Oscar's Record Quarter and the Math Behind It

Introduction

In 2024, ACA insurers denied about 85 million in-network claims. Of those, 262,982 were appealed, fewer than 1 in 100. And of the appeals that got filed, insurers reversed 34% of them. So when one of those rare patients pushed back, the denial flipped about a third of the time.

On May 6, Oscar Health CEO Mark Bertolini reported the highest quarterly profit in the company's history: net income of $678.996 million on a 70.5% medical loss ratio, meaning Oscar kept almost 30 cents of every premium dollar instead of spending it on care. The same KFF data shows Oscar had the highest denial rate among large insurers in the country: 25%. Bertolini's $45 million equity package has no metric for how many claims got approved.

The number that makes the whole thing work

Start with your own claim. If your insurer denied something last year and you let it go, there was about a 1-in-3 chance you'd have won an appeal. The insurer was counting on you not knowing that.

Here's why the odds don't scare them. Run the KFF numbers together and a picture falls out that doesn't appear in any of the coverage. Of 85 million denied claims, fewer than 1% get appealed, and about a third of those appeals succeed. That leaves something on the order of 29 million claims a year that had a real shot on appeal and never got one filed. A 34% reversal rate means denials are wrong at a scale this large, and the wrongness almost never costs the insurer anything, because the patient walks away before the correction ever happens.

That's the math that makes aggressive denial rational. A wrong denial costs the patient a delayed treatment or a procedure quietly abandoned, while the savings stay with the insurer. As long as the appeal rate stays under 1%, every denied dollar is mostly a kept dollar. Bertolini's 70.5% loss ratio is what that looks like on an income statement: premiums retained instead of care delivered, booked as a record quarter.

"Change the outcome of the appeal"

If denials were just honest mistakes, you'd expect a company that found out it was wrong a third of the time to fix the front end. At least one company looked at the back end instead.

The Senate Permanent Subcommittee on Investigations spent a year pulling more than 280,000 pages from UnitedHealth, Humana, and CVS, the three largest Medicare Advantage insurers. The report, released in October 2024, documents a UnitedHealthcare working group that met in December 2022 to explore using machine learning on, in the report's words, "the appeal population." The stated aim was to "change the outcome of the appeal," a verbatim quote from the report (pp. 6-7, 28). The group wanted to predict which denied claims were likely to be appealed and which of those appeals were likely to be overturned, then work back from there. Nothing in that plan involved fixing the denials that were wrong.

The same report tracks what UnitedHealth's denial numbers did once it deployed an algorithm called nH Predict. Its denial rate for skilled nursing care went from 1.4% in 2019 to 12.6% in 2022, a ninefold jump in the first full year the tool was running. Investigators also found a naviHealth instruction to customer service staff that read, verbatim: "IMPORTANT: Do NOT guide providers or give providers answers to the questions." The Senate's own framing was that the insurers were "substituting judgment about medical necessity with a calculation about financial gain."

CVS shows the calculation directly. The company's internal projection for an AI tool called Post-Acute Analytics started at $10 to $15 million in savings over three years. By November 2021, the projection was $77.3 million over the same period, roughly five to six times the original estimate. CVS documents also show the company "deprioritized" a plan to reduce its prior authorization volume because the cost of doing so was "too large to move forward."

There's a separate, sharper number circulating, and it needs a clear label. In an active federal lawsuit, Lokken v. UnitedHealth Group (No. 0:23-cv-03514, D. Minn.), the plaintiffs allege nH Predict carried a 90% error rate and that only 0.2% of policyholders ever appealed. The complaint argues UnitedHealth knew this and kept the system running because the math worked. Those are allegations, not findings. The case is still in discovery, and in March 2026 a federal magistrate ordered UnitedHealth to produce documents on how the algorithm was built and whether it was "designed to supplant physician decision-making." UnitedHealth says nH Predict is a care-support tool, not a claims-decision engine. But the pattern the plaintiffs describe, high error and near-zero appeals and the system stays on anyway, is the same pattern the Senate documented from primary sources.

The regulator that has the data and the $48,869 fine

Every one of those denial rates exists in public because the government collects them. The Centers for Medicare & Medicaid Services runs the Transparency in Coverage reporting that produced the 85-million figure and the 25% Oscar rate, all of it pulled from the federal data by KFF. So CMS has both the data and the enforcement authority, and it has never once used that authority to fine an insurer for high ACA marketplace denial rates.

What it does fine for is bookkeeping. On May 1, 2026, CMS posted a round of penalties: UnitedHealth Group, $48,869; CVS Health, $753,805; Centene, $380,785. Every one was categorized as "Contract Administration." Against a company that booked more than $400 billion in revenue last year, $48,869 is a parking ticket. None of it touched denial rates.

The current playbook substitutes a press event for an audit. In June 2025, CMS Administrator Mehmet Oz and HHS Secretary Robert F. Kennedy Jr. stood up with the insurance lobby and announced that 48 insurers had voluntarily pledged to "simplify" prior authorization. There's no enforcement mechanism in a pledge. The AMA pointed out that insurers made nearly identical promises in 2018 and 2023. By April 2026, the industry reported it had eliminated 11% of prior authorizations, a figure that comes from the insurers themselves and can't be independently checked.

Congress has a version with teeth, and it's worth naming because of what it reveals. The Patient Refunds for Bad Denials Act, introduced April 22 by Reps. Angie Craig and Pat Ryan, would let HHS fine any insurer whose denial rate tops 25%. The penalty is $10 million at the line, plus $2 million for each point above it, paid to the affected patients rather than the Treasury. At a 25% denial rate, Oscar Health would sit exactly at that threshold. The bill has no Republican co-sponsors and hasn't moved.

Who Benefits

The beneficiary is the insurer, and the benefit is money that flows through a specific mechanism most people never see.

The ACA requires individual-market insurers to spend at least 80% of premiums on actual care. That floor is also a target the other way: every claim an insurer denies is a dollar that stays on its side of the medical loss ratio. Oscar's 70.5% Q1 loss ratio means it spent 70.5 cents of each premium dollar on care and held the rest. Wall Street read the held money as a record quarter, not as care withheld. The harm gets spread thin across millions of patients who each lost a single claim, while the benefit lands in one earnings line. The patient carries the full cost of a wrong denial and the insurer carries none of it, which is why the model keeps running.

Bertolini's pay is built to track the kept money and nothing else. His compensation runs on a $1.3 million base, a target bonus of 150%, and that $45 million equity grant. For 2024 the bonus was weighted half on adjusted EBITDA and 30% on premium revenue; for 2025 onward Oscar swapped EBITDA for operating margin after shareholder pushback. The metric changed. What stayed constant across both versions is that no part of his bonus depends on how many claims Oscar approved, how fast it paid them, or how often its denials got reversed. He's paid to optimize what gets measured, and nobody put the denials on that list.

Oz and the political apparatus around him get a different currency: cover. After the killing of UnitedHealthcare CEO Brian Thompson in December 2024 turned insurance denials into raw public anger, a voluntary pledge let the administration hold a podium and announce reform without forcing a single structural change or absorbing a single industry complaint. Actually using the denial data would have meant a fight with the industry; the pledge got the same headline for the price of a press conference.

The policy fix that copies the problem

The part I keep getting stuck on is what CMS is doing next. While it declines to penalize the denial rates its own data exposes, it has started building the denial incentive into the program it directly controls.

In January 2026, CMS launched the WISeR model (Wasteful and Inappropriate Service Reduction), a six-year program now running in New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. It applies AI-driven prior authorization to traditional Medicare, the side that historically didn't have much of it. Six technology companies are running the reviews. Their compensation is tied to a share of the savings from "demonstrated reduction in spending." Pay the vendor more when it approves less. That is the exact financial structure the Senate documented inside Medicare Advantage, the one that took UnitedHealth's skilled-nursing denials up ninefold, now installed by the regulator into the one corner of Medicare that was mostly free of it.

CMS says vendors are incentivized "to get the determination right," not to deny, and that it will monitor denial rates. It already has that same monitoring power over the ACA marketplace, where the denial rate hit 25% at one insurer and the agency's loudest response on record was a $48,869 fine for paperwork.

In Medicare Advantage, when an appealed prior-authorization denial actually gets to an external review, about 82% are overturned, according to a 2026 Health Affairs study from Stanford researchers. Once an independent reviewer actually looks, the denial turns out wrong far more often than it's right. The whole system runs on almost nobody forcing that look in the first place.

The Bottom Line

The industry has already priced denial rates, which is why nobody on that side is racing to fix them. A 25% denial rate plus a sub-1% appeal rate is a working business model, and at least one company put a working group on protecting the math when too many appeals started landing. The data that proves all of it is published by a regulator that treats high denials as something to report rather than something to stop, and that has now wired the same pay-to-deny incentive into traditional Medicare.

The Patient Refunds for Bad Denials Act draws the line at 25% and sends the fines to patients instead of the Treasury. It would put Oscar Health right at the threshold on day one. The open question is whether a denial rate ever gets a price the insurer actually pays. Until it does, the next AI tool just gets better at finding the patients who won't appeal.

Sources

1. KFF, "Claims Denials and Appeals in ACA Marketplace Plans in 2024" (March 24, 2026) — 85M in-network denials, 19% denial rate, Oscar 25% (highest among large parent companies), 262,982 appeals filed, 66% upheld / 34% reversed. https://www.kff.org/patient-consumer-protections/claims-denials-and-appeals-in-aca-marketplace-plans-in-2024/
2. Oscar Health, Q1 2026 results (8-K / BusinessWire, May 6, 2026) — net income attributable to Oscar $678.996M, revenue $4.647B, MLR 70.5%, 3,174,489 members. https://www.businesswire.com/news/home/20260506886599/en/
3. U.S. Senate Permanent Subcommittee on Investigations, "Refusal of Recovery: How Medicare Advantage Insurers Have Denied Patients Access to Post-Acute Care" (October 17, 2024) — "change the outcome of the appeal" (pp. 6-7, 28); UHC skilled nursing denial rate 1.4% (2019) → 12.6% (2022); naviHealth CSR instruction; CVS $10–15M → $77.3M projection; "substituting judgment about medical necessity with a calculation about financial gain." https://www.hsgac.senate.gov/wp-content/uploads/2024.10.17-PSI-Majority-Staff-Report-on-Medicare-Advantage.pdf
4. Estate of Gene B. Lokken v. UnitedHealth Group, Inc., No. 0:23-cv-03514 (D. Minn.) — plaintiffs allege 90% nH Predict error rate and 0.2% appeal rate; March 2026 discovery order. Active litigation; figures are allegations. https://www.courtlistener.com/docket/67888463/estate-of-gene-b-lokken-v-unitedhealth-group-inc/
5. CMS, Part C and Part D Enforcement Actions — May 1, 2026 penalties: UnitedHealth Group $48,869; CVS Health $753,805; Centene $380,785, all "Contract Administration"; no fines for ACA marketplace denial rates; June 2025 voluntary prior-authorization pledge (Oz / RFK Jr.). https://www.cms.gov/medicare/audits-compliance/part-c-d/part-c-and-part-d-enforcement-actions
6. Oscar Health, DEF 14A proxy statements and 8-K (SEC EDGAR, CIK 1568651) — Bertolini $1.3M base, 150% target bonus, $45M equity package; 2024 bonus weighted on adjusted EBITDA (50%) and premiums (30%), changed to operating margin for 2025+; no denial-rate metric. https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=1568651
7. CMS Innovation Center, WISeR Model FAQ — six-year model (2026–2031), six states, six AI vendors, compensation tied to "demonstrated reduction in spending." https://www.cms.gov/priorities/innovation/files/document/wiser-model-frequently-asked-questions
8. Mello, Trotsyuk, Djiberou Mahamadou, Char, "The AI Arms Race In Health Insurance Utilization Review," Health Affairs 45(1), pp. 6–13 (2026) — ~82% overturn rate on appealed Medicare Advantage prior-authorization denials. https://www.healthaffairs.org/doi/10.1377/hlthaff.2025.00897
9. Reps. Angie Craig and Pat Ryan, Patient Refunds for Bad Denials Act (introduced April 22, 2026) — fines insurers with denial rates above 25%; $10M base + $2M per point above threshold, paid to patients. https://jayapal.house.gov/2026/04/22/