The Burden of Proof.

Because Ontario funds autism therapy outside its universal health system, and the programme it built in place of universal coverage behaves like an insurance plan: families register with a privately operated administrator, wait for a funding determination, spend against an individual allocation, and keep receipts for seven years to justify it. 71,263 of 91,974 registered children — 77.5% — are waiting for that funding.

The United States is where the insurance model has run longest, been measured most closely, and left the clearest paper trail: federal transparency data on denial rates, inspector-general audits, peer-reviewed evaluations of what coverage mandates actually cost. It is the best-documented example of what happens to disabled children when the entity deciding a claim is also the entity that keeps the money.

The American evidence in Parts I through V is the control group. Ontario has adopted the administrative architecture without the insurer, and Part VII traces what that has produced here: a $385M annual gap between the $965M budget and the $1.35B the Financial Accountability Office’s methodology implies, absorbed by a queue rather than by a denial letter.

In 2024, US Affordable Care Act marketplace insurers denied 19% of in-network claims — roughly 85 million of 451 million submitted — with individual insurers ranging from 3% to 36%. Only about 5% of denials cited medical necessity. More than a third carried no specific reason at all.

Those figures, drawn from federal transparency data compiled by KFF, describe the machine at rest — its normal operating state, not a scandal year. The same dataset contains the number that explains why the machine persists: of the roughly 85 million denied claims, families appealed 262,982 — under one percent. Of those few appeals, insurers upheld their own denials 66% of the time.

Public-administration researchers Pamela Herd and Donald Moynihan call this administrative burden: the learning costs of decoding what is covered, the compliance costs of assembling documentation, and the psychological costs of being treated as a suspect rather than a claimant. In health care those burdens are not evenly distributed. A national survey published in Health Services Research found 73% of insured US adults performed at least one insurance administrative task in a year — and about one in four delayed or went without care because of one, a barrier the authors found comparable to cost itself. Disabled respondents were nearly three times as likely to carry these tasks.

For autism families the burden compounds, because the service being contested is not an event but a schedule — therapy hours, week after week, each block of hours a fresh opportunity for the payer to ask the parent to prove, again, that their child is still autistic and still improving, or not improving fast enough, or improving too fast to need help. Every re-authorization cycle re-runs the same trial with the same defendant: the family.

The obvious question about that funnel is why almost nobody appeals. The answer is not that patients accept the decision. In a KFF survey of insured consumers, only 40% believed they had a legal right to have a denial reviewed by a government agency or an independent expert; 51% were unsure. Among marketplace enrollees — the population in the denial figures above — awareness fell to 34%.

A right that two-thirds of the people holding it do not know they have is a formality. The appeals process satisfies the requirement that a remedy exist while the denial rate behaves as though one does not.

In 2023, ProPublica and The Capitol Forum published internal Cigna documents describing a review system called PXDX, under which company doctors could deny batches of claims without opening the patient files. Over two months in 2022, the system was used to deny 300,000 claims — an average of 1.2 seconds of review per claim. Cigna told reporters the process was legal and applied to low-cost claims. That defence is worth sitting with: the proving apparatus pointed at families operates in seconds when pointed by the payer.

For autism specifically, ProPublica’s 2024 reporting on UnitedHealth Group’s Optum unit described internal strategy documents for limiting spending on applied behavior analysis — the most commonly prescribed autism therapy — in the Medicaid plans it manages. The documents describe applying a “more rigorous” clinical review to ABA in roughly 20 states, culling providers from the network, and authorizing fewer therapy hours than clinicians requested — while the company projected spending about $290 million on ABA in its Medicaid plans that year, and its parent reported $22 billion in annual net profit. A related Optum program for limiting outpatient therapy was ruled unlawful by courts or regulators in three states, per the same reporting.

None of this machinery is confined to one company. The American Medical Association’s 2024 survey found practices completing 39 prior-authorization requests per physician per week, with 93% of physicians reporting that prior authorization delays necessary care. When federal inspectors audited Medicare Advantage prior-authorization denials for post-acute care, they found 95% of appealed nursing-facility denials were overturned. And the federal agencies that police mental-health parity reported to Congress in 2026 that the most common violation remains exactly the one autism families know best: stricter prior authorization and utilization review for behavioral care than for medical care.

The denial statistics in Part II describe paperwork. This is what the paperwork does to a childhood. In a cohort of 4,145 children referred for ABA inside a single large US health system, researchers followed what happened next: 13% never received any ABA at all, 66% were still in therapy at twelve months, 46% at twenty-four, and 28% received the full clinically recommended dose — the dose the evidence base for the therapy is actually built on. A separate study of a different Kaiser region found 23% of referred children never initiated treatment, and that only 15% received at least 80% of their recommended hours.

The second clock starts earlier and runs longer. Autism can be identified far sooner than it usually is: Pierce and colleagues found that a diagnosis made at around 14 months is already largely stable, with stability rising through the second year. The median American child is diagnosed at 47 months. Then, in the publicly funded systems studied by Yingling and colleagues, an average of nearly three more years passes between diagnosis and the start of intensive intervention.

Put those together and the child who could have started at eighteen months frequently starts near seven. That interval is not neutral waiting; it is the interval the clinical evidence identifies as the one that matters most. The Early Start Denver Model trial recorded a 17.6-point IQ gain against 7.0 in the comparison group for toddlers starting between 18 and 30 months, and a later cost analysis of the same model found it cost about $14,000 more per year while it ran and saved roughly $19,000 per child per year afterwards. The system is not only failing the child. It is declining a return it has already been shown.

Mothers of children with autism earn 56% less than mothers of children with no health limitation — about $14,755 a year — and their families earn 28% less overall, a penalty the researchers attribute to the unpaid work of arranging and supervising care that no one else will coordinate.

The direct costs stack on top. In national US data, families of children with autism were more likely than other special-needs families to spend ten or more hours a week providing or coordinating care and to have paid more than $1,000 out of pocket in a year. Even in states that forced insurers to cover autism therapy, families in the highest-spending fifth still paid over $200 a month out of pocket for ASD-specific services.

The legislative record is its own indictment. Insurers did not cover autism therapy until every state and DC passed a law compelling them to, a campaign that ran from Indiana in 2001 to Tennessee in 2019 — and when those mandates arrived, use of autism services among commercially insured children measurably rose, which is another way of saying children had been going without. Where mandates carried age caps, service use fell the month the cap hit, by coverage rule rather than by clinical need. And state parity laws — the promise that behavioral care would be treated like medical care — showed no demonstrable improvement in access for autism families in the peer-reviewed evaluation; families in strict-parity states were 61% more likely to report four-figure out-of-pocket spending.

Even a mandate, an in-network provider, and a willing clinician do not close the loop. In a large California health system operating under a state mandate, 13% of children referred for ABA never received a single hour of it, and only 28% received a full clinical dose. The apparatus of proving, authorizing, re-authorizing and reconciling consumes families before it ever pays them.

Autism coverage in the United States was not granted; it was extracted. Beginning with Indiana in 2001 and ending with Tennessee in 2019, every state and the District of Columbia eventually passed a law forcing insurers to cover autism services — a state-by-state campaign fought over eighteen years against the documented opposition of the industry’s trade association. AHIP’s public argument — that mandates raise premiums and thereby “reduce access to affordable coverage” — reframed paying for disabled children’s therapy as a threat to everyone else’s insurance.

The industry’s weapon in those fights was the cost projection, and the record now lets us score them. In Oklahoma in 2009, an actuary retained for House leadership warned legislators that the proposed mandate, “Nick’s Law,” would raise premiums by 7.8% to 19.8%; the state’s own employee insurance board put it at 1% or less, and the bill died in committee. The peer-reviewed estimate published that same year by Bouder, Spielman and Mandell landed at about 1%, with an upper bound of 2.31%. The insurer-side figure was high by roughly an order of magnitude, and it worked anyway: Oklahoma parents later filed a professional-conduct complaint over the analysis.

What the mandates actually cost is now measurable. Across state-employee plans the reported cost of adding autism coverage was $0.15 per member per month in the first year and $0.31 in the second — figures collected by Autism Speaks, an advocacy organization, and reported in the Harvard Law & Policy Review, so worth treating as directional rather than independent actuarial work. The independent confirmation is Pennsylvania’s: after its mandate, actual autism-related spending came to $6,289 per child in 2012 — 17% of the $36,000 annual cap insurers had warned about. Meanwhile the peer-reviewed evaluations found exactly what compulsion was supposed to achieve: treated prevalence rose, service use and spending per child increased, and the board-certified behaviour-analyst workforce grew from 1.34 to 29.88 per 100,000 children between 2003 and 2017.

There is a catch that makes the entire mandate story narrower than it sounds. State insurance mandates do not reach self-funded employer plans, which are governed by the federal ERISA statute instead. In 2025, 67% of covered US workers — and 80% of those at large firms — were in exactly such plans. The hardest-won consumer protection in autism coverage does not legally apply to two-thirds of the people who assume it protects them.

The lobbying that defends this architecture is a line item. In 2024 the insurance industry spent approximately $155 million on federal lobbying — one industry, one year, one country. The Affordable Care Act’s medical-loss-ratio rules cap what insurers may retain at 15–20% of premium for administration and profit; on the industry’s scale that retained share is measured in hundreds of billions. A company the size of UnitedHealth Group — $447.6 billion in 2025 revenues — does not need to deny any individual claim in bad faith for the arithmetic to work. It needs only a system in which proving is slow, appealing is rare, and the default is no.

The counter-model is universal health care, and the comparison is largely settled in the literature. Himmelstein, Campbell and Woolhandler measured administrative overhead across the US and Canada in Annals of Internal Medicine: in 2017, health care administration consumed 34.2% of US health spending ($2,497 per person) versus 17.0% in single-payer Canada ($551 per person). The largest gap was precisely where the proving happens: insurer overhead ran $844 per American against $146 per Canadian.

A universal system operates on a fundamentally different legal and economic logic. The coverage pool is the population. The premium is the tax base. Because the payer has no commercial imperative to minimize payouts and no competitors to defend against, the machinery of proving — prior authorization, concurrent review, eligibility renewal — is radically reduced. In a universal system, the diagnosis is the eligibility. The proving apparatus largely disappears because its economic function does.

This does not make universal systems flawless; they are characterized by wait times for elective and specialist care, a tradeoff Canada accepts in exchange for eliminating financial ruin at the point of service. But they are uniquely efficient at precisely the thing the insurance model is bad at: moving resources to patients with chronic, complex, or developmental needs without forcing them to assemble a defensive dossier.

The OAP currently reports 91,974 registered children. Of those, 71,263 — 77.5% — are waiting for core clinical funding.

The Financial Accountability Office of Ontario (FAO) calculated that to actually serve the waitlist under the current model would require an annual budget of roughly $1.35B. The province’s 2026-27 allocation is $965M. That $385M gap is the mathematical reason the waitlist exists. The programme does not have the funds to authorize the care it promises.

Rather than sending 71,263 denial letters, the system absorbs the shortfall through duration. This is the danger of administrative burden removed from its American context: it obscures the structural failure. An insurer’s denial is an explicit, actionable refusal. A five-year waitlist is a passive refusal that masquerades as an administrative process. For the developing child, the distinction is entirely theoretical.