Experts weigh AI-assisted colonoscopy benefits

You’re reading the web edition of STAT’s Health Tech newsletter, our guide to how technology is transforming the life sciences. Sign up to get it delivered in your inbox every Tuesday and Thursday.

A panel of experts at a leading speciality society announced in a new clinical guideline that they have decided not to decide whether to recommend AI-assisted colonoscopies. 

After reviewing studies and using existing information to model outcomes, experts enlisted by the American Gastroenterological Association determined that using AI definitely increases adenoma detection rate (ADR), or colonoscopies that find polyps. But they figure, with low certainty, that using AI in screening only leads to 2 fewer colorectal cancer-related deaths per 10,000 people over 10 years. Citing a close call and fuzzy evidence, the panel decided not to issue a recommendation.

The unsatisfying shrug of conclusion underscores the trade-offs involved in enlisting new technology to screen for disease. Usually, the ADR is an important measure of quality as it is associated with better outcomes. But using technology to boost ADR can lead to more testing, more procedures, and higher health care costs, and we don’t really know that it meaningfully reduces the number of cancers and deaths. The authors note that the increase in ADR is “primarily attributable to detection of diminutive to small polyps with low potential for progression.” Authors Benjamin Lebwohl and Shahnaz Sultan told me by email that there is a long time lag between finding these extra polyps and the potential benefit, so these data will not be available soon. Short of hard data on mortality, evidence that AI helps find more problematic polyps might sway the decision-makers.

The guideline itself is actually an interesting read. Apparently the authors released a draft recommending AI-assisted screening but ultimately walked it back after receiving comments. The non-recommendation was arrived at by vote because there was no consensus on the panel. The guideline will be revisited in a year or two when there is new data available.

Nvidia announces platform for AI-powered robots in health care and more

This week Nvidia held its GTC event, and consequently, a few big names in health care got to briefly bask in the electronic chip maker’s limelight. As the main supplier of chips used to train large language models and other AI technologies, Nvidia props up, in one way or another, a lot of the AI tech we talk about.

Announcements were lofty and occasionally inscrutable:

• Buried in Microsoft’s press release was a quote from Epic Systems VP of cloud and analytics Drew McCombs who said the company plans to run open source generative AI models on an offering from Nvidia and Microsoft. He also noted that the company is working with UW Health and UC San Diego Health on “researching methods to evaluate clinical summaries with these advanced models.” Collaboration between the medical centers, Microsoft and Epic goes back to 2023.
• Elsewhere, GE HealthCare announced it’s working with Nvidia to “reimagine diagnostic imaging with autonomous X-ray and ultrasound solutions.” The company said it hopes that Nvidia’s new platform for AI-powered health care robots and synthetic data will “help to automate repetitive tasks performed by a technologist in the patient exam room.” The company cites this effort as way to address radiology staff shortages.
• While we’re on Nvidia, earlier this month, at the TD Cowen Healthcare Conference, the chip giant’s VP of healthcare Kimberly Powell pitched a vision for how life science research companies — such as its partners Genentech and Relation Therapeutics — can use NVIDIA’s BioNeMo platform “in the loop” to squeeze the most value out of data and their scientists’ knowledge: “Every single experiment that you do, every idea your scientist has, can now be captured and codified to build upon an institutional knowledge that otherwise is maybe sitting in an electronic lab notebook somewhere. But how could you represent even those decades of electronic lab notebooks in a model, and then every single experiment you make, put it back in? And so, we’ve invented a platform called BioNeMo to serve just that.” (Brittany Trang contributed to this item.)

Medicare advisors suggest telehealth isn’t fading

As expected, the stop-gap funding bill passed by Congress last week included an extension of Medicare coverage for telehealth. Brace yourself, we’ll be talking about this again in September.

The day before that bill passed, the Medicare Payment Advisory Commission released its March report to Congress. It noted that while some changes to the pandemic care landscape have reverted, others, like telehealth, may prove longer lasting. They note that in their 2024 survey of beneficiaries, 33% reported using telehealth, adding that as more claims from outside the pandemic become available they’ll continue to monitor the impact.

Survey responses don’t necessarily reflect reality or even Medicare cost impact. As an interesting counterpoint, the most recent data published by Medicare says that in 2023, 24% of beneficiaries used telehealth at least once.

Oura, EpiWatch, and the legacy of Apple’s ResearchKit

This week, an app that uses the Apple Watch for detecting seizures announced clearance from the Food and Drug Administration. The summary documents haven’t been posted to the FDA website yet. Originally developed by at Johns Hopkins, EpiWatch was one of the very earliest research projects launched on Apple’s ResearchKit platform in 2015. EpiWatch been spun out into a company that’s developing care programs around the detection tech. This is a similar trajectory to a Parkinson’s research project announced in 2015, which has also blossomed into a few companies.

If you go back to that 2015 ResearchKit press release above, you’ll see a quote from Ricky Bloomfield about Autism & Beyond, a study using Apple’s tech to study young children’s mental health in hopes of developing screening tools for autism and other conditions. Bloomfield was then a director of mobile technology strategy and assistant professor at Duke University.

In 2016, Bloomfield went to Apple where he worked for over eight years. Until now. Today, he was announced as the new chief medical officer of smart ring maker Oura. The hire comes as Oura seeks to evolve from a wellness fad to a player in the health care space. Oura in particular cites Bloomfield’s work on data interoperability at Duke as something they’re excited about.

A hospital at home merger

Medically Home and DispatchHealth announced a plan to merge. The former is one of the key companies helping health systems to provide hospital at home services. A pandemic-era Medicare program that allowed hospitals to deliver inpatient care in people’s homes was last week extended until September, though, I was told the timing is coincidental. DispatchHealth also provides hospital at home care, but it’s biggest and longest running service delivers same-day at-home care for serious and chronic medical conditions that helps avoid trips to the emergency department.

Amazon exits

After abruptly leaving his post as chief medical officer at Amazon Pharmacy in February, Vin Gupta revealed he’s joining consultancy Manatt, along with former CDC director Mandy Cohen, as the firm aims to build a public health communications practice.

Earlier this month, Trent Green, CEO of Amazon’s One Medical said he would be leaving the company in April to take a new job as the CEO of analytics company National Research Corp. He held the One Medical CEO job for less than two years.

What we’re reading

• HHS agency responsible for health care quality research threatened with mass layoffs, STAT
• Healthtech VCs look to AI for next wave of returns, WSJ
• ‘We’ve vanished’: U.S. aid cuts leave health workers around the world reeling, STAT