demonstration also carried passive enrollment, identified as the factor most associated with higher take-up and which MACPAC records is not possible outside the three-way contract. Both went away on 2025-12-31, and what remains is a graded plan type in which the fastest growing tier is the least integrated one.

The third movement is that the failure is invisible in the headline. Nearly half of this population is now in a plan named for it, which reads like progress and is reported as such. The number that describes accountability for both programs is 3 percent, it is published in one table by one commission, and the source that would corroborate it independently could not be opened. Meanwhile the flagship joint data reference on this population, published in December 2025, reports no integration tier enrollment at all — it reports a proxy and calls it the potential for greater integration. When the measure of the problem is not in the book that measures the population, nothing forces the question.

What observation would mean it is solved?

Candidates — (a) the share of dual eligible individuals in a plan accountable for both programs rises well above 3 percent (b) a federal instrument obliges every state to operate an integrated arrangement, with a date attached (c) the operational burden falls, so that one card, one assessment and one appeals process become the ordinary case rather than the exception.

(a) alone is weaker than it looks. The share moves when definitions move, not only when care changes. The 2025 tightening of the fully integrated definition pushed plans holding about 110,000 enrollees down a tier without any change in the care those people received, and the closure of the demonstration moved about 160,000 people up one. A number that can rise or fall by six figures on a regulatory reclassification is not, by itself, an observation about people.

(b) alone counts paper. The demonstration was exactly this kind of instrument and it enrolled between 5 and 62 percent of eligible people across the nine capitated states. A mandate that produces plans nobody joins reproduces the current position under a different name, and the 2027 alignment rule already carries exceptions and applies only where Medicaid managed care serves this population.

(c) alone is unmeasured. No source opened here reports how many people in this population hold two cards, sit two assessments or file into two appeals systems. The remedies are precisely scoped to applicable integrated plans, which was 358,000 people in July 2024, and the size of the group on the other side of that line is stated nowhere as a count. The three have to be read together, and (a) has to be read against the definition it was computed under.

What is it connected to?

Fills with researchlong term services and supports capacity and the Medicaid home care queue, Medicare Advantage market conduct and marketing oversight, Medicare Savings Program take-up, and comparable dual coverage arrangements in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• No federal target and no deadline. No source opened here names a share of this population that should be integrated, a date, or a definition of success. The MedPAC chapter is a comparative assessment mandated by Congress and sets no destination; the MACPAC recommendation asks only for a strategy document. The statutory report to Congress where a target would most plausibly live returned HTTP 403 to every attempt, so this is stated as what was not found rather than as an absolute.
• No two sources define integration the same way, and the choice moves the answer by roughly a factor of two. One counts PACE and demonstration plans only and reaches 5 percent. Another counts three D-SNP tiers plus those two and reaches 46 percent, 3 percent and 3 percent depending on the tier. A third works from a unified appeals definition. None states which definition it is departing from, and nothing opened here reconciles them.
• Nothing restates the narrowest category after the closure. Demonstration plans were 256,000 of roughly 316,000 in the single coverage arrangement category as of July 2024. After 2025-12-31 that category is essentially PACE alone, about 60,000 people. No source opened here publishes a post closure share on that definition.
• The counting basis is almost never in the sentence that carries the headline number. 13.6 million is an ever enrolled count for CY2022. 12.1 million is a July 2024 point in time. 11.69 million is the same month minus people without both Part A and Part B and minus territory residents. About 12 million is a 2023 survey figure. The spread is roughly 1.9 million, larger than the entire integrated population under any definition, and only the data book states its basis inline.
• Nobody counts the burden the problem is named for. No source opened here reports how many people hold two insurance cards, sit two health risk assessments, or file into two appeals systems. The remedies are scoped exactly; the population on the other side of the scope is never sized.
• What integration costs a state is never priced. The MACPAC recommendation asks Congress for additional federal funding to states with no dollar figure, no per state estimate and no appropriation named. The only price attached anywhere in this round is the 2011 design contracts of up to USD 1 million each to 15 states. The claimed barrier is unpriced by the body that claims it.
• The 3.4 million partial benefit duals have no integration policy, and no source says so. Every instrument opened here is written for full benefit duals — the fully integrated status, exclusively aligned enrollment, the 2027 one plan rule, the strategy recommendation, the new enrollment period. The 2027 rule permits states to allow a separate D-SNP for partial benefit duals, which is a carve out rather than a plan. Nothing opened here states what is intended for these 3.4 million, or that nothing is.
• There is no accepted verdict on the largest attempt ever made. The official evaluator reports little impact on Medicare expenditure for the capitated model and a significant reduction for the Washington managed fee-for-service model. MedPAC reports that the demonstrations tended to result in statistically significant increases in Medicare spending, without splitting the models. The stated CMS reasons for ending the program concern plan permanence and regulatory convergence rather than savings. Twelve and a half years, a peak of roughly 400,000 to 450,000 people, and no reconciled finding in anything opened here.

See the evidence

No primary federal program document was read from the agency that runs the program. Every CMS-hosted page attempted in this round returned HTTP 403, so the demonstration facts, the stated reasons for closing it and the coordinating office report are all absent as primary sources and are carried instead by the two congressional commissions, the contracted evaluator, one federal rule text and trade analysis. Only one source was read end to end — the March 2026 MedPAC chapter. The rest were opened and text-extracted rather than read whole: the December 2025 joint data book was searched for integration tier terms, which is how its silence on them was established rather than assumed; the three MACPAC documents and the March 2020 GAO report were extracted and read for the passages cited here; and the raw text of the 2025 rule was read only from the beginning, because the fetcher stops there. That last limit has a consequence worth naming — the look-alike and plan benefit package provisions of that rule could not be confirmed from the rule itself, so they appear here on the authority of MedPAC instead. Where sources overlap they agree: the demonstration end date of 2025-12-31 appears in four opened sources with seven states named identically in two, and the tier structure appears in two MedPAC products. But those two are one analysis published twice rather than two findings, and the 358,000 exclusively aligned figure rests on a single table whose independent corroboration was refused. Where sources disagree the disagreement is left visible rather than resolved — the full benefit share is 71 percent in one source, 74 percent in another and about 75 percent in a third, none of which states its data year beside the split; the effect of the demonstrations on Medicare spending is a reduction, no impact or an increase depending on which document and which model is being described, and the two statements have different scopes and are not printed here as a contradiction; and the headline population count is 11.69, 12.1, about 12, or 13.6 million depending on a counting basis that only one of those four sources states inline. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 5,997,000–11,017,000 As of July 2024

Sensitivity The width of the range is not sampling error and is not disagreement about how many people exist. It is the width of the word integrated: the same July 2024 table yields 5,997,000 or 11,017,000 depending on which published definition of an integrated arrangement is applied, and no source opened in this round chooses between them. An independent check on the narrow side points the same way: KFF found just 5 percent of this population in a single coverage arrangement in March 2021, counting only PACE and Medicare-Medicaid Plans, which implies roughly 11.6 million outside, slightly above the high bound as the narrowest definition applied to an earlier year should produce. What this number cannot count: people who would qualify as dually eligible and are not enrolled, because no current national take-up estimate for the Medicare Savings Programs was found in this round and the 2025 reconciliation law froze the rule that would have raised enrollment; people whose dual status is episodic, since 16 percent of duals were not dually eligible in all twelve months of CY2022 and a July snapshot counts them once or not at all; people enrolled in look-alike plans that carry no Medicaid contract at all, whose total enrollment is nowhere published; and the 3.4 million partial benefit duals, who sit inside both bounds by default rather than because any instrument defines integration for them. The limit in the opposite direction is that the low bound counts as integrated 3.2 million people in plans that owe them no Medicaid service, so the honest reading is that the true figure sits far nearer the high bound than the low one. The last published measurement is July 2024; the closure of the demonstration on 2025-12-31 and the 2025 reclassification of plans holding about 110,000 enrollees roughly offset each other at the high end, but no source opened here restates the figure for 2026 and none is computed.

Regional breakdown No state level distribution of exclusively aligned enrollment could be confirmed. The Integrated Care Resource Center monthly tables, which publish that distribution by plan and by state, returned HTTP 403 on both the site root and the specific tables. The state level figures that were confirmed measure different things and cannot be combined into a breakdown: D-SNPs are offered in 47 states and DC as of 2026, PACE was available in 32 states and DC as of 2023, and the KFF single coverage arrangement share ranges from 34 percent in Rhode Island to below 1 percent in 32 states and DC on March 2021 data. Splitting the national residual by state population share was not done and would not be a measurement.