UnitedHealthcare's preliminary 2027 Medicare Advantage exit list covers 34 counties across 12 states and more than 20,000 members, a fraction of the 109 counties and 180,000 members it cut for the current plan year (Modern Healthcare, August 2026). Presbyterian Health Plan is dropping roughly 30,000 members and Humana 600,000.

A Medicare Advantage plan does not fail nationally. It fails county by county, on the same benchmark and rebate arithmetic a bid actuary runs every June.

Key Takeaways

  • County benchmarks are set in four quartiles of local fee-for-service spending: 95%, 100%, 107.5% and 115%. A bid above the benchmark earns no rebate and cannot bill the difference back to CMS.
  • A star-tier downgrade costs more than the bonus. Dropping from the 65% rebate share to 50% forfeits 15 percentage points of every dollar of benchmark-to-bid spread in every county the contract serves.
  • The aggregate picture is generous: average benchmarks run 124% of fee-for-service, and CMS paid plans an estimated $76 billion more in 2026 than the same beneficiaries would have cost in traditional Medicare.
  • Presbyterian's Medicare Advantage book lost more than $59 million in 2025, with no national risk pool to absorb it, which is why a provider-sponsored plan exits where a diversified carrier absorbs another cycle.
  • CY2027 excludes diagnoses from unlinked chart reviews and audio-only telehealth, which bites hardest in the low-density counties already sitting in the lower benchmark quartiles.

The Three Confirmed Retreats

UnitedHealthcare's list spans Minnesota, Nebraska, Wyoming, Montana and eight other states, and remains preliminary until CMS's October 1 plan-finder release. Its 2026 list also started smaller and grew before finalization, ending at 109 counties across 16 states and 180,000 members. "We need a model that is sustainable and allows us to bring care to folks in those areas in a cost-effective way," said Bobby Hunter, UnitedHealth Group's CEO of government programs (Reuters via Investing.com, October 2025). Total enrollment fell by roughly 1.1 million over 2026.

Presbyterian Health Plan is discontinuing most of its Medicare Advantage business for 2027 and cutting 150 administrative jobs, affecting roughly 30,000 members while retaining a Dual Plus Special Needs Plan covering 13,000. The book lost more than $59 million in 2025, and CEO Rishi Sikka tied the decision to "the loss profile last year" (Healthcare Finance News, June 2026). A provider-sponsored plan has no national risk pool to smooth a county shortfall against.

Humana's exit is the largest by count: plans covering roughly 600,000 members, about 8% of its 7.2 million base, with an expected recapture of 240,000 at its historical 40% rate. CFO Celeste Mellet described the cuts as trimming "the lower tail of profitability and return," a framing that fits the county arithmetic better than a member count does. The recapture-selection side of that move is a separate question from which counties stop clearing a bid.

Two Levers Decide Whether a County Clears

Every bid is judged against a county-specific benchmark, the maximum CMS will pay for an average-risk enrollee there, set as a percentage of local fee-for-service spending. CMS ranks counties by per-capita FFS cost into four quartiles: the highest-spending quartile gets 95% of FFS, the middle two get 100% and 107.5%, and the lowest-spending quartile gets 115% (Congressional Research Service, 2026). Bid below the benchmark and the plan keeps a rebate. Bid above and it gets nothing.

Rural counties sit in the low-FFS quartiles, which reads as generous on paper. The problem is the denominator. In a county with a few thousand members, one high-cost enrollee or a modest utilization swing moves actual experience further than the benchmark tolerance allows.

MechanismValue / RuleSource
County benchmark quartiles (% of local FFS)95% / 100% / 107.5% / 115%Congressional Research Service, 2026
CY2027 benchmark rebasing window2020–2024 claims (drops 2019)CMS, April 2026
CY2027 net average payment change2.48% (4.98% incl. risk-score trend)CMS, April 2026
CY2026 net average payment change (comparison)5.06%CMS, 2026
Star-rating rebate share, <3.5 / 3.5–4.5 / ≥4.5 stars50% / 65% / 70% of benchmark-bid gapKFF, 2026
MA membership in 4+ star contracts, 2026 vs. 202563.5% vs. 64.1%Healthcare Dive, 2026
National MA benchmark average, 2026124% of FFS ($76B above FFS spend)MedPAC, cited by actuary.info

The CY2027 rebasing rolled the claims window forward, dropping 2019 and adding 2024, with one year of transition before a reclassified county's lower benchmark applies in full. The finalized rate announcement projects a 2.48% net average increase, or 4.98% including risk-score trend, down from 5.06% for CY2026 (CMS, April 2026). A national average near 2.5% still leaves individual counties with negative benchmark movement.

The second lever is the star-rating rebate share: the portion of the benchmark-to-bid gap a plan keeps to fund supplemental benefits. Statute sets it at 50% below 3.5 stars, 65% from 3.5 to 4.5, and 70% at 4.5 and above (KFF, 2026). A contract falling from 65% to 50% loses 15 percentage points on every dollar of spread, which in most markets exceeds the lost bonus.

That compression is happening at scale. CMS's 2026 star ratings raised the 4-star cut point on roughly 60% of measures, and membership in 4-star-or-better contracts slipped to 63.5% from 64.1%, while the bonus-qualified share fell to 68% from 75%, the lowest since 2018, even as bonus spending rose to $13.4 billion from $12.7 billion. Elevance's suit over $115 million of disputed 2026 bonus dollars shows how one measure removal moves a rebate tier across every county a contract serves.

The exit follows from what the rebate funds. Fewer rebate dollars means thinner dental, vision, hearing and cost-sharing benefits. A thick urban market absorbs that and still fields a competitive package. A county with a few thousand members faces a benefit package too thin to retain enrollment or a bid too rich to clear, which is where a plan leaves rather than compete on economics it cannot win. That aggregate benchmarks run 124% of fee-for-service nationally is what makes these county-level exceptions worth reading closely.

The Risk Score Compresses in the Same Counties

A third mechanism lands on top, through the model that converts a county's enrollee mix into the score the bid is paid against.

The CMS-HCC V28 model reached full phase-in for the current year after running one-third weighted in 2024 and two-thirds in 2025, and CMS declined to refresh its calibration for CY2027, retaining coefficients built on 2018 diagnoses and 2019 expenditures to "allow the MA market more time to adjust." The revenue exposure that created ahead of the June bid deadline is separate from the two narrower exclusions CMS did finalize: diagnoses from unlinked chart reviews and from audio-only telehealth no longer count.

Both exclusions concentrate where the exits are. Rural and low-provider-density counties lean on telehealth, including audio-only visits, to close annual wellness and diagnosis-capture gaps, and those are disproportionately the counties in the lower benchmark quartiles where each risk-score point carries more weight in the bid.

The pattern is selective rather than a retreat. Oliver Wyman describes carriers moving into a "portfolio management posture" for 2027, trimming counties and curating networks rather than leaving the program, with scaled plans concentrated in high-benchmark urban markets realizing 3% to 5% efficiency gains against a dispersed footprint (Oliver Wyman, March 2026).

What it leaves behind is a pool nobody has priced. Three carriers' exits plus the 180,000 members already displaced for 2026 land inside the same October 15 to December 7 window, and a plan that gains enrollment because a competitor left a shared county inherits a population sorted by who shopped and who was passively crosswalked. That is the same selection mismatch that decides whether Humana's 240,000 recaptured members skew healthier or sicker than the 360,000 it loses, arriving in books whose 2027 bids were built on 2026 experience data from a different population.

Further Reading

Sources