06-reference/research

state medicaid behavioral health denial rates by category

2026-10-17·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
behavioral-healthdenial-ratesmedicaid-managed-carefamily-venturercm

One state publishes Medicaid behavioral-health denial rates by service category, and it carries the first public ABA number: 9.31%, below the non-behavioral-health average in its own table

The question

Verbatim: "Does any state Medicaid agency, Medicaid MCO, or other state parity report publish behavioral-health denial rates BY SERVICE CATEGORY -- and does Virginia's pattern (SUD denials driven ~30% by preauthorization failure, not medical necessity) replicate in NY DFS, CA DMHC, IL or DE?"

Context: [[2026-10-04-behavioral-health-claim-denial-rate-primary-sources]] found exactly one primary source, and it covers commercial fully-insured lives only (16 carriers, 2.48M lives), excluding Medicaid managed care and self-funded ERISA plans. The hard number we had was drawn from the payer line that matters least to an Applied Behavior Analysis (ABA) or substance use disorder (SUD) provider. Acronyms used throughout: ABA = Applied Behavior Analysis; SUD = substance use disorder; MH = mental health; BH = behavioral health; M/S = medical/surgical; MCO = managed care organization; PA = prior authorization; RCM = revenue cycle management; EOB = explanation of benefits; EQRO = External Quality Review Organization; MCPAR = Managed Care Program Annual Report; T-MSIS = Transformed Medicaid Statistical Information System; TAF = T-MSIS Analytic Files; CARC = claim adjustment reason code; ASAM = American Society of Addiction Medicine; DFS = New York Department of Financial Services; DMHC = California Department of Managed Health Care; HFS = Illinois Department of Healthcare and Family Services; HSAG = Health Services Advisory Group; MACPAC = Medicaid and CHIP Payment and Access Commission; OIG = HHS Office of Inspector General.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The question had two halves and they resolve in opposite directions. The by-category reporting exists -- Louisiana publishes it quarterly, for Medicaid managed care, with Applied Behavior Analysis broken out as its own line, and has done so since roughly calendar 2019. That is a real answer, not a null, and it retires the parent brief's cleanest finding. There is now a public ABA denial rate. It is 9.31% across 109,991 adjudicated line-claims in a single quarter, and it sits below the 16.79% non-behavioral-health rate in the same table and below Primary Care - Pediatrics in the same table. The vendor claim of 15-30% ABA initial denials at "3-5x the medical average" is no longer merely unsourced, as the parent brief concluded. On the payer line that carries most ABA volume it is contradicted by the only public measurement of it. The venture premise should be treated as falsified on the ABA leg rather than open.

The cause half of the question resolves into something more useful than a yes or no. The Virginia preauthorization-driven mix does not look like a generalizable market fact, and the reason is that the cause mix is an artifact of which denial event you measure. Claim-denial datasets put administrative and authorization causes on top and medical necessity near zero; Virginia at 32.9% versus 1.4%, Illinois Medicaid at 24.04% versus 0.10%. Prior-authorization datasets put medical necessity on top and administrative near the bottom; Colorado at 86% versus 14%, New Hampshire at 95.4% versus 2.1%. Both are true of the same providers. They are counting different moments in the lifecycle. Any product pitch built on "most behavioral-health denials are authorization failures, not clinical refusals" is defensible only if the pitch also says which measurement it rests on, and the honest version is that no published dataset crosses a behavioral-health service category with a denial cause on the Medicaid line at all. Louisiana comes closest by accident: its behavioral-health prior-authorization denial rates run 0.0% to 6.1% at most ASAM levels of care while its behavioral-health claim denial rates run 9% to 15%, which is consistent with the authorization gate approving and the claim dying afterward. That is my inference from two tables in one report, flagged as such, not a published finding.

What this does to where the product has to sit is the most actionable result. The dominant term in the Medicaid data is not service line, it is payer. ABA denial rates inside one state program, one quarter, under one set of state rules, span 5.0% to 35.9% across six MCOs -- a 7.2x spread, larger than any behavioral-health-versus-medical gap anywhere in this sweep including Virginia's. The service-line framing that every competitor in [[2026-10-01-national-rcm-eob-ai-behavioral-health-scan]] markets on is the weaker signal; payer-specific adjudication variance is the stronger one. A product positioned on "behavioral health gets denied more" is positioned on a claim the Medicaid data contradicts. A product positioned on "your denial rate at this payer is multiples of your rate at that payer for the same service, and here is the rule you are missing" is positioned on something the data supports and that providers cannot see for themselves, because the comparison requires a book of business wider than any one practice. That is first-pass claim construction and payer-rule benchmarking, upstream of the EOB, not appeals or remittance parsing downstream of it. It also makes the claim lifecycle question concrete rather than philosophical: the leak is at authorization capture and payer-specific claim assembly, and the measurement that would price it is the one nobody publishes.

Two dated catalysts make this a live window rather than a settled null, and both cut in the venture's favor. MCPAR Topic XIII prior-authorization indicators became mandatory in June 2026, and 42 CFR 438.210(f) plan-website postings began March 2026. Both produce Medicaid prior-authorization denial rates per plan with no service-category and no behavioral-health dimension, because the Centers for Medicare & Medicaid Services declined that granularity on the record. So within a year there will be a universal public payer-level denial rate and still no public behavioral-health service-category breakdown -- precisely the shape of a defensible data asset. The proprietary-remittance test from [[2026-10-04-behavioral-health-claim-denial-rate-primary-sources]] therefore survives and gets cheaper: a multi-site ABA operator's remittance history can now be benchmarked against a published Medicaid comparator in Louisiana and a published parity comparator in Illinois, which was not possible two weeks ago. The first question to Michael Holzum is unchanged and the framing is better: not "is behavioral health denied more," which the public data answers no, but "what is your denial rate by payer, and how far is your worst payer from your best."

Why this is in the vault

This closes two of the three follow-ups left open by [[2026-10-04-behavioral-health-claim-denial-rate-primary-sources]] on the family co-founder venture thread, and it changes the go/no-go test rather than merely informing it: the ABA elevated-denial premise is now contradicted on the Medicaid line by Louisiana's Act 233 series, and the product thesis moves from service-line positioning to payer-variance benchmarking upstream of the explanation of benefits.

Open follow-ups

Related

Sources

Per-state and per-stream verdicts

Jurisdiction / stream BH denial rates by service category Cause mix published Payer lines covered Document
Louisiana YES -- ABA, BH-Residential, All Other Specialized BH, Behavioral Hospital, per MCO, quarterly NO -- zero denial-reason tables in the report Medicaid managed care only (6 MCOs) Healthy Louisiana Claims Report, Act 233, CY2025-Q2, Feb 2026
Illinois PARTIAL (strong) -- MH/SUD vs M/S x inpatient/outpatient/emergency x network status, statewide and per MCO NO in the parity report ("Data were not available"); a 7-reason mix exists in a separate hospital-claims report with no BH breakout Medicaid managed care (6 MCOs) HFS/HSAG 2023 2024 Mental Health Parity Analysis Summary Report, Dec 2024
New York NO from New York itself. Commercial-only DFS section 343 workbooks are Virginia-shaped but explicitly exclude Medicaid. The only published NY Medicaid BH rates by service come from a federal auditor NO -- the external-appeal cause axis is the statutory appeal grounds, and preauth/administrative denials are not externally appealable in NY Commercial fully-insured (DFS parity); Medicaid (OIG audit only) DFS MH/SUD parity reports + reporting instructions; OIG A-02-24-01011
California NO -- clean null on Medi-Cal. DMHC Behavioral Health Investigations exclude Medi-Cal by their own footnote and cite plans for not tracking BH denial rates PARTIAL, commercial only -- Covered California QHP filings split BH-only claims with a reason mix Commercial (Covered CA, DMHC BHI); Medi-Cal has no denial-rate publication DMHC BHI Phases 1-3; Covered California denied-claims filings
Delaware NO -- hard null on both lines; DMMA declined a request to require MCO claims-denial data NO -- the only administrative-vs-medical-necessity split published statewide is dental, one vendor, 120 denials Commercial market-conduct exam populations only (counts, not rates) Delaware Quality Strategy 2023; 2024 EQR technical report
Virginia (carried forward) YES by service type for MH/SUD/M/S + top-three-reason table YES -- the only jurisdiction with both Commercial fully-insured only -- excludes Medicaid and self-funded Virginia Bureau of Insurance 2024 report
CMS federal (MCPAR, T-MSIS/TAF, CMS-0057-F) NO. MCPAR has inpatient/outpatient BH indicators but for appeals, and reason and service type cannot be cross-tabulated. T-MSIS captures denials; TAF strips them. 438.210(f) is "aggregated for all items and services" NO for Medicaid. The ACA transparency PUF has a BH medical-necessity denial field, Marketplace only -- does not answer the Medicaid question Medicaid managed care (MCPAR); Marketplace (ACA PUF) 42 CFR 438.210(f); MCPAR appeals/grievances technical guidance, Aug 2024; TAF Claims RIF Codebook v1.8
EQRO reports NO -- near-universal null across 16+ annual technical reports (HSAG, IPRO, Comagine, Myers and Stauffer, MetaStar), including behavioral-health carve-out reports in Arizona and New York. Exceptions are statute-driven focused studies, not the 438.364 report PARTIAL in focused studies: New Hampshire 5-reason mix on BH drugs, Colorado 86/14 medical-necessity-vs-administrative on SUD Medicaid MACPAC Jan 2023 scan corroborates the null

Vault

Primary, verified at source by me (downloaded, extracted, tables read)

Primary, verified at source by sub-agents and not re-opened by me -- treat as one step removed

Research budget used: QMD 1 of 5 (lex + vec on vault priors). Zero WebSearch and zero WebFetch calls from the parent; the web sweep was fanned out to five parallel sub-agents (New York, California, Illinois+Delaware, CMS federal streams, EQRO cross-state), each running its own searches and retrievals, per the AGENTS.md rule on routing long artifacts away from parent context. The parent then independently re-retrieved and re-read the four load-bearing documents by curl plus pdftotext and recomputed every Louisiana figure used here, satisfying the primary-source gate. One sub-agent claim was checked and rejected: Louisiana's Table 1.9 is average days to adjudicate, not a denial-cause mix.

Not reached: Massachusetts MBHP behavioral-health-only EQR technical reports 2020-2025 and Pennsylvania HealthChoices behavioral-health annual technical reports (both HTTP 403); Virginia DMAS RD267's companion "Top Claim Denial Reasons by Provider Type" spreadsheet, which would convert Virginia's Medicaid side to a full behavioral-health rate plus reason mix; Illinois HFS quarterly QBR behavioral-health denial-reason data (collected, never published); New York OMH monthly and quarterly behavioral-health denial reports (collected, never published); roughly 20 New York plan-level behavioral-health focused-survey PDFs that are scanned images without a text layer; Texas UMCM 5.24.1 and 5.24.6 behavioral-health denial collections (collected, never published).