A denial is a bet you won't appeal.

The letter that says your claim has been denied is usually one page. Near the bottom, in a paragraph the law requires, it tells you that you have the right to appeal. That is the safest paragraph the insurer will ever print. In 2024, insurers selling plans on HealthCare.gov denied roughly 85 million in-network claims, and consumers appealed fewer than 1 percent of them. In Medicare Advantage, where denials do get appealed, 80.7 percent of those appeals won. I want to convince you that these two numbers belong in the same sentence, because together they describe a transaction. When an insurer denies a claim, it is betting you will not come back. The bet almost always pays.

Let me concede the strongest objection first, because it is real. "Denial" is a sloppy word. It covers duplicate submissions, miskeyed codes, claims sent to the wrong payer, and services the plan never covered, and insurers argue, correctly, that billing departments fix much of this without anyone filing an appeal. Here is the trouble with that defense: the federal data already accounts for it. KFF's denial figures count only final adjudication status. A claim denied in March and paid in April shows up as paid. The 85 million stayed denied. And of the reasons insurers themselves gave for those denials, 5 percent involved medical necessity. The rest was plumbing, and the plumbing's errors are profitable by default.

Nineteen percent, before anyone blinks

Start with the one corner of American health insurance where denial data is public: plans sold on HealthCare.gov. In 2024, the 157 insurers on that platform received 451 million in-network claims and denied about 85 million, a 19 percent denial rate, one claim in five, essentially unchanged from the year before. Remember what these plans are. They run on a federal rulebook, file their numbers with CMS, and answer to regulators in every state they sell in. One in five is the rate with the lights on.

The average tells you less than the spread. Across those 157 insurers, in-network denial rates ran from 3 percent to 36 percent. Same regulated product, same federal rulebook, same kinds of patients, and one insurer says no twelve times as often as another. If denial rates measured medicine, they could not vary twelvefold between companies selling the same policy. A spread that wide is measuring the insurers, not the patients.

The insurers' own stated reasons point the same direction. The single largest category of denial reasons in 2024 was "other," a label covering 36 percent, which is to say the most common explanation for refusing payment was no explanation. Administrative reasons took another 25 percent. Excluded services, 13 percent. Missing prior authorization or referral, 9 percent. Medical necessity, the one category that contains an actual judgment about whether a person needed care, was 5 percent.

None of this is exotic. In a January 2026 KFF poll, a third of insured adults said an insurer had refused to cover a service or medication their doctor prescribed within the past two years. The denial letter is among the most widely distributed documents in American life, and the appeal paragraph at the bottom of it is among the least used.

What a wrongful denial costs the insurer

Now run the arithmetic nobody prints in the denial letter. Of those 85 million denied in-network claims, consumers appealed at least 262,982. Three in a thousand. When consumers did appeal, insurers reviewed their own decision and upheld it 66 percent of the time. So the overturned remainder, the denials the insurer itself admitted were wrong once someone asked, worked out to roughly one in every thousand denials issued.

Put that in an actuary's terms. Suppose an insurer denies a thousand dollars of claims it should have paid. The probability anyone appeals is about 0.3 percent. The probability the appeal succeeds is about one in three. The expected cost of the wrongful denial, through the consumer appeal channel, is about one dollar. The other 999 dollars is kept, by default, because default is what people do.

Past the internal appeal, the law provides an independent external review. In 2024, HealthCare.gov consumers took at least 5,881 denials to external review. Against 85 million denials, that is about one in fourteen thousand. The system has a court of appeals, and statistically, no one has ever been there.

None of this requires anyone at an insurer to twirl a mustache. Each individual surrender is rational. A $400 denial is not worth six hours of phone trees, document hunts, and a sixty-day wait, so the rational person eats it, and 85 million rational people eating it sums to a very large number that never appears on any income statement as what it is. The insurer does not need to be right. It needs you to be tired. And on the ledger, an unappealed wrongful denial is identical to a correct one.

Four in five denials fail the check

So what happens when somebody does come back? For that, look at Medicare Advantage, the one market where the appeal pipeline is measured end to end. In 2024, MA insurers processed nearly 53 million prior authorization requests, about 1.7 per enrollee, and fully or partially denied 4.1 million of them. That is a 7.7 percent denial rate, up from 6.4 percent in 2023. Of those denials, 11.5 percent were appealed. Of those appeals, 80.7 percent won.

Sit with the 80.7 for a second. These are not soft reversals on technicalities. KFF's read is that overturned appeals represent care a physician ordered and care ultimately deemed necessary, delayed by the extra step of appealing. Nor was 2024 an outlier: in every year KFF has examined since 2019, more than eight in ten MA appeals overturned the initial denial. The appeal rate is higher here than on the marketplace, 11.5 percent against 0.3, largely because in MA the appellant is often a hospital or a physician's office with staff paid to fight, not a patient alone at a kitchen table. Even so, 88.5 percent of denials were never challenged by anyone. Multiply the gates and roughly nine of every hundred MA denials are ever reversed. A decision that fails review four times out of five survives, overall, more than nine times out of ten.

The obvious rejoinder is selection: the denials that get appealed are presumably the strongest cases, so 80.7 percent says little about the denials nobody contests. Fair, and it is why the most important study here is the one with no selection in it. In 2022, the HHS Office of Inspector General published a review of a stratified random sample of denials issued by the 15 largest MA organizations during one week of June 2019, checked case by case against Medicare's own coverage rules. Among prior authorization denials, 13 percent were for requests that met those rules and should have been approved. Among payment denials, 18 percent. Not the strongest cases. Random ones. Applied naively to 2024's volume, the 13 percent floor alone would mean over half a million wrongly denied requests a year, in one insurance market, before anyone appeals anything.

The biggest market reports nothing

Everything above, every number, comes from two comparatively small corners of the insurance world: HealthCare.gov plans, where about 16 million people picked coverage for 2024, and Medicare Advantage, covering about 33 million. The largest source of health coverage in the United States is neither. It is the employer plan, covering 154 million people under 65, nine HealthCare.gov marketplaces stacked on top of each other. For those 154 million there is no public denial rate, no appeal rate, no overturn rate. The shelf is bare.

The measurement is missing because a law was left half-built. The Affordable Care Act required insurers to report claims denials and related transparency data, and a companion provision extends the requirement to employer group plans. Sixteen years on, the federal government has implemented it only for plans sold on HealthCare.gov. KFF's own brief puts it flatly: for the rest of the market, the requirements "still have not been implemented at the federal level." What we know about denials comes from the one shelf where someone installed a light, and the warehouse behind it is dark.

There is no particular reason to assume employer plans behave better in the dark. The same insurers administer them, often on the same claims systems, and the third of insured adults reporting a denial in KFF's polling were not drawn only from the measured markets. What the dark guarantees is narrower and worse: an employer deciding between carriers cannot compare their denial rates, an employee cannot know whether their plan denies at 3 percent or 36, and a wager on exhaustion runs in the one market where nobody can even count the bets.

Two honest objections, and what would change my mind

The first objection I have already named: selection. An 80.7 percent overturn rate among appealed denials is a ceiling, not a base rate, and anyone quoting it as "80 percent of denials are wrong" is misreading it. The honest base-rate evidence is the OIG's random sample, and it says 13 to 18 percent, which is both far below 80 and far above anything a system should tolerate in a domain where the error is someone's chemotherapy authorization. The bet framing does not need 80. It needs the wrongful-denial rate to sit meaningfully above the reversal rate, and 13 percent against a 9 percent reversal rate clears that bar with room to spare.

The second objection is that appeal rates understate contestation, because providers fix most denials by correcting and resubmitting claims rather than filing formal appeals. Also true, and also already in the data: KFF's 85 million counts only claims whose final status was denied, after the resubmission churn resolved. What the objection gets right is that my one-dollar figure measures a single channel. Provider billing teams claw back real money through rework, and insurers bear real administrative costs running the machine. What it cannot rescue is the patient's position. Resubmission is the provider's channel, run on the provider's behalf. The only lever the patient holds is the appeal, and the patient pulls it three times in a thousand.

Here is what would change my mind. CMS has begun a pilot collecting prior authorization data at the plan and service level, with expansion to all MA plans anticipated in 2027; if that data showed denials concentrated in genuinely low-value care, the exhaustion read weakens. If the group-plan transparency provision were ever implemented and employer-plan denial rates came in near the marketplace's best insurers, with appeal outcomes near coin-flip rather than four in five, the wager framing would be wrong, and I would write that paper. The prediction this paper makes is falsifiable: wherever denial decisions finally get measured and challenged at scale, a large fraction will not survive the contact. Everything published so far points one way.