How insurers use AI

Executives of the nation's largest health insurance companies regularly highlight their use of AI tools in earnings calls and meetings with investors. But where they see opportunities for efficiency and increased savings, regulators see a potential crisis in the making.

As my colleague Casey Ross writes in a new story, insurance watchdogs are concerned that the rapid uptake of AI is outpacing their ability to scrutinize these tools or even understand how they are being applied to decisions about patients' coverage and care. A recent report by consumer representatives to the National Association of Insurance Commissioners called for urgent action to protect patients against discrimination and harm, concluding: "The importance of acting now cannot be overstated."

Meanwhile, health insurers argue AI is helping to speed up decision times and reduce delays that commonly annoy consumers. In the story, Casey unpacks what large companies like UnitedHealth, Elevance, CVS Health/Aetna, Cigna, and Centene use AI for, and the intensifying debate over regulation.

data

Questions about 23andMe's bankruptcy

23andMe filed for bankruptcy, and the company's co-founder Anne Wojcicki stepped down as CEO. The latest twist in the story for the genetic testing company follows Wojcicki's failed efforts to take the business private. Now she'll have a chance to buy the assets at auction like anyone else.

As STAT's Matthew Herper writes, the bankruptcy is a bad outcome for everyone involved, including the millions of consumers who trusted the company with their genetic data that will now be sold to the highest bidder. California's attorney general issued a public aler t to consumers urging them to consider instructing 23andMe to delete their data. Matt's story asks a number of important questions about what might happen to the data and also what all that data might actually be worth.

News on two DTx business models

A couple of updates from digital therapeutics companies remind us of two distinct models for getting software-based medical treatments into the hands of patients.

On one side, Big Health published an update on its efforts to partner with health systems and providers, including telehealth companies, who want to offer the company's apps to their patients. The company's CEO Yael Berman described how Big Health is working with Henry Ford Health to make its Food and Drug Administration-cleared treatment for insomnia, SleepioRx, available to patients. In early returns, 39% of patients offered the app enrolled and 43% used a clinically effective dose. The company declined to disclose actual numbers for competitive reasons but said they have seen "significant traction" and that more partnerships are forthcoming.

Elsewhere, Click Therapeutics announced it raised a Series C round from Dassault Systèmes. We'll set aside that not spelling out the funding amount is always a funny way to announce an investment round. Click has been doing everything it can to publicize the potential of draft FDA guidance that would allow drug makers to mention related software on their product labels. Assuming drugmakers ever get excited by the idea, Click would be very well positioned to help them add apps to their drugs. The company sets a very high standard for clinical evidence, and many of the treatments in Click's pipeline are developed in partnership with drugmakers. The only product Click has brought to market, a treatment for depression, was developed with Otsuka. No wonder Click needs more money — drug trials are very expensive.

research

AI regulation perspectives

A few recent papers on AI regulation caught my eye:

Researchers writing in NPJ Digital Medicine say they were able to make large language models produce device-like outputs. Here they lean on the FDA's guidance on what kinds of decision support products count as medical devices regulated by the agency. When presented with requests for help with "time-critical emergencies," 100% of GPT-4 and 52% of Llama-3 responses resembled device decision support. The authors make some interesting observations about regulating LLMs, including that traditional approvals for single indications may not be appropriate.
A viewpoint in JAMA Network Open examines the outlook for health AI regulation in coming years noting the Trump administration's rollback of Biden-era executive orders and cutbacks to FDA staff. More broadly, authors point out that FDA has limited ability to regulate all that might cause harm. They argue that the federal health department ought to have more power to regulate health AI, which would require lawmakers to act. So in the short term, "work should focus on strengthening private governance, especially in health care organizations."

More around STAT

Check out more exclusive coverage with a STAT+ subscription

Read premium in-depth biotech, pharma, policy, and life science coverage and analysis with all of our STAT+ articles.

What we're reading

Cancer research, long protected, feels 'devastating' effects under Trump, STAT
OIG's remote patient monitoring audits are here: What you need to know, Nixon Law