300,000 Denials Without a Single Doctor Looking
🔴 REAL INCIDENT: Cigna's PXDX algorithm mass-denied medical claims (2023-2024)
What Happened
In March 2023, a class-action lawsuit revealed something that millions of health insurance customers never knew: their medical claims were being denied by an algorithm—not reviewed by a doctor.
The lawsuit alleged that Cigna, one of America's largest health insurers, deployed a system called "PXDX" (procedure-to-diagnosis) that automatically rejected claims when the procedure code didn't match an expected diagnosis code.
The numbers were staggering:
- Over 300,000 claims denied in a two-month period
- Average review time: 1.2 seconds per claim
- Doctors allegedly signed off on batches of 50 claims at a time
- Patients received denial letters that appeared to come from physician review
The algorithm wasn't assisting doctors in making decisions. It was making the decisions. The doctors were rubber stamps.
How PXDX Worked
Insurance claims include procedure codes (what treatment was provided) and diagnosis codes (why it was needed). Legitimate medical review involves a doctor examining whether the treatment was medically necessary for the patient's condition.
PXDX did something simpler—and more dangerous.
The system maintained a database of "acceptable" procedure-diagnosis combinations. If your claim matched an approved pairing, it was paid. If it didn't match, it was flagged for denial.
The algorithm didn't consider:
- Your medical history
- Your doctor's clinical judgment
- Unusual presentations of common conditions
- New treatments for existing diagnoses
- Individual patient circumstances
It just checked: does this code combination appear in our approved list?
If not, denied.
The 1.2-Second "Review"
Under federal and state law, health insurers must have a physician review claims before denying them on medical necessity grounds. The denial letter patients receive typically states that a doctor reviewed the claim.
According to the lawsuit, that review was a fiction.
The alleged process:
1. PXDX automatically flags claims for denial
2. A physician is presented with a batch of 50 flagged claims
3. The physician has approximately 1.2 seconds per claim to "review"
4. The physician approves the batch
5. Patients receive letters saying a physician reviewed their claim
1.2 seconds isn't review. It's not even reading. It's clicking "approve" on decisions an algorithm already made.
Real People, Real Harm
The claims being denied weren't cosmetic procedures or experimental treatments. They were routine medical care:
Cancer screenings denied because the procedure code didn't match the "expected" diagnosis (the whole point of screening is that you don't have a diagnosis yet)
Physical therapy denied because the algorithm's database didn't link the patient's condition to that treatment
Diagnostic tests denied even when ordered by physicians, because the code combination wasn't in the approved list
Follow-up care denied when patients' conditions didn't fit neatly into the algorithm's categories
Most patients, faced with a denial letter citing "medical necessity," assumed a doctor had actually evaluated their case. They didn't know an algorithm had decided in milliseconds.
Some paid out of pocket. Some went without care. Some fought lengthy appeals processes—against an automated system designed to deny.
The Broader Pattern
Cigna isn't alone. Algorithmic claims processing is industry-standard:
UnitedHealth's NaviHealth: A separate lawsuit alleged that UnitedHealth used an AI tool to predict how long patients should need post-acute care, then denied coverage beyond that prediction—even when patients' doctors said they needed more time to recover.
Industry-wide adoption: A 2024 survey found that roughly 75% of health plans use AI for prior authorization approvals. About 8-12% use AI specifically to support denials.
The economics: Denying claims saves insurers money. Automated denials save even more—you can reject at scale, 24/7, with minimal human overhead. The patients who appeal are a manageable subset.
Why It Took So Long to Surface
Opacity: Patients received denial letters that looked like medical decisions. The algorithm's role was invisible.
Complexity: The procedure-to-diagnosis mapping system sounds technical and reasonable. Understanding why it fails requires medical expertise.
Power asymmetry: Individual patients fighting billion-dollar insurers face overwhelming odds. Most give up.
Regulatory lag: Health insurance is heavily regulated, but regulators struggled to keep pace with algorithmic decision-making that technically complied with letter-of-the-law requirements.
Information asymmetry: The lawsuit only happened because internal documents revealed the 1.2-second review times. Without that evidence, the algorithmic denials would have continued invisibly.
The Legal and Regulatory Response
The class-action lawsuit seeks to represent millions of Cigna customers whose claims were denied through PXDX.
At the state level, new laws are emerging:
- Texas (2025): Prohibits using automated systems as the sole basis for medical necessity denials without human oversight
- Arizona and Maryland: Similar laws requiring human review for AI-assisted denials
- New York: Proposed legislation mandating disclosure when AI is used in claims decisions
Federal agencies have signaled increased scrutiny. The HHS has emphasized that "there are no exceptions to civil rights laws for algorithms" in healthcare.
But the fundamental tension remains: automated denial is profitable. Human review is expensive. Until the penalties for algorithmic harm exceed the savings, the incentives favor the machines.
The Root Cause
PXDX wasn't built to provide medical care. It was built to manage costs.
The algorithm optimized for what it was designed to optimize for: identifying claims that didn't fit expected patterns and flagging them for denial. From a cost-control perspective, it worked perfectly.
From a healthcare perspective, it was a machine for denying care to people who needed it—with a veneer of medical legitimacy that even the patients couldn't see through.
The physicians who "reviewed" the denials weren't practicing medicine. They were processing decisions made by a system that understood procedure codes but nothing about patients.
How It Could Have Been Prevented
Meaningful physician review: If a claim requires denial on medical grounds, an actual physician should actually review it—with enough time to understand the case.
Transparency: Patients should know when algorithms are involved in their care decisions. "This claim was processed by an automated system" should be disclosed.
Outcome auditing: Insurers should track what happens to patients whose claims are denied. If denial leads to worse outcomes, the algorithm is causing harm.
Regulatory oversight: Regulators need visibility into algorithmic decision-making, including denial rates, review times, and appeal outcomes.
Appeal mechanisms that work: When appealing an algorithmic denial, patients shouldn't be fighting the same algorithm. Human review means human review.
The Lesson
An algorithm that denies healthcare claims in 1.2 seconds isn't assisting medical decision-making. It's replacing it.
The doctor's name on the denial letter provides legal cover. The algorithm provides deniability. The patient provides the suffering.
When AI systems make life-and-death decisions, governance isn't optional. The question isn't whether automation improves efficiency. It's whether efficiency should be the goal when someone's cancer screening is on the line.
What decisions are your AI systems making about people's lives—and who's actually reviewing them?
Sources: