The Parity Test

Parity has been federal law since 2008. The Mental Health Parity and Addiction Equity Act says a health plan may not treat mental-health and substance-use benefits worse than it treats medical and surgical benefits — not through harder limits, and not through the softer machinery of prior authorization, network design, and medical-necessity review. What parity has never had is a scoreboard. The denial side of the promise has been effectively unmeasurable: in the public reporting that exists, a behavioral-health denial and a knee-surgery denial fall into the same undifferentiated bucket.

That is beginning to change, and this piece is a stake in the ground before it does. Two separate regulatory gears — one at the Departments of Labor, Health and Human Services, and the Treasury, one at CMS — are set to make behavioral-health denials countable on their own for the first time. Neither is fully settled. One is under an enforcement pause; the other is a proposal. We lay out exactly what each requires, where each stands, and what would become visible if both turn. We will update this page when the data lands.

The evidence of a parity gap has, until now, come from the access side rather than the denial side — from how often patients are pushed out of network. A 2024 study by RTI International economists, using one of the largest commercial-claims databases for 2019–2021, found that patients went out of network for a behavioral-health office visit 3.4 times more often than for a medical or surgical office visit; the gap was 8.9 times for seeing a psychiatrist.¹ The same study found in-network reimbursement was, on average, 22% higher for medical and surgical clinicians than for behavioral-health clinicians — a payment gap the authors tie directly to why behavioral providers stay out of network.¹

Those are access measures inferred from claims. What no public dataset has isolated is the denial itself — how often a behavioral-health claim is refused, and on what grounds, relative to a medical one at the same plan. The federal transparency file that ACA marketplace issuers already report does not distinguish between the two, so a 20% headline denial rate says nothing about whether behavioral care is denied at twice that or half it.² That is the specific blindness both new regimes are aimed at.

On September 9, 2024, the three Departments issued a final rule under MHPAEA, published in the Federal Register on September 23.³ Alongside detailed content requirements for the nonquantitative treatment limitation (NQTL) comparative analyses that plans must produce on request, the rule added something new: plans and issuers must collect and evaluate relevant outcomes data to assess whether an NQTL — such as prior authorization or a medical-necessity standard — produces materially worse access to behavioral care than to medical care.⁴

The rule names the number and percentage of claims denials as relevant data for that evaluation, and for network-design limitations points to in- and out-of-network utilization, network-adequacy metrics, and provider reimbursement rates.⁴ Where the data show a material difference in access, the plan must take reasonable action to address it. In the agencies’ framing, a demonstrated difference in access becomes a strong signal of a compliance problem — which makes the denial data itself, for the first time, an enforcement input rather than a footnote.

The requirement is not evenly timed, and it is currently on hold. The rule took effect November 22, 2024, with the general comparative-analysis provisions applying for plan years beginning on or after January 1, 2025, and the data-evaluation provisions delayed to plan years beginning on or after January 1, 2026.³ Then, on May 15, 2025, the Departments announced they would not enforce the provisions that are new in the 2024 rule — pending the outcome of litigation brought by the ERISA Industry Committee (filed January 17, 2025), plus an additional 18 months.⁵

Two things are worth holding precisely. First, the non-enforcement is a pause on the new provisions, not a repeal — MHPAEA’s underlying statutory duties, the 2013 rule, and the comparative-analysis mandate that Congress added in the Consolidated Appropriations Act, 2021 all remain in force.⁵ Second, a non-enforcement posture is not the same as a data vacuum: the outcomes-data expectation is written, plans that build to it will be generating denial-rate comparisons internally, and the pause has an end. This gear is real; its clock is uncertain.

The second gear touches the public number directly. Every year, KFF analyzes the transparency data that issuers of qualified health plans on HealthCare.gov must report to CMS. Its brief on 2024 data found that these plans denied 19% of in-network claims and, combined with out-of-network, about 20% of all claims — but noted that the file, as reported, cannot separate a behavioral-health denial from any other.²

In its 2024 marketplace brief, KFF flags the coming change: “CMS proposes to include requirements for issuers to report data separately for claims that were and were not related to behavioral health, and to report data regarding pre-service claims.”² The change is tied to plan year 2027 — the vehicle is the HHS Notice of Benefit and Payment Parameters for 2027, proposed in February 2026, with the final rule issued May 15, 2026.⁶ Because it is a proposal rendered final through that rulemaking, the precise contours of the behavioral split are exactly the kind of detail this page will confirm against the reporting instructions when the first data actually publishes.

The two gears do different work. MHPAEA’s outcomes-data duty operates plan-by-plan and inward — it makes a plan account for its own behavioral denial patterns to regulators. The CMS split operates market-wide and outward — it puts a behavioral denial rate into a public file anyone can read. Together, if both turn, they convert parity from a principle into a measured quantity.

For a provider or a billing team, the near-term operative facts are narrow. The behavioral split is a reporting change, not a coverage change — it does not, by itself, alter how any single claim is adjudicated. And the MHPAEA outcomes-data requirement is, for now, unenforced. What changes is the evidentiary landscape around a behavioral-health appeal: for the first time, the argument that a plan denies behavioral care at a rate out of step with its medical book may have a public number behind it rather than an inference.

Three dates carry the story forward. The ERIC litigation’s resolution sets the MHPAEA enforcement clock — until it lands, the outcomes-data duty is written but dormant. Plan year 2027 is when issuers are slated to report behavioral claims separately, meaning the first split file would surface in the transparency cycle that follows. And the first KFF-style analysis of that file is when a behavioral-health denial rate becomes a citable figure. This page is written before any of those has happened. We will treat it as provisional until the numbers exist — and revise it against the actual reporting instructions and the actual data, not the proposal.

For the specialty context behind these rules, see our behavioral-health denial guide and the coverage explainer on mental-health parity denials.