Why Behavioral Health Claim Denials Keep Rising - and How Payers' AI Is Making It Worse

If your denial rate is creeping up year over year, the instinct is to look inward: a biller who left, a coding habit that slipped, a clinician behind on notes. Look harder before you reorganize the team. The most important change of the last three years didn't happen on your side of the table. It happened on the payer's.

Behavioral health claim denials already run 15–25% nationally, roughly two to three times the 5–10% rate medical specialties see, and the trend line is pointing up. The reason isn't only that behavioral health is hard to bill (it is). It's that payers have industrialized the denial itself. Claims and prior-authorization requests that used to land on a human reviewer's desk now pass through automated systems that adjudicate at machine speed and scale. Some reject in roughly a second. Some reject without a clinician opening the file at all. For a treatment center running thousands of claims a month across IOP, PHP, residential, and outpatient, that shift is the difference between an A/R you can staff and one you can't.

This guide is for the RCM lead who needs the honest picture: how bad denials actually are in behavioral health, the documented evidence that payer automation is driving the increase, the balanced distribution of denial causes you can still control, what the rework actually costs your center, and how to fight automated denials without burning out your team. Evidence-led, not conspiratorial. Automation isn't the villain. One-sided automation is the problem, and the durable answer is to meet it on equal footing.

Sources: KFF - Claims Denials and Appeals in ACA Marketplace Plans (2024) · ICANotes - Behavioral Health Billing Metrics & KPIs (2025)

How Bad It Is in Behavioral Health

Start with the baseline, because the gap between behavioral health and the rest of medicine is the whole story.

A clean general-medical practice runs a 90–95% clean-claims rate and a 5–10% denial rate. Behavioral health sits well below that on both. Outpatient psychotherapy lands around 78–85% clean and 15–20% denied. Move up the level-of-care ladder and it gets worse. IOP and PHP, the higher-revenue, prior-auth-heavy services a treatment center depends on, routinely deny at 20–30%. Substance-use and dual-diagnosis claims trail close behind, because medical necessity for continued stay gets challenged constantly.

Sources: SimiTree - Behavioral Health Clean Claims Rate · Supahealth aggregate data from 200+ behavioral health practices.

Two things matter here. First, behavioral health carries a structurally higher denial rate before anyone touches automation. Parity enforcement is uneven, level-of-care billing is complex, and payers scrutinize these claims harder. Second, the trend is moving the wrong way. Industry trackers including Experian's State of Claims and HFMA's denial reporting have flagged denial volumes climbing year over year across healthcare, and behavioral health rides at the top of that curve. A center that held steady at 18% denials two years ago is, in many cases, watching that number drift toward the mid-20s without changing a single internal process.

That last point is the one most billing teams underestimate. When your denial rate rises while your processes stay constant, the variable that changed is external.

The New Driver: Payers Are Industrializing Denials With AI

Here's what changed on the other side of the table.

Payers have invested heavily in automation to review and adjudicate claims and prior-authorization requests. The documented record shows that automation operating at a speed and scale no human review department could match. This isn't speculation. It's in court filings, investigative reporting, and a Senate committee report.

The clearest single example is Cigna's "PxDx" system. According to ProPublica's investigation, the system let Cigna's doctors reject more than 300,000 claims over a two-month period, spending an average of about 1.2 seconds on each, with denials issued without opening the patient file. One physician described the workflow as clicking and submitting in batches: "We literally click and submit… it takes about 10 seconds to do 50 at a time." When a claim can be denied in roughly the time it takes to read this sentence, the economics of the appeal flip entirely against the provider.

Then there's nH Predict, the algorithm UnitedHealth acquired with naviHealth and used to guide post-acute care decisions. The lawsuit Estate of Gene B. Lokken v. UnitedHealth Group, filed in November 2023, alleged a roughly 90% error rate. The suit claims about nine in ten of the denials the model drove were reversed when patients appealed. Phrase that carefully: it's an allegation in active litigation, not an adjudicated fact. But a 90% reversal rate, if it holds, describes a system that wasn't built to be right. It was built to deny, knowing most denials are never challenged.

That "never challenged" assumption is the engine. The Senate Permanent Subcommittee on Investigations released a majority-staff report in October 2024. It found that major Medicare Advantage insurers (UnitedHealth, Humana, and CVS/Aetna) used AI and algorithmic tools to deny prior authorization for post-acute care, with denial rates rising as the automation scaled. The pattern is consistent across all three examples: automate the denial, issue it fast, and rely on the fact that most providers and patients won't fight back.

The downstream cost lands on clinical care, and the American Medical Association has measured it. In the AMA's annual prior-authorization physician survey (roughly 1,000 practicing physicians), 95% reported that prior authorization causes care delays, 79% said it leads patients to abandon treatment, and 26%, more than one in four, reported that a prior-authorization delay led to a serious adverse event for a patient in their care. For behavioral health, where treatment continuity is the intervention, a delay isn't an inconvenience. It's a clinical setback.

Sources: ProPublica - How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them · STAT - UnitedHealth used a deeply flawed AI algorithm to deny care, lawsuit alleges · AMA - Prior Authorization

One honest counterweight before moving on: automation is not inherently bad. The same systems that deny fast can also approve legitimate claims fast, and not every denial is wrong. Plenty are correct catches of real coding or eligibility errors. The problem isn't that payers use AI. It's the asymmetry. Payers adjudicate at machine scale while most providers still appeal by hand, one claim at a time. That imbalance is what's pushing denial rates up, and it's the imbalance a treatment center has to close.

Why Behavioral Health Gets Hit Hardest

Automated adjudication hits every specialty, but it hits behavioral health harder, for three structural reasons.

Parity enforcement is uneven. The Mental Health Parity and Addiction Equity Act requires payers to cover behavioral health on par with medical care, but enforcement is inconsistent and payers apply tighter utilization management to behavioral claims than the law arguably allows. That tighter management is exactly the kind of rule-based scrutiny that automates well, which means behavioral claims pass through more automated review gates than a comparable medical claim.

Prior authorization is denser. Behavioral health, and especially the level-of-care services a treatment center runs, lives and dies on prior auth. IOP, PHP, residential, and SUD treatment require initial authorizations, concurrent reviews, and continued-stay justifications at every step. Every one of those touchpoints is a place an automated system can say no. A medical practice billing office visits has a handful of auth triggers. A residential SUD program has dozens per patient per episode.

Level-of-care billing invites concurrent review. When a payer's algorithm can pull a patient's length of stay, last documented assessment scores, and ASAM criteria and compare them against a coverage threshold, it can flag a continued-stay claim for denial automatically. The clinical reality is that recovery isn't linear and a plateau isn't a reason to discharge. That doesn't fit neatly into a threshold check. So the denial fires, and your team is left documenting medical necessity after the fact to a system that already decided.

Put those together and behavioral health is the specialty most exposed to automated denial: more review gates, more auth touchpoints, and clinical judgment that resists the threshold logic automation runs on.

The Real Distribution of Denial Causes

Now the balance correction, because it would be easy to read everything above and conclude that payers cause 100% of denials. They don't. Payer automation is the new accelerant, but the underlying causes of denial are distributed across your whole revenue cycle. Most of them are things you can influence.

Most denial articles pin everything on one cause, usually documentation, because that's the cause the article's product happens to solve. The honest picture is messier. KFF's 2024 analysis of in-network ACA marketplace denials found the breakdown skewed heavily toward "other/unspecified" (36%) and administrative reasons (25%), with prior auth/referral and medical necessity making up smaller named shares. Those are healthcare-wide numbers. For behavioral health, prior auth, medical necessity, and documentation all weigh heavier.

Sources: KFF - Claims Denials and Appeals in ACA Marketplace Plans (2024) · HHS OIG - Prior Authorization Denials in Medicaid Managed Care

Two takeaways drive everything that follows. First, documentation is one of several big buckets, not the only one. Combined medical-necessity and documentation denials typically run 20–35% of total denials. Meaningful, but not 60%-plus. Second, the buckets have different owners. Prior auth is a front-desk-plus-clinician problem. Coding errors are a biller problem. Eligibility is a verification problem. Documentation is a clinician problem. Cut your denial rate in half and you'll find you had to fix all of them at once. Pour all your energy into one bucket and the improvement caps fast, because the other six keep firing.

This is why the payer-AI story and the internal-causes story aren't in tension. Payer automation raises the cost of every denial by making them faster and harder to appeal. Your internal process determines how many denials you hand the payer's algorithm to act on. You can't change the payer's speed. You can change how many clean claims you feed it.

The Rework Tax at a Treatment Center

Every denial carries a rework cost, and at treatment-center volume that cost compounds into a line item that should have its own budget.

HFMA's analysis puts rework at roughly $48 per Medicare Advantage denial and $64 per commercial denial. That's staff time to research, correct, resubmit, or appeal, even when the claim eventually pays. Industry estimates for behavioral health land in a similar $25–$70 range per reworked claim. For a center submitting 4,000 claims a month at a 20% denial rate, that's 800 denials a month. At $50 average rework, you're spending $40,000 a month, roughly half a million dollars a year, just to recover money you already earned. That's before counting the cash that ages out and never comes back.

Sources: HFMA - Navigating the Rising Tide of Denials · KFF - Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024

The KFF numbers in that table are the quiet scandal. In Medicare Advantage, only about 11.7% of denied requests are appealed, but 81.7% of those appeals succeed. Read that twice. More than four out of five appealed denials get overturned, which means the original denial was wrong four out of five times. And yet nearly 88% of denials are never appealed at all. They age past timely-filing windows and get written off. The payer's automation isn't betting that its denials are right. It's betting that you won't fight them. At a center, that bet pays off because your team physically cannot hand-appeal every denial. There aren't enough hours.

This is the asymmetry made concrete. The denial costs the payer a fraction of a second to issue. It costs you $50 and 45 minutes to overturn. Multiply by 800 a month and the only rational response is to stop the denials upstream, not chase them downstream.

What You Can't Control vs. What You Can

Sorting denials into "controllable" and "not" is the single most clarifying exercise a treatment-center RCM lead can run. It tells you where to spend and where to stop spending.

What you can't control:

• Payer policy and coverage rules. When Aetna moves a code to non-covered or tightens a medical-necessity threshold, you don't get a vote. You can catch the change and adapt, but you can't make the payer pay.
• The payer's adjudication speed and automation. PxDx-style systems aren't going away. Prior-auth algorithms aren't going away. The Senate report and ongoing litigation may shape the rules eventually, but not on your billing-cycle timeline.
• Whether a clinically valid continued stay matches a coverage threshold. Sometimes the algorithm says no to care that's genuinely necessary. You fight that with documentation and appeals, not by preventing the denial first.

What you can control:

• Whether eligibility and benefits are verified before the visit. This is fully yours.
• Whether prior auth exists, covers the right code and modality, and is current. Yours.
• Whether the claim is coded correctly: CPT, modifier, POS, time, bundling. Yours.
• Whether the note establishes medical necessity with severity, functional impairment, and progress measures. Yours.
• Whether the claim is filed inside the payer's window. Yours.
• Whether you catch payer bulletins before the policy change shows up as a denial spike. Yours.

The strategic point: you can't out-argue an algorithm one denial at a time, but you can shrink the surface area you expose to it. Every clean claim is one the algorithm has nothing to flag. The goal isn't to win more appeals. It's to give the payer's automation fewer legitimate reasons to fire in the first place.

Fighting Automated Denials: The Upstream Prevention Playbook

Prevention is a sequence, not a single tool. Skip a step and the others can't cover for it. Here's the playbook, ordered by where it intercepts the denial.

1. Verify eligibility and benefits in real time, repeatedly. Check at scheduling, again 24–48 hours before the visit, and again at check-in. Re-verify the whole panel in January and July when plans turn over. For auth-heavy services, confirm not just that an auth exists but that it covers the exact code, modality, and date range you're about to bill. This closes most eligibility and a chunk of prior-auth denials before they exist.

2. Capture medical necessity in structured documentation. Every note that supports a billable session should carry presenting problem, severity with numbers (PHQ-9, GAD-7, functional ratings), functional impairment, treatment-plan linkage with the auth reference, a progress metric, and start/stop time for time-based codes. Real-time gap alerts at note signing fix the problem in the moment, not three weeks later when the EOB arrives. This is the documentation lever: one bucket, not all of them.

3. Scrub every claim pre-submission against payer and state rules. A scrubber holding current payer-by-payer and state-by-state guidelines catches the errors automation loves to deny: "Time documented is 47 minutes, bill 90834, not 90837." "Texas Medicaid moved telehealth to POS 10, so POS 02 will deny." "Aetna requires modifier 95 here; it's missing." This is where you stop feeding the payer's algorithm easy rejections.

4. Make same-day or next-day note completion a hard standard. This closes most timely-filing leakage and pulls days-in-A/R down by 5–10 days. It's the cheapest denial-prevention move available, and it costs nothing but discipline.

5. Monitor the top five payers' bulletins. A 15-minute weekly review catches the policy changes that drive surprise denial spikes. You can't control the change, but catching it early turns a month of denials into a one-time adjustment.

6. Appeal selectively and route smartly. For the denials that still land, route by code and payer, appeal with matched clinical context, and track resubmission cycles. Given that ~82% of appealed MA denials succeed, the bottleneck isn't whether to appeal. It's having the capacity to appeal at all. That capacity problem is the one automation on your side actually solves.

Sources: HFMA - Navigating Medical Necessity Denials · CMS - Prior Authorization and Pre-Claim Review Program Statistics, FY 2024

AI / Agentic Systems in Denial Management

AI in revenue cycle management has moved past "better OCR." What's emerging now is agentic systems that operate full workflows end-to-end, with human review at the edges. For denials, that shift matters more than for almost any other RCM problem. As the cause distribution above shows, a denial is rarely a single-point failure. It traces back to a benefits check that missed a session limit, a code that didn't match the documented time, a note that didn't establish medical necessity, or a payer rule that changed last quarter. To stop denials, you have to fix all of those places at once, continuously. And you have to do it at the speed the payer's automation already runs.

That's the logic behind Supabill. It isn't a single denial tool. It's an integrated system of behavioral-health-trained agents working together, 24×7, across the whole revenue cycle. The point is symmetry: if payers adjudicate with automation, the provider side needs automation too, or the asymmetry never closes.

Here's how the pieces reinforce each other. The benefits verification agent pulls eligibility before the first visit. It works across 5,000+ payers, operating payer portals directly, and even places a voice call to a payer rep when a portal doesn't surface what's needed. It flags session limits, prior-auth requirements, and copay structure. When it learns a payer caps psychotherapy at 20 sessions a year, the claims agent starts flagging that patient's submissions as they approach visit 18, before an avoidable denial fires. The claims agent holds state-by-state and payer-by-payer rules in its core database and scrubs every claim pre-submission: time-vs-code, modifier, POS, prior-auth match, bundled-code conflicts. Whatever still gets denied flows to the denials management agent, which routes by code and payer, drafts appeals against matched clinical context, and tracks resubmission cycles, so the ~82% of appeals that would succeed actually get filed instead of aging out. And when that agent learns a payer is rejecting a code for a specific missing element, the signal feeds back into the documentation guardrails in Supanote, so the note prompts for that element next time. Each agent makes the others smarter: a denial reason learned once becomes a denial prevented everywhere.

Honest limits. None of this overrides a payer's policy. When Aetna moves a code to non-covered, the system catches the change and stops submitting, but it can't make Aetna pay. It can't make a wrong diagnosis right or turn a clinically thin session into a billable one. It won't rescue a broken clinical workflow. What an integrated system does fix is the high-volume, rules-based, cross-team work humans get wrong simply because there's too much of it to hold in working memory across every payer, every code, and every state. That's exactly the surface area the payer's automation is built to exploit.

If you're interested, book a demo here to learn more.

Quick Wins to Lower Your Denial Rate

Things a treatment-center RCM lead can start this week.

1. Audit your last 50 denials and bucket them into the seven causes. The shape tells you where to invest, and it's usually not the bucket you assumed. Centers braced for a documentation problem often find prior auth and eligibility leading.

2. Re-verify eligibility 24–48 hours before every visit. If your system can't automate it, batch-check tomorrow's schedule each afternoon. This single workflow change closes most eligibility denials.

3. Pull your appeal rate and your appeal success rate. If you're appealing far fewer than you win (and most centers are), you're leaving overturnable money on the table. That gap is a capacity problem, and it's the clearest case for automating appeals.

4. Standardize a note template with required medical-necessity fields: presenting problem, severity with numbers, functional impairment, plan plus auth reference, progress metric, start/stop time. Required, not optional.

5. Enforce same-day or next-day note completion as a hard standard. Closes most timely-filing leakage and tightens A/R by a week or more.

6. Subscribe to bulletins for your top five payers and review them 15 minutes a week. It's the cheapest defense against surprise denial spikes from policy changes.

FAQ

Q: Are payers really using AI to deny my claims, or is that just hype?

A: It's documented, not hype. ProPublica reported Cigna's PxDx system rejected 300,000+ claims in two months at about 1.2 seconds each. A lawsuit alleges UnitedHealth's nH Predict algorithm carried a roughly 90% error rate on the denials it drove. The Senate Permanent Subcommittee on Investigations found in 2024 that major Medicare Advantage insurers used AI to deny prior auth for post-acute care. What you should not assume is that every denial you get came from an algorithm. Many are routine, and some are correct catches of real errors. The accurate framing is that automation has made denials faster, cheaper for payers to issue, and harder for you to appeal.

Q: My denial rate went up but my processes didn't change. What happened?

A: That's the tell. When internal processes hold constant and denials rise, the variable that changed is external: tighter payer rules, more automated review gates, or a policy change you didn't catch. Audit a sample of the new denials by cause and payer. If they cluster around a specific payer or code, you're likely seeing a policy or automation change on their side, not a regression on yours.

Q: Which denials are actually worth appealing?

A: Appeal where the math works and the documentation supports you. In Medicare Advantage, ~82% of appealed denials succeed, so the merit usually favors appealing. The real constraint is staff capacity. Prioritize high-dollar denials (PHP, IOP, longer-session codes), administrative and medical-necessity denials with strong supporting notes, and any payer where you see a repeating pattern worth batch-appealing. Stop chasing $40 reimbursements that cost $60 in rework unless they reveal a fixable systemic issue.

Q: How does denial rate vary by payer - Medicare Advantage vs. commercial vs. Medicaid?

A: Medicare Advantage denies prior auth around 7.7% nationally, ranging from 4.2% (Elevance) to 12.8% (UnitedHealth Group) per KFF, and it's the fastest-growing denial source. Commercial PPO behavioral health typically runs 12–20% depending on the carve-out. Medicaid is bimodal: state-direct often pays cleanly, but Medicaid MCOs can deny at 20–30% for higher levels of care. Track per payer; your action plan depends on which dominates your mix.

Q: Why does behavioral health get denied so much more than medical?

A: Three structural reasons. Parity enforcement is uneven, so payers apply tighter utilization management to behavioral claims. Prior authorization is far denser for level-of-care services: IOP, PHP, residential, and SUD carry initial auths, concurrent reviews, and continued-stay requirements at every step. And level-of-care billing invites threshold-based concurrent review, which automation handles easily and clinical reality resists.

Q: If payer automation is the driver, is there any point fixing my internal process?

A: Yes. It's the whole game. You can't change the payer's adjudication speed, but you control how many denials you hand their algorithm to act on. Every clean claim is one the system has nothing legitimate to flag. Prevention doesn't beat the algorithm in an argument; it removes the openings the algorithm exploits.

Q: What's the difference between a denial and a rejection?

A: A rejection is kicked back pre-adjudication (formatting, a missing identifier, an eligibility mismatch) and often corrects in minutes. A denial is processed and refused (prior auth, medical necessity, benefit limits) and requires a formal appeal. Clean-claims rate captures both; denial rate captures only the second. Track them separately so you know whether your problem is upstream data or payer adjudication.

Q: Are "soft denials" - silent downcoding - common in behavioral health?

A: Yes, and most centers don't track them. A 90837 paid at the 90834 rate is a soft denial: no formal rejection, just less money. At 8–10% of claims silently downgraded by $30–$50, that's $2,400–$5,000 lost per 1,000 claims. Track expected versus actual reimbursement by CPT monthly to surface the pattern, then batch-appeal the recoverable ones.

Q: How much can structured documentation actually move my denial rate?

A: It's one lever of several, not a cure-all. Notes missing severity, functional impairment, time, and treatment-plan linkage drive 10–18% of denials directly and feed another 10–18% through medical necessity. Structured templates with required fields typically pull combined documentation-plus-medical-necessity denials down 30–50% within six months. They do nothing for prior-auth, eligibility, or coding-error denials. Those need their own workflows.

Q: Should I invest in prior-auth tooling, claims scrubbing, or both?

A: Both. They cover different failures. Prior-auth tooling (or a benefits agent) handles the front-end check that an auth exists and matches before the visit. Claims scrubbing handles the back-end check that the claim matches payer rules. Auth tooling won't catch a wrong modifier; a scrubber won't catch a plan change. Centers stuck above 15% denials usually have one but not the other.

Q: How long does it take to move a treatment center's denial rate from 20% to 10%?

A: With tighter eligibility and auth workflows, structured documentation, and pre-submission scrubbing, most centers see 3–5 points of reduction within 90 days and reach 10–12% by month six. Getting into single digits typically requires agentic RCM tooling and 9–12 months of denial-pattern learning across your specific payer mix, because the last few points come from payer-specific edge cases that only reveal themselves over time.

Q: Will the Senate report or the lawsuits force payers to stop using AI for denials?

A: Don't plan your A/R around it. Regulatory and legal pressure may reshape the rules eventually, but it won't move on your billing-cycle timeline, and automation that speeds legitimate approvals is unlikely to disappear entirely. Treat payer automation as a permanent feature of the landscape and build your operation to match it, rather than waiting for it to be regulated away.

References

• American Medical Association. (2024). Prior authorization practice resources and physician survey. https://www.ama-assn.org/practice-management/prior-authorization
• CareCloud Continuum. (2025). How Much Are Denials Really Costing You? https://carecloud.com/continuum/how-much-are-denials-really-costing-you/
• Centers for Medicare & Medicaid Services. (2024). Prior Authorization and Pre-Claim Review Program Statistics, FY 2024. https://www.cms.gov/files/document/pre-claim-review-program-statistics-document-fy-24.pdf
• Experian Health. (2025). State of Claims Report 2025. https://www.experian.com/blogs/healthcare/healthcare-claim-denials-statistics-state-of-claims-report/
• Global AHS. (2025). Behavioral Health Billing Denial Rate Benchmarks. https://www.globalahs.com/behavioral-health-billing-denial-rate-benchmarks/
• Healthcare Dive. (2025). More insurance claims denials are being overturned upon appeal. https://www.healthcaredive.com/news/insurance-denials-overturned-appeal-new-york-study-JAMA/817490/
• HFMA. (2025). Navigating Medical Necessity Denials: Strategies for Successful Resolution. https://www.hfma.org/revenue-cycle/navigating-medical-necessity-denials-strategies-for-successful-resolution/
• HFMA. (2025). Navigating the Rising Tide of Denials. https://www.hfma.org/revenue-cycle/denials-management/navigating-the-rising-tide-of-denials/
• HFMA. (2024). Claim Integrity Task Force: Standardizing Denial Metrics. https://www.hfma.org/guidance/standardizing-denial-metrics-revenue-cycle-benchmarking-process-improvement/
• HHS Office of Inspector General. (2023). High Rates of Prior Authorization Denials by Some Plans and Limited State Oversight Raise Concerns About Access to Care in Medicaid Managed Care (OEI-09-19-00350). https://oig.hhs.gov/reports/all/2023/high-rates-of-prior-authorization-denials-by-some-plans-and-limited-state-oversight-raise-concerns-about-access-to-care-in-medicaid-managed-care/
• ICANotes. (2025). Behavioral Health Billing Metrics & KPIs 2025. https://www.icanotes.com/2025/09/24/behavioral-health-billing-metrics-kpis/
• KFF. (2025). Claims Denials and Appeals in ACA Marketplace Plans in 2024. https://www.kff.org/patient-consumer-protections/claims-denials-and-appeals-in-aca-marketplace-plans-in-2024/
• KFF. (2025). Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024. https://www.kff.org/medicare/medicare-advantage-insurers-made-nearly-53-million-prior-authorization-determinations-in-2024/
• ProPublica. (2023). How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them. https://www.propublica.org/article/cigna-pxdx-medical-health-insurance-rejection-claims
• SimiTree. (2025). How to Calculate and Improve Your Behavioral Health Clean Claims Rate. https://simitreehc.com/simitree-blog/how-to-calculate-and-improve-your-behavioral-health-clean-claims-rate/
• STAT News. (2023). UnitedHealth used a deeply flawed AI algorithm to deny care to seriously ill patients, lawsuit alleges. https://www.statnews.com/2023/11/14/unitedhealth-algorithm-medicare-advantage-investigation/
• U.S. Senate Permanent Subcommittee on Investigations. (2024). Refusal of Recovery: How Medicare Advantage Insurers Have Denied Patients Access to Post-Acute Care. https://www.hsgac.senate.gov/subcommittees/investigations/
• Supahealth. (2026). Aggregate billing and denials data from 200+ behavioral health practices. Internal dataset.