Medicaid Work Rule's 2.3M Exits Hit Self-Insured 2027 Assumptions

Judge Richard Stearns of the federal district court in Massachusetts declined on July 30 to pause the Medicaid community engagement rule, rejecting a preliminary injunction sought by 25 states and the District of Columbia, and the rule took legal effect the next morning (Healthcare Dive, July 2026). CMS projects the requirement will remove 2.3 million people from Medicaid in fiscal 2027, and a measurable share of them will land, sicker than average, on self-insured employer plans whose 2027 assumptions were set before any of this existed.

What Became Enforceable on July 31

The interim final rule, CMS-2454-IFC, implements the community engagement requirement Congress wrote into last summer's budget reconciliation law. Non-pregnant adults aged 19 through 64 in the Medicaid expansion group will need to document 80 hours per month of qualifying activity, employment, education, work programs or community service, or meet equivalent income thresholds, as a condition of eligibility. The rule carries exemptions for pregnant and postpartum women, medically frail individuals and caregivers, and it lets states request a temporary good faith effort exemption when their verification systems are not operationally ready (Foley Hoag, June 2026). Every state that expanded Medicaid, plus the District of Columbia, has to comply no later than January 1, 2027.

The timeline was compressed by design. CMS issued the rule June 1, published it in the Federal Register June 3, and set both the effective date and the close of the comment period on the same day, July 31 (CMS fact sheet, June 2026). The agency's own projection puts the enrollment effect at 2.3 million fewer Medicaid enrollees in fiscal 2027, rising above 3 million in later years. The Congressional Budget Office runs the longer arithmetic higher still: 5.3 million more uninsured by 2034 from the work requirement provision alone, the largest single Medicaid item in the law, carrying a $325 billion reduction in federal Medicaid spending over 2025 through 2034 (CBO, via Georgetown CCF, August 2025). Judge Stearns was not persuaded the states faced irreparable harm before a decision on the merits, so those numbers are no longer contingent on litigation for the 2027 plan year. Appeals will continue. Renewal calendars will not wait for them.

The Migration Channel: New Lives With Deferred-Care Profiles

The first channel runs through the plan's own enrollment file. A large fraction of the adults subject to the requirement already work; the coverage losses CMS and CBO project come substantially from documentation and reporting failures rather than from refusal to work. A worker disenrolled from Medicaid whose employer offers coverage has a qualifying life event and a strong incentive to take the employer plan, and that migration reprices three assumptions at once. Frequency rises because the plan is adding lives, many of them carrying diagnoses that went unmanaged while coverage churned. Severity rises because Medicaid-eligible populations carry higher chronic disease and behavioral health burdens than the average enrolled employee. The expected claim ratio for the 2027 plan year moves with both, on a cohort whose experience does not yet exist in any credibility database the plan's actuary can query.

The reserving effect is more specific than a general loading. Populations entering coverage after a gap front-load utilization: deferred imaging, delayed procedures, prescriptions restarted, specialist backlogs cleared in the first months of eligibility. For a self-insured plan that adds a disenrollment-driven cohort through 2027, the early development periods of the 2027 incurral year will run steeper than the completion factors fitted to 2024 through 2026 experience, and an IBNR estimate that applies historical completion patterns to the new mix will understate the liability precisely when the enrollment file says exposure is growing. The understatement also tends to surface late rather than early: the first valuation after the cohort arrives still leans on thin, immature paid data, so the miss shows up as adverse development in the second and third valuation cycles, after the reserve has already been booked and reported. That is the same mechanism, from the receiving side, that actuary.info worked through for the payer side in July: the managed care capitation reset, where the acuity of the residual Medicaid pool rises as healthier compliers exit, and the ACA morbidity load now appearing in 2027 rate filings. The self-insured book is the third destination for the same displaced morbidity, and the only one of the three whose actuaries mostly have not filed anything yet.

The migration will not arrive as a January 1 block, either. Disenrollments begin when states start enforcing, and most states will process them through existing redetermination cycles, which spreads the exits across the calendar year. Each disenrollment is a qualifying life event that opens a 60-day special enrollment window on the employer side, and many of the affected workers then face a plan waiting period before coverage starts. Stack those lags and the employer plan's exposure accretes month by month from February onward rather than resetting once at renewal. Per-member-per-month claim costs, aggregate corridor accrual and lag-triangle exposure bases all inherit that partial-year shape, and a reserving model that treats 2027 membership as a single annualized figure will misstate both the denominator and the development pattern in the same direction.

Setting an Expected Claim Ratio With No Experience to Rate

The entering cohort has no claims history any credibility formula can use, which forces the 2027 expected claim ratio selection onto proxies, and the available proxies disagree with each other. Medicaid managed care encounter data describes the cohort's morbidity but at Medicaid reimbursement rates, which run well below the commercial and self-funded allowed amounts the plan will actually pay for the same services; grossing encounter-based cost up to commercial fee schedules is a repricing exercise with its own error bar. Manual rates built from commercial books describe the right fee schedule applied to the wrong population. The defensible selection brackets the answer instead of pretending precision: a baseline ECR carried from the existing population, a loaded scenario that applies a first-year morbidity and deferred-care factor to the projected entrants, and an explicit statement of which enrollment assumption moves the result most. That structure also survives the audit and stop-loss questions that follow a bad year, because it shows the range was considered before the experience arrived, not reconstructed afterward.

The Second Channel: Uncompensated Care Repriced Into Commercial Rates

Not every disenrolled worker lands on an employer plan; CBO's 5.3 million figure is, by definition, people who end up with nothing. Their care does not stop, it becomes uncompensated, and hospitals recover uncompensated care the way they always have, through the commercial rates they negotiate with the payers that can pay. The AFL-CIO, working from analyses prepared for health plans during the reconciliation debate, put the pass-through at $182 to $485 of added annual cost per person across the 179 million people with employment-based coverage (AFL-CIO, 2025). The width of that range is itself the actuarial point: nobody pricing a 2027 self-insured renewal can currently distinguish a 40-basis-point trend add from one three times that size, because the split between channel one and channel two depends on employer offer rates, state implementation behavior and take-up decisions that have not happened yet.

The two channels also compound rather than offset. A plan whose workforce includes formerly Medicaid-enrolled employees absorbs channel one directly in its own claims, then pays channel two again through negotiated hospital rates that spread everyone else's uncompensated care across the commercial book. There is no experience credit anywhere in that transfer. The 9.9% medical trend Segal published for 2027, a 15-year high before any of this arrives, is the baseline the shift lands on top of, not a number that already contains it: the survey's respondents reported their assumptions before the rule survived its first court test.

The two channels run on different clocks, which is easy to miss when both get summarized as one trend add. Channel one hits claims as soon as the new lives enroll and start clearing deferred care, inside the 2027 plan year. Channel two moves at the speed of hospital contracting: uncompensated care absorbed in 2027 gets recovered through commercial rate negotiations that run on one-to-three-year cycles, so its trend effect lands mostly in 2028 and 2029 renewals. A sponsor that prices 2027 for the enrollment shift and then treats the resulting experience as the new normal will walk into the second channel's rate effect a year later with the same surprise twice.

Stop-Loss Attachments Set on a 2026 Census

Specific and aggregate attachment points for 2027 are being quoted right now, off 2026 census files and 2026 claim experience. A plan that adds even three to five percent more lives mid-year, concentrated in a cohort with above-average expected claims, erodes its aggregate corridor from both directions: the attachment was sized to a smaller exposure base, and the per-life expected claims of the additions run higher than the base rate the corridor assumed. On the specific side, a cohort carrying unmanaged chronic conditions and first-year catch-up utilization is disproportionately likely to produce claims that reach a fixed specific deductible, the leveraged mechanics our flat-deductible analysis traced for ordinary trend, applied here to a mix shift instead. Stop-loss underwriters saw the same rule effective date everyone else did, and 2027 quotes that arrive with enrollment-growth interrogatories or mid-year census re-rate provisions are the market pricing this in. A plan sponsor who signs a 2027 stop-loss contract without asking how mid-year Medicaid-migration enrollment is treated has left the question to the carrier's claims department, which is the most expensive place to discover the answer.

State Implementation Spreads the Exposure Unevenly

The rule sets a floor, not a uniform rollout. States choose how aggressively to verify, how to operationalize the medically frail exemption, whether to seek the good faith effort exemption, and how work-rule checks interact with the redetermination cadence they already run, choices that determine whether disenrollment arrives as a January cliff or a rolling attrition through 2027. KFF is tracking state-by-state implementation, and the variation already visible in state readiness means a self-insured plan with employees in a dozen expansion states holds a dozen different exposure profiles under one set of plan assumptions. Multi-state sponsors in retail, food service, home care and logistics, the industries with the heaviest overlap between hourly workforces and Medicaid enrollment, sit at the top of that distribution. A single national enrollment-growth assumption is the wrong shape for this risk even if its level turns out right.

The sponsor's own contribution design then decides how much of the exposure arrives through which channel. A worker losing Medicaid weighs the employee premium contribution against going uninsured, and for low-wage workers that contribution has historically suppressed take-up even when coverage is offered. A plan with rich employer contributions will pull more of the displaced population into channel one, buying claims exposure directly; a plan whose employee share prices those workers out pushes them into channel two, where the cost returns through hospital rates without ever appearing on the enrollment file. Neither design avoids the cost. The contribution schedule just chooses which line of the renewal it shows up on, and dependent tiers make the same choice again for spouses and children.

The Renewal File Between Now and January

Five months separate the rule's effective date from the state implementation deadline, and most of the 2027 self-insured renewal decisions in between will be made on experience data that contains none of this. The renewal file that holds up is the one that shows its arithmetic: an explicit enrollment-growth scenario tied to the sponsor's own wage and state mix, an expected claim ratio revised for the entering cohort rather than carried forward, completion factors flagged for the steeper early development a deferred-care population produces, and a stop-loss placement that answers the mid-year enrollment question in the contract rather than in the claim file. The watch items are concrete: the appeal of the Stearns ruling, which states request good faith effort exemptions before January, and the first quarterly enrollment prints of 2027, which will show whether CMS's 2.3 million or a different number is the one the receiving plans actually absorb. The plans that modeled the range will adjust a parameter. The plans that assumed zero will call it an unforeseeable shock, and it was not.