AI Health
Friday Roundup
The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.
March 28, 2025
In this week’s Duke AI Health Friday Roundup: LLM framework for more flexible robots; PROBAST updates guidelines for AI predictive models; getting a handle on future dementia incidence; finding new targets for drugs with AI; NC Medicaid pilots cost more at first, then yield savings; 23andMe bankruptcy leaves consumers and researchers weighing options, risks; study queries docs’ opinions about ambient AI transcription in clinic; much more:
AI, STATISTICS & DATA SCIENCE
- “Our approach successfully combined the cognitive abilities of LLMs with the sensorimotor skills of robots, enabling our robot to interpret a high-order verbal command and execute a complex long-horizon task while adeptly managing uncertainties. We used the LLM, augmented with feedback loops and RAG [retrieval-augmented generation], to write expressive code and facilitate the manipulation sub-tasks required by the robot to achieve the high-level goal (making a hot beverage). ELLMER allowed real-time adaptation to environmental changes and leveraged a repository of precise solutions via RAG.”A research article published in Nature Machine Intelligence by Mon-Williams and colleagues describes the incorporation of large language model frameworks to equip robots with a greater degree of flexibility and adaptability to unpredictable factors in environments and tasks.
- “The lifesaving drug regimen wasn’t thought up by the doctor, or any person. It had been spit out by an artificial intelligence model….In labs around the world, scientists are using A.I. to search among existing medicines for treatments that work for rare diseases. Drug repurposing, as it’s called, is not new, but the use of machine learning is speeding up the process — and could expand the treatment possibilities for people with rare diseases and few options.” A feature article by the New York Times’ Kate Morgan explores the nascent world of AI-driven research into repurposing existing drugs for new uses.
- “Since PROBAST’s introduction in 2019, much progress has been made in the methodology for prediction modelling and in the use of artificial intelligence, including machine learning, techniques. An update to PROBAST-2019 is thus needed. This article describes the development of PROBAST+AI. PROBAST+AI consists of two distinctive parts: model development and model evaluation. For model development, PROBAST+AI users assess quality and applicability using 16 targeted signalling questions. For model evaluation, PROBAST+AI users assess the risk of bias and applicability using 18 targeted signalling questions.” In a new paper published in BMJ, Moons and colleagues introduce an updated version of the PROBAST-2019 guidelines for assessing quality, bias, and applicability of prediction models.
- “…when systems can control multiple information sources simultaneously, potential for harm explodes. For example, an agent with access to both private communications and public platforms could share personal information on social media. That information might not be true, but it would fly under the radar of traditional fact-checking mechanisms and could be amplified with further sharing to create serious reputational damage. We imagine that “It wasn’t me—it was my agent!!” will soon be a common refrain to excuse bad outcomes.” In an opinion article for MIT Technology Review, Margaret Mitchell, Avijit Ghosh, Sasha Luccioni, and Giada Pistilli sound a note of caution with regard to deploying semi- or fully autonomous AI “agents” in the real world.
- “Facilitators to adoption included ease of use, ease of editing, and generally positive physician perspectives on tool quality. Barriers to adoption included accessibility challenges, such as limited functionality with non-English speaking patients and lack of access for physicians without a compatible mobile device. Physician perspectives on accuracy and style were largely negative, particularly regarding the length of notes and the editing requirements associated with the ambient AI scribe tool. Despite these concerns, some physicians noted that the longer content might provide a more comprehensive capture of the encounter that otherwise might not have been documented.” A qualitative research article published in JAMA Network Open by Shah and colleagues explores physicians’ experiences and feelings regarding the use of ambient AI “scribes” in clinical settings.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “That is not to say the concept behind these companies—a much broader collection of data— is wrong. My contention is that it is flawed because it is indiscriminate as to who undergoes testing and what tests (such as total body MRI or multi-cancer early detection) and when they are performed. Ironically, despite what might be thought of as a ‘kitchen sink’ approach, many tests that could be particularly useful have not been incorporated…” At his Ground Truths Substack page, Eric Topol evaluates the recent emergence of companies aiming to extend human lifespan or reverse aging processes.
- “…over the last 4 decades, education levels have risen, smoking rates have dropped and treatment of cardiovascular risks and hearing loss has improved all of which may explain why dementia rates dropped. But on the other hand, rising rates of obesity, diabetes and sedentary lifestyle may lead to increased dementia rates in the future….Some 40% of the risk of dementia may be potentially modifiable so its crucial to understand how attributable risks are changing over successive generations.” In an interview published on the Duke Social Sciences Research Institute website, Duke researchers discuss the implications of a recently published paper in which they investigated projected rates of dementia incidence in the U.S.
- “In hospital settings it’s common to use patient-controlled analgesia (PCA) pumps for management of severe pain. These infuse a solution containing morphine together with nonopioid analgesics directly into a patient’s intravenous connection, and allow them to deal with their own experience of pain and their own tolerance for it….But a couple of years ago questions were raised about one commonly used mixture.” At his Science blog In the Pipeline, Derek Lowe describes recent research highlighting chemical processes in which patient-controlled analgesia pumps can, depending on the painkillers being used, form new pharmaceutically active compounds under certain conditions.
COMMUNICATION, Health Equity & Policy
- “It’s not yet clear what the sale will mean for the academic and industry scientists who have collaborated with the company and relied on its data for research on human health and evolution. And some are calling for the company’s clients to safeguard their data. Last week, responding to securities filings from 23andMe indicating financial trouble, California Attorney General Rob Bonta issued a consumer alert asking Californians to “consider” having the company delete their genetic data and destroy samples of genetic material, and instructing them on how to do so.” Science’s Kelly Servick reports on consumer-oriented genetic testing company 23andMe entering bankruptcy, and what the means for the millions of people whose genetic information resides in the company’s databases, now up for sale.
- “My dad spoke up first. He said that when the doctors had told Mum she was “in heart failure,” she thought she was about to die. That had sent her into a tailspin, bordering on panic….She was feeling much better, breathing easily and back in sinus rhythm. In fact, she felt better than she had in weeks, if not months. Why, then, were the doctors intimating with their diagnosis that she was about to die?” A perspective article in the New England Journal of Medicine by Jason Gurney highlights the hidden role that choice of descriptions can play when communicating with patients and their families.
- “The clawed-back funds were largely being used for Covid testing, vaccination, community health workers and initiatives to address Covid health disparities among high-risk and underserved populations, including racial and ethnic minority populations and rural communities, as well as global Covid projects, according to talking points CDC leadership emailed to agency departments Tuesday.” NBC News’ Brandy Zadrozny reports that the Centers for Disease Control and Prevention is taking back more than $11B in funding allocated to state and local health departments.
- “…we found that the time of HOP enrollment was one of initially greater spending, but that subsequently there was a significantly decreasing trend in spending relative to what would have been expected in the absence of the intervention. Overall, these observations are consistent with an association between health-related social needs and triggering events that lead to greater need for expensive health care services, and with an association between services to address those needs and less subsequent spending.” A research article by Berkowitz and colleagues, published in JAMA, reports findings from a study of Medicaid Healthy Opportunities Pilot programs in North Carolina.
