06-reference/research

behavioral health claim denial rate primary sources

2026-10-04·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
behavioral-healthdenial-ratesfamily-venturercmprimary-sources

A state regulator's table puts SUD denials at 24.9%, but on a pharmacy-heavy base; the "3-5x medical average" holds only for SUD office visits in that table; the ABA number has no primary source at all

The question

Verbatim: "Source the ABA and SUD claim-denial-rate claims (15-30% and ~25%) to a payer, state Medicaid, or peer-reviewed source instead of vendor marketing -- the family RCM/EOB venture premise rests entirely on that spread being real."

Context: [[2026-10-01-national-rcm-eob-ai-behavioral-health-scan]] could only source those two figures to vendor content pages, and flagged that every competitor in the lane markets against an elevated behavioral-health denial rate. The family co-founder venture is an open 3-6 year problem-space search, not a funded or decided plan; this brief tests one premise inside it. Acronyms used throughout: ABA = Applied Behavior Analysis; SUD = substance use disorder; MH = mental health; RCM = revenue cycle management; EOB = explanation of benefits; OMFS = oral and maxillofacial surgery; CMS = Centers for Medicare & Medicaid Services; ERISA = Employee Retirement Income Security Act; AMA = American Medical Association; DFS = New York Department of Financial Services; DMHC = California Department of Managed Health Care.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The premise is weakened, not collapsed, and the part that survives is narrower and more actionable than the part that died. What died is the headline arithmetic: there is no primary source for behavioral health denials running 3-5x the medical average, and the one state regulator that publishes the comparison in a single table shows a 2.4-point gap at the aggregate and a mental-health rate essentially at parity. What survives is a specific, well-evidenced claim: SUD claims in Virginia's 2023 fully-insured commercial data are denied at roughly 2.7x the medical/surgical rate once pharmacy transactions are excluded (one state, one year, in a series whose overall denial rate moved from 13.6% to 22.3% across three reports), and on the same non-Rx basis 32.9% of SUD denials are preauthorization failures against 5.0% on the medical/surgical side. That is a real operational asymmetry with a named cause.

The cause is the finding. Medical necessity accounts for 1.4% of categorized SUD denials in Virginia. The behavioral-health denial problem, as the only primary source that measures it describes it, is not payers refusing care. It is providers failing to obtain or document authorization and billing incorrectly, in a service line where preauthorization requirements are unusually dense. That reframes a product positioned here. An EOB-parsing and denial-appeal layer is aimed downstream of where the money leaks; the leak is upstream, at authorization capture and first-pass claim construction. It also reframes any sales story, because "we recover your denials" is a weaker pitch than "we stop you from generating the third of denials you cause yourself," and the second one is the one the primary data supports.

The appeal data makes the same point from the other direction. Virginia providers appealed 0.06% of denied claims; HealthCare.gov providers appealed under 1%. Any revenue model that assumes appeal volume as the unit of work is modeling an activity that nearly nobody performs. Rework and resubmission, which neither dataset measures, is where the labor actually goes. That is a gap worth naming rather than filling with an estimate.

Finally, the null result on ABA should be read as a market fact, not just a research failure. No public dataset reached in this pass reports claim denial rates for applied behavior analysis specifically. CMS transparency reporting cannot produce one by construction for 2024 data; KFF notes CMS will add a behavioral-health / non-behavioral-health split of claims received and denied for plan year 2027, which will give a national behavioral-health denial rate, though still not an ABA-specific one. No state Medicaid managed care denial reporting by service line surfaced. The nearest adjacent federal requirement, the 2024 CMS interoperability and prior-authorization rule (CMS-0057-F), has qualified-health-plan issuers publicly reporting prior-authorization metrics on their websites by 2026-03-31 (KFF), which is prior-authorization volume and outcome, not claim denials by service line. The peer-reviewed ABA literature that most wants a high number publishes none. This means every competitor in [[2026-10-01-national-rcm-eob-ai-behavioral-health-scan]] that markets against an ABA denial rate is asserting something unverifiable, and so would the family venture. For a 3-6 year horizon decision, the correct posture is that the ABA leg of the thesis is currently unevidenced in public data, and the only way to price it is proprietary access to a multi-site ABA operator's remittance history. Michael Holzum is an executive at ABA Centers of America ([[03-contacts/michael-holzum]]), a multi-site ABA operator whose own holding company shuttered an AI RCM effort before demo ([[hcls/problem-log]] 2026-09-16 row) and who was exploring a career change as of 2026-07-21. Its remittance history is the employer's data, not his; whether any of it could be examined, and under what terms, is the first question, and the debrief on why the prior effort died is the second. That is the cheapest next test available to the family group, if it ever tests this, and it is also the asymmetry that would make the venture defensible, since nobody selling into this market can see what that data would show.

Why this is in the vault

This narrows the single most load-bearing open item on the family co-founder venture thread flagged in [[2026-10-01-national-rcm-eob-ai-behavioral-health-scan]]: whether the elevated-denial-rate wedge every competitor markets on is real. It changes what a go/no-go on that venture would have to test, moving the question from "is behavioral health denied more" to "can we get remittance data from an ABA operator to price a lane no public dataset can price today."

Open follow-ups

Related

Sources

Vault

Primary

Peer-reviewed

Secondary / analyst

Vendor marketing (carried forward, still unsourced)

Research budget used: QMD 1 of 5 (lex + vec on vault denial-rate priors); WebSearch 3 of 3 (behavioral health denial rate primary/peer-reviewed; ABA/autism Medicaid managed care denial rate; KFF/CMS transparency and Medicaid MCO denial reporting); WebFetch 2 of 3 (Virginia 2024 PDF, extracted locally with pdftotext after the fetch returned raw PDF; KFF 2024 report). One Playwright retrieval for the PMC article after reCAPTCHA blocked WebFetch.

Not reached: state Medicaid managed care denial reporting by service line; parity reports from states other than Virginia; the JAMA Internal Medicine letter at source; any ABA-specific denial rate from any source of any class.