Anwar would bar AI from making health care decisions

In an ever-changing, tech-driven world, artificial intelligence systems are becoming increasingly pervasive each year. At the same time, lawmakers across the country and in Connecticut are looking to legislation as a way to regulate it, especially in the health care field.

One such bill, proposed by state Sen. Saud Anwar, D-South Windsor, aims to amend the state's code to prohibit health insurance carriers from using AI to determine patient care "to safeguard patient access to testing, medications and procedures."

He said the bill comes after a ProPublica investigation found how Connecticut-based Cigna Insurance was refusing care to patients with the help of AI-guided algorithms during the prior authorization process. According to the investigation, Cigna doctors denied over 300,000 requests for payments using an automated system driven by AI, spending an average of 1.2 seconds on each case, over two months in 2022. Since then, there been a flurry of lawsuits against large health insurance companies -- including UnitedHealth -- alleging the improper use of AI to deny health care faster, following a similar investigation.

Anwar, who is a medical doctor specializing in diseases of the lungs, said these quick denials of care increase "stock value at the cost of human beings," as patients "will die ... some of them will live in agony and pain" while waiting to access the care they need.

Patients "deserve a human being on the other side to be able to look at things, rather than an artificial intelligence algorithm, which is basically trained to deny care, so that some of the times insurance companies can make more profits," Anwar said. "That is unethical, immoral and wrong for the patients."

It's not new for health payers to use algorithms in their administrative process; AI is the next natural step for that technology, said Paul Kidwell, senior vice president of policy at the Connecticut Hospital Association. However, the trend needs a watchful eye to ensure transparency when the algorithms are used in this way.

He said every patient and the care they need is different, so it makes staff wonder who makes the prior authorization denial decisions in "two seconds," given the complexity of each person's care.

A recent American Medical Association survey found that three in five physicians nationwide are concerned about using AI to increase prior authorization denials, overriding "good medical judgment and systematically denying patients coverage for necessary medical care."

When providers go through the "exhausting" and "unnecessarily complex" process of appealing, Kidwell said most denials are ultimately overturned, costing time and money for the hospital and creating barriers to care for patients.

Susan Halpin, executive director of the Connecticut Association of Health Plans, said health carriers do not use AI to make health decisions with those remaining in the "domain of humans."

"Given the time-sensitivity of all health care interactions, and the gravity of decisions made by the minute, health carriers vary in their deployment of [artificial intelligence systems], but all rely on the work of a Human-in-the-Loop, for critical decision making," Halpin said in an email statement to CT Insider.

She added that the Connecticut Insurance Department oversees health carriers working in the state. The carriers are also subject to the unfair insurance practice act and anti-discrimination regulations and must comply with a bulletin notice on AI systems issued in February 2024.

At the same time, Haplin said AI is "positively transforming" the health care delivery system. She said AI systems that are "responsibly developed and deployed ... will yield accessible, effective, affordable and enhanced health care solutions for all consumers."

She said some of the consumer benefits include improved engagement, clarity, and access to benefits and financial services. It also helped improve access to services through advances in telemedicine and timely data with wearable monitoring devices.

Artificial intelligence systems "and machine learning administrative uses, such as payment integrity and claims adjudication, can help address some of the challenges between payers and providers by improving accuracy and efficiency," Halpin said in an email statement.

Connecticut is not the first or last state to introduce legislation aimed at AI oversight. According to the National Conference of State Legislatures, at least 40 states have proposed or passed legislation on AI regulations, many relating to health care, during the 2024 legislative season. State lawmakers have tried passing AI regulation bills in the past, but they have been met with resistance and discussions.

Anwar's bill was referred to the Joint Committee on Insurance and Real Estate in late January. He said he imagines seeing pushback on the bill from representatives of the insurance industry who have "significant influence in the legislature" based on past experiences.

Kidwell said that information about how the AI system is taught and used, and the type of oversight in place are some of the fundamental answers hospitals and their staff need in order to understand the process better.

"Our general view of AI is it can be a really important powerful tool ... so how do we do both of those things, make sure that we're really getting the benefit while making sure that it's fundamentally there to support patient, it's not creating barriers to access," he said.