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.
Earlier this week, the Peterson Health Technology Institute (PHTI) issued a report that assesses health systems’ experience with AI ambient scribes so far and zeroing in on the technology’s “financial and operational implications.” The report says about 60 scribes are adopted in the market, and while many pitch themselves as time savers, the PHTI finds that evidence of time savings is limited. The report also explores how scribes may result in higher health care costs. My colleague Brittany Trang brings us this cheat sheet:
- PHTI made some great tables summarizing the companies in the AI scribe space and the studies that have been conducted so far on them. But if you’re short on time, skip to the “Looking Ahead” section on page 28, which covers the basics: Health systems adopt AI scribes to address provider burnout, which is why hardly anyone has put any effort into measuring any other sort of return on investment. But as the technology gets more popular, health systems will need to justify the spending.
- The report warns in multiple places that AI tools could increase costs in several ways: “On the one hand, enhanced documentation quality could lead to higher reimbursements, potentially offsetting expenses — but also leading to unintended downstream consequences for patients and the market overall. On the other hand, the cumulative costs of the software may be greater than any savings achieved through improved efficiency, reduced administrative burden, or reduced clinician attrition,” write the authors.
- PHTI notes that the industry “is grappling with the immense potential of these tools and the evolving understanding of their strengths and weaknesses. Critically, there is a need for more standardized methodologies and metrics to understand performance across a range of indicators and for more research to understand their long-term impact on efficiency.” The chart above describes some metrics that might be used in evaluation as well as an early look at what the evidence tells us today.
‘Black boxy’ telehealth-pharma partnerships
As drugmakers like Pfizer and Eli Lilly turn to telehealth platforms as a way to get their treatments to patients, the practice is drawing scrutiny from Senators who want to determine whether the relationships violate the federal anti-kickback statute. But the deals also caught the eye of academics Ateev Mehrotra, Olivier Wouters, and Erin Fuse Brown. In a new paper in the New England Journal of Medicine, the Brown University researchers explore how the partnerships might increase access to care but may also lead to inappropriate prescribing.
STAT’s Katie Palmer caught up with Mehrotra and Wouters who explained the challenge of assessing the impact of the “black boxy” deals without more information. Read more here
A diagnostic chatbot and other new ‘breakthrough’ devices
We updated our tracker of experimental medical devices that have received “breakthrough” status from the Food and Drug Administration because they may offer more effective treatment or diagnosis than the current standard of care. The designation can ease a device’s path to marketing authorization and can help with publicity and generate interest from investors. As STAT’s Lizzy Lawrence writes, at least 15 manufacturers announced a breakthrough designation between December and mid-March.
New breakthrough devices include PathChat DX, a generative AI chatbot from Modella AI that helps pathologists diagnose cases. The device is based on a model developed at Mass General Brigham. Other additions: A breast cancer test, a breast cancer imaging tool, and two Alzheimer’s blood tests. Read more here
AI drug development updates
- Life sciences data company Apheris announced that its consortium of drugmakers will contribute to OpenFold3, a Columbia University-based open source protein folding model similar to AphaFold. Specifically, AbbVie and Johnson & Johnson will contribute structural data. Read Brittany’s story explaining why this is significant.
- In a study presented at a conference this week, AstraZeneca used an AI model to analyze CT scans and better predict survival rates of lung cancer patients. Developed by Altis Labs, the technology is also being tested in breast cancer and colorectal cancer scans and could aid drugmakers in designing clinical trials. Read more here
- At its big health event last week, Google teased TxGemma, a batch of large language models designed for drug development based on the work of Google DeepMind. It’s now released the models along with a paper analyzing their performance.
What we’re reading
- Vaccine critic’s apparent selection to head HHS autism study shocks experts, STAT
- In genetics, a tense coexistence of mainstream and fringe views, UNDARK
- In a first, Eli Lilly to connect patients to telehealth providers of Alzheimer’s care, STAT
link