Breaking News

More bad news for Epic, inside a ChatGPT experiment, & diabetes tech under legal scrutiny

August 29, 2023
Reporter, STAT Health Tech Writer
Good morning, health tech readers! I'm back from a couple weeks off — let me know what I've been missing at mohana.ravindranath@statnews.com. And if you'd like to stay up to date at HLTH in October, sign up for our pop-up newsletter here
medical devices

Insulet sues a rival for allegedly stealing diabetes tech secrets

In her latest dispatch, Lizzy Lawrence gives us a window into a heated legal battle that could have implications for the diabetes technology industry as well as a planned merger.

This month, insulin pump manufacturer Insulet filed a lawsuit against South Korean company EOEFlow, the target of Medtronic's planned $738 million acquisition, for stealing the design for its only product, a patch pump. 

Experts tell Lizzy the lawsuit reflects just how much a threat the planned acquisition poses to Insulet and its Omnipod product. What happens next could predict Insulet's trajectory: If the merger goes through and if regulators approve EOFlow's pump for patients in the U.S., Insulet may see its position threatened. But a lawsuit could also force Medtronic to postpone or rethink the acquisition. 

Some background: Insulet, now worth $13 billion, released its own tubeless, adhesive insulin pump in 2005. In 2016, EOEFlow hired several of the company's senior executives, and Insulet alleges that's when it started copying its pump design. 

"If Medtronic actually had a good patch pump, a good tubeless option, that would be tough to fight against," Debbie Wang, a Morningstar medical device analyst, told Lizzy. Read more from Lizzy here. 


Artificial intelligence

More bad news for Epic's sepsis model

Electronic health record vendor Epic has for years touted the ability of its AI model to give clinicians early warning of the life-threatening condition known as sepsis. But a new study by researchers at North Carolina-based Atrium Health found it to be less useful than other commonly-used models, such as SIRS (Systemic Inflammatory Response Syndrome), Casey Ross tells us.

The researchers reported that Epic's product was more accurate at the highest scoring thresholds, when it was most confident that a patient had sepsis. But they also found that those thresholds were only reached after clinicians had already taken steps to treat the condition. The new findings confirm earlier research by the University of Michigan and reporting by STAT — that first flagged the model's questionable performance. Perhaps more troubling than news of the model's weaknesses is that it took about six years for researchers to smoke them out.


From the field

Inside NYU Langone's ChatGPT experiments

On the ground in New York, Mario Aguilar reports on how one health system is putting generative AI and large language models into practice today.  At a recent "prompt-a-thon" at NYU Langone, a small group including a music therapist and a medical student were tasked with creatively analyzing medical records using the health system's HIPAA-compliant version of ChatGPT, which can generate text in response to queries, with an eye toward equity. (Microsoft makes the tool NYU Langone uses.) 

Among them was Christine Gonzalez, who studies implicit bias in medicine. She noticed the system couldn't identify any instances of bias in the text of patients' records, which suggested the importance of keeping humans in the loop, she said. 

"The ideas aren't going to come from me, they're going to come from everyday folks who are thinking about their own problems, who are doing things for themselves," Yindalon Aphinyanaphongs, an assistant professor who leads the predictive analytics unit in NYU Langone's department of informatics. "And one advantage to GPT is that it's incredibly democratizing with a low barrier to entry." 



Artificial intelligence

Why LLMs fall short in medicine — for now

Speaking of AI's limitations, experts Jenna Wiens, Rada Mihalcea and Brahmajee Nallamothu join in the ongoing concerns in a new STAT First Opinion piece: While they're exciting, large language models may not fundamentally transform doctors' visits like supporters hope, they argue.

But artificial intelligence could be especially helpful if it's used to gather data from health records or other sources at scale — analogous to the massive amounts of text and visual data that sites like Wikipedia, Flickr and others generate for training, they write. 

But health data isn't readily accessible online, and training data sets have blind spots, they note, calling for better data sharing or federated learning systems in which data isn't directly  shared but can be used to update models. Read more here


Telehealth

Research: Virtual mental health care remains up

In the past three years, use of mental health care — and related spending — increased by about 39% and 53% respectively among commercially insured people, a RAND analysis of claims data finds. Authors outlined their findings in a research letter in JAMA Health Forum. Virtual care drove those spikes: Telehealth visits for mental health care has stabilized at about 10 times pre-pandemic levels between 2020 and 2022. Spending on telehealth remained stable during that window, but spending on in-person mental health care sank to prepandemic levels, the claims suggest. 

These findings hint that patients are continuing to spend on mental health care, years after the pandemic's onset. But it's also possible that insurers may reassess their own spending, especially if increased utilization makes plans have to spend more, the authors warned. 


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

  • How telehealth has changed migraine treatment, GQ
  • Why AI can't replace humans in medicine, Time
  • FTC has paused its challenge to the Amgen-Horizon merger, Bloomberg

Thanks for reading! More on Thursday - Mohana

Mohana Ravindranath is a Bay Area correspondent covering health tech at STAT and has made it her mission to separate out hype from reality in health care.


Enjoying STAT Health Tech? Tell us about your experience
Continue reading the latest health & science news with the STAT app
Download on the App Store or get it on Google Play
STAT
STAT, 1 Exchange Place, Boston, MA
©2023, All Rights Reserved.

No comments