Insurance coverage denials have risen in recent years in the US, driven in part by automated algorithms powered by AI – and some recently launched artificial intelligence tools may fight back by generating automatic appeals.

But to see more lasting change, health experts say that the health insurance system needs greater reform to control high prices and ensure coverage.

UnitedHealth, Humana and Cigna are facing class-action lawsuits alleging the insurers relied upon algorithms to deny lifesaving care.

One of the lawsuits alleges that Cigna denied more than 300,000 claims in a two-month period, which amounts to about 1.2 seconds for each physician-reviewed claim. Such a practice is aided by algorithms, the lawsuit said.

In 2020, UnitedHealth Group acquired naviHealth and its algorithm for predicting care, called nH Predict, which UnitedHealth uses and also contracts out to other insurers, including Humana. (A spokesperson for UnitedHealth Group denied that the algorithm is used to make coverage decisions; Humana did not respond to a request for comment.)

The lawsuit against them alleged that nH Predict has a 90% error rate, meaning nine out of 10 denials are reversed upon appeal – but that vanishingly few patients (about 0.2%) appeal their denied claims, leading them to pay bills out of pocket or forgo necessary treatment.

That figure tracks with a survey by the non-profit KFF, which found less than 0.2% of people purchasing insurance through HealthCare.gov appeal in-network claims that are denied.

When it comes to prior authorization, a practice in which doctors and patients must receive an insurance company’s approval before starting care or medications, less than 10% of denied requests in Medicare Advantage plans (Medicare-approved plans from a private company) were appealed in 2022, according to another KFF survey.

Doctor’s offices now have entire departments devoted to processing and appealing prior authorization decisions.

Nearly half of US adults say they have unexpectedly received a medical bill or been charged a co-payment, according to a survey from the Commonwealth Fund.

Four out of five said these delays caused worry and anxiety, and nearly half said their condition worsened because of delayed care. Most did not know they could appeal a denial.

But for those who do try to appeal, the process can be so labyrinthine, they are forced to give up.

Two of her three sons have severe food allergies, so Deirdre O’Reilly was worried about sending one of them out of state to college. When he had a reaction, he went to the emergency room like usual.

But this time, the insurance company denied coverage for the entire visit – nearly $5,000, according to a denial letter reviewed by the Guardian. O’Reilly tried to appeal four times, and each time, the insurer, BlueCross BlueShield of Vermont, gave her a different reason, she said.

“My son didn’t have a choice – he was going to die if he didn’t go to the nearest emergency room,” O’Reilly said.

She should know; she’s an intensive care physician at the University of Vermont. She has seen denials like these happen to her own patients, such as premature infants who have oxygen equipment denied.

“It’s gotten out of control. It’s changed a tremendous amount in the 20 years I’ve been a physician,” she said. “I can’t believe that people have to go through this just to get healthcare covered – things that are basic needs.”

And many people don’t have the same medical expertise and the time or resources for lengthy appeal processes.

“I was tenacious,” she said. “But at some point, I could only fight so much.”

A spokesperson from BlueShield Vermont said in a statement that she could not comment on an individual’s health record, but denied the use of algorithms in managing care. “Most” prior authorization decisions were made by the insurer’s team of doctors and nurses based on national guidelines, she said.

Vermont is one of several states that recently passed legislation to reduce the strain of prior authorizations.

Automated denials in particular have faced increased scrutiny by federal and state lawmakers.

UnitedHealthcare, CVS and Humana – the three largest providers of Medicare Advantage, together providing almost 60% of all Medicare Advantage coverage – reject prior authorization claims at high rates using technology and automation, according to a US Senate report released in October.

Appealing these denials costs more than $7.2bn in administrative costs for providers each year, according to an analysis of data from the US Centers for Medicare and Medicaid Services.

The agency recently announced new rules to regulate prior authorization for Medicare Advantage plans.

For those looking for details on why a claim was denied, ProPublica launched a service to help patients submit records requests.

Some patients and companies have developed AI tools to appeal denials in a “battle of the bots”.

Companies have launched new generative AI tools to help hospitals and patients draft appeal letters, while one open-source large language model developed by an engineer promises to help patients “Fight Health Insurance”.

“Nobody loves the system we had a few years ago – which also used algorithms, just simpler ones,” said Michelle Mello, professor of health policy at the Stanford University School of Medicine. “And now nobody loves it with AI involved. But I think there are constructive roles for improved algorithms to play.”

AI can help make sure forms are coded and formatted according to each insurer’s specifications, she said – making sure the requests aren’t kicked back for being incomplete. It could also be used by insurance companies to approve insurance requests more quickly.

Most denials happen because of mistakes in filling out or filing the form, Andrew Witty, CEO of UnitedHealth Group, said last week on an earnings call during which executives said UnitedHealthcare’s revenues in 2024 neared $300bn, with the company expecting that figure to rise to $340bn in 2025.

Witty estimated that 85% of denied claims could be avoided “through technology in a more standardized approach across the industry”.

Changing to an industry standard, instead of each company having different forms and processes, is especially important, Witty said.

But having human oversight of automated processes is a needed change, experts said.

“Those algorithms aren’t always getting it right, and so I think there’s a fear that more of the human aspect is being pulled out of the system,” said Mika Hamer, assistant professor of health policy and management at the University of Maryland School of Public Health.

California recently enacted legislation to prohibit AI making coverage decisions and to require physician oversight.

But only addressing AI doesn’t fix some of the issues underlying the decision to automate, Hamer said – including sky-high prices for medical care and medications.

“One out of every $5 of the US GDP is spent on healthcare,” said Hamer. “It’s an absolutely massive system. It’s going to take a massive overhaul.”