The Evidence: 300,000 Denials, Two Months, 1.2 Seconds Each
Do the arithmetic before the outrage. As of August 15, 2026, the most-cited number in the entire AI-claims debate is 1.2 seconds — the average time a Cigna medical director spent per claim when signing off on batches of denials, according to internal documents reported by ProPublica and STAT News in March 2023. Over roughly two months, Cigna doctors denied more than 300,000 requests for payment through a system called PXDX (Procedure-to-Diagnosis). One doctor alone accounted for 121,000 of them.
Run 1.2 seconds across 300,000 denials and you get about 100 hours of total human review — spread across a team, over two months, for a decision volume that would occupy a single reviewer for two and a half full-time work weeks if they never ate, slept, or opened a file. That is the number nobody put in a headline, and it is the one that matters. The scandal was never that a computer flagged the claims. It was that the human signature attached to those flags was, functionally, a batch operation.
According to refresh, which surfaced this case again as regulators sharpen their rules on algorithmic coverage decisions, the PXDX story has become the reference point for every subsequent fight over automated claims management — and it is still being litigated in the court of public understanding as much as in court.
What Cigna Says Back — and Why the Distinction Is Not Trivial
Here is where a careful skeptic should push back, because the sources genuinely diverge and most coverage flattens it.
ProPublica and STAT framed PXDX as a mechanism enabling denials at scale with negligible human review. Cigna's public statement framed the same system as a routine post-service review tool that verifies whether a billed procedure matches the submitted diagnosis — and insisted it does not deny care and does not use an algorithm to make treatment decisions at the point of service. Reuters, covering the resulting litigation, reported Cigna's legal defense along similar lines.
Both framings can be technically accurate at once, and that is the uncomfortable part. A post-service coding check is not a doctor refusing your surgery. But "we only denied the payment, not the care" lands differently depending on where you sit in the transaction. If the procedure already happened and the payment is refused, the bill does not evaporate — it relocates. It lands on the patient, or on the provider who then bills the patient. The distinction between denying care and denying payment is meaningful to a compliance department and close to meaningless to someone holding a five-figure invoice.
Our read: the divergence is less about facts than about which noun the denial attaches to. Cigna is defending the noun. Patients experience the verb.
The Legal Question Is About Time, Not Technology
The class action that followed — Kisting-Leung v. Cigna, Case No. 2:23-cv-01477, filed July 24, 2023 in the Eastern District of California — did not allege that using software is illegal. It alleged that Cigna violated California law requiring insurers to conduct thorough, fair, and objective investigations of claims.
That is a subtle and important shift. The legal theory targets the duration and depth of review, not the existence of an algorithm. Patient advocates made the point plainly: whatever standard "thorough, fair, and objective" sets, 1.2 seconds does not clear it. Former Cigna officials quoted in the original reporting described the process as rubber-stamping rather than individual medical judgment.
Chart: Figures as reported by ProPublica/STAT News (March 2023) and Cigna disclosures from that period. Bars are scaled for visual comparison across different units — denials versus covered members — and are not a direct ratio.
The membership figure is the context most readers miss. Cigna covered roughly 18 million U.S. members in its health plans around that period. Against that base, 300,000 denials in two months is not a rounding error and not a statistical inevitability either — it is a volume that only becomes manageable if review is automated somewhere. Which is precisely the accountability question: at what volume does "human in the loop" stop meaning anything?
Where the Coverage Gap Actually Opens
Now the part that affects your policy coverage rather than your blood pressure.
The gap is not in what your plan covers. It is in what happens between a covered service and a paid claim — the space where a coding mismatch turns a legitimate treatment into a rejected line item. Your policy document lists exclusions to check, and most people read those. Almost nobody reads how the plan describes its review process, because it is usually one bland sentence about medical necessity determinations.
Here is the practical asymmetry, and it is the strongest argument for appealing. A denial that took 1.2 seconds to issue takes a human being hours to contest. That imbalance is the business model of automated denial at scale — not malice, just arithmetic. If a system can generate refusals a thousand times faster than any individual can challenge them, the majority of wrong denials simply survive because nobody has the time. This is the same structural pattern Smart AI Trends flagged in the debate over meaningful human oversight: a review requirement that exists on paper collapses if the reviewer's throughput makes genuine review physically impossible.
Regulators noticed. CMS finalized rules effective in 2024 (CMS-4201-F) restricting how Medicare Advantage plans may use algorithms to deny coverage, requiring that decisions rest on individual patient circumstances rather than a model's output alone. Separately, class actions filed in late 2023 against UnitedHealth and Humana alleged that the nH Predict tool wrongly denied Medicare Advantage post-acute care claims with a reported high override rate. Note the boundary, though: those CMS rules cover Medicare Advantage. If you are on an employer plan or an ACA marketplace plan, that specific protection is not yours — your recourse runs through state insurance law and your plan's internal and external appeal process instead.
The Cheaper Move Most People Skip
The instinct after a denial is to call a lawyer or pay the bill. Both are expensive. There is a middle path that costs almost nothing and works more often than people expect.
Start by requesting the denial reason code and the specific policy language relied upon — in writing. A PXDX-style denial is a diagnosis-to-procedure mismatch, which means the fix is frequently a corrected billing code from your provider's office, not a legal argument about medical necessity. That single request separates a coding problem (cheap, fast, fixable by a billing clerk) from a coverage dispute (slow, contested). Most people never make the distinction and end up fighting the wrong battle.
If it is genuinely a coverage dispute, use the external review your plan is required to offer after internal appeals are exhausted — an independent reviewer outside the insurer. It is free to the consumer in most cases, and it takes the decision away from the party with the financial interest. And when comparing plans during open enrollment, insurance comparison shopping on premium alone misses this entirely: two plans with identical deductibles (the amount you pay out of pocket before insurance kicks in) can have very different denial and appeal-overturn track records. Ask your state's department of insurance for complaint data on the carrier before you enroll. That is a genuine insurance savings lever that has nothing to do with the sticker price.
One caution on the rider question: there is no add-on you can buy that immunizes you against a claim denial. Anyone selling that framing is scare-selling. The protection here is procedural, not a product.
Bottom Line
On balance, our analysis is that the PXDX episode will be remembered less as an AI story than as a timekeeping story. The enduring legal test emerging from Kisting-Leung and the CMS rulemaking is not "did an algorithm decide?" but "could a human plausibly have decided in the time recorded?" That is a far more portable standard, and it is the one insurers' risk assessment and claims management systems are now being quietly rebuilt around — because a defensible audit trail of review time is cheaper than losing a class action. The likely outcome is not less automation. It is automation with a stopwatch attached.
For the individual member, the takeaway is narrower and more useful: a fast denial is not a final one, and the first question to ask is whether the problem is a code or a coverage decision.
Frequently Asked Questions
What is Cigna's PXDX system and how does it work?
PXDX stands for Procedure-to-Diagnosis. According to the March 2023 ProPublica/STAT News reporting, it flags mismatches between a submitted diagnosis and the tests or procedures accepted for that diagnosis, then batches the flagged claims so a medical director can reject them in bulk with a single electronic signature. Cigna has disputed the characterization, describing it as a post-service coding verification tool that does not deny care.
Is it legal for insurers to use AI to deny health insurance claims?
Using software to process claims is not itself prohibited. The legal challenges focus on whether the human review attached to it satisfies state requirements for thorough, fair, and objective investigation. As of August 15, 2026, CMS rules effective in 2024 (CMS-4201-F) restrict Medicare Advantage plans from denying coverage based on algorithmic output without regard to individual patient circumstances. Rules for employer and marketplace plans vary by state — consult a licensed agent or your state insurance department.
What can I do if my health insurance claim was wrongly denied in 2026?
Request the denial code and the exact policy language in writing, confirm with your provider's billing office whether the diagnosis and procedure codes were submitted correctly, file an internal appeal within your plan's deadline, and if that fails, request an independent external review. Many denials resolve at the coding stage without ever becoming a coverage argument.
Which health insurers have been sued over AI claim denials?
Cigna faced Kisting-Leung v. Cigna (E.D. Cal., filed July 24, 2023) over PXDX. Separately, class actions were filed in late 2023 against UnitedHealth and Humana concerning the nH Predict tool and Medicare Advantage post-acute care denials. Litigation status changes over time — verify current dockets before relying on any case outcome.
Explore Our Network
Disclaimer: This article is editorial commentary for informational purposes only and does not constitute insurance, legal, or medical advice. It reflects analysis of publicly reported information, not independent testing or review of any insurer's systems. Always consult a licensed insurance agent for personalized guidance. Research based on publicly available sources current as of August 15, 2026.
No comments:
Post a Comment