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.
April 25, 2025
In this week’s Duke AI Health Friday Roundup: Duke scientists use AI to create protein “shrinkray”; LLM anonymizes chart data; revisiting Mendel’s peas; funding cut for Women’s Health Study; medical education paradigm for AI regulation; no evidence of cognitive decline in older digital tech users; creating a better evidence base for digital technology effects on users; much more:
AI, STATISTICS & DATA SCIENCE
- “Singh and postdoctoral fellow Kapil Devkota, PhD, named it Raygun in homage to movies like Honey, I Shrunk the Kids, in which a scientist invents a ‘shrinkray’ that blasts a laser to miniaturize objects. ‘Kapil and I and the other co-authors are all sci-fi fans, and there is this trope of a laser gun that can shrink or expand things,’” Singh said. “And we thought, wouldn’t it be cool to have something like that in real life.’” In a story for Duke School of Medicine’s Magnify online magazine, Angela Spivey reports on work by a team at Duke headed by research Rohit Singh in adapting LLM-style technology to build custom proteins.
- “As AI tools in clinical workflow abound, it has become clear that understanding the impact of these systems on tangible targets, such as reduced health system costs or improved quality of care, is complex and often nonintuitive….Ultimately, adoption of AI systems will hinge on the improvements they offer to health systems, and a nuanced understanding of health care workflows in their entirety will be crucial to appropriately identify the bottlenecks that should be targeted for intervention.” An editorial published in JAMA Network Open by Mandair and colleagues commenting on a recent randomized trial of an AI-based oncologist-notification system examines some of the current challenges of evaluating AI technologies in clinical workflows.
- “The removal of PII [personally identifying information] from medical text data is critical to ensure the privacy of sensitive patient information, but it presents a formidable challenge for data use and sharing. In this article, we benchmarked eight locally deployable LLMs for personal identifier extraction from 250 real-world clinical letters and their subsequent anonymization. In addition, we present the LLM-Anonymizer, a comprehensive, automated pipeline that streamlines this process for secure deidentification of medical text data on local hardware and with a user-friendly interface…” A research article published in NEJM AI by Wiest and colleagues debuts a privacy-preserving LLM for extracting unstructured free text information from patient charts (H/T @smcgrath.phd).
- “In the ad, Cluely works like magic. It instantly understands the situational context and the user doesn’t have to do anything. In reality, we spend the first couple minutes troubleshooting Cluely-related audio problems. The AI can’t intuit what I need to know even though I gave it some context ahead of the call. There’s no being discreet when you have to type prompts with a clacky mechanical keyboard. The few times I try, it’s obvious my eyes are wandering to the side of my screen. And whenever I shoot off a prompt, the AI takes forever to generate a response.” The Verge’s Victoria Song takes a closer look at Cluely, a startup AI application that created a recent social-media stir with its exhortation to “cheat on everything.”
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Here we present a sequence-based population genomic map and a trait-based phenotypic variation map, coupled with haplotype–phenotype association analyses across a wide range of traits in one of the world’s major Pisum germplasm collections. We focus on elucidating the gene identities and genomic context of alleles underlying Mendel’s seven well-known traits. We further demonstrate how this approach can be expanded to uncover the molecular basis of a wide range of agronomic and horticultural characters with breeding potential.” Revisiting the classics: in an article published in Nature, Feng and colleagues explore the genetic elements underlying the phenotypic traits that Gregor Mendel tracked in his foundational experiments with pea varieties.
- “Funded by the National Heart, Lung, and Blood Institute, the WHI has been studying women’s health since 1991, an era when observational studies suggested that hormone therapy to replace falling levels of estrogen and progesterone would improve cardiovascular health, among other targets. The goal was to more rigorously test that idea in randomized clinical trials in dozens of centers across the country, and more than 161,000 women were eventually enrolled.” STAT News’ Elizabeth Cooney reports that the long-running Women’s Health Initiative study is slated for steep funding cuts amid changes at NIH.
- “Technology engagement was associated with reduced odds of cognitive decline in middle-aged and older adults. There was no credible evidence from the longitudinal studies, or the meta-analysis as a whole, for widespread digital ‘brain drain’ or ‘digital dementia’ as a result of general, natural uses of digital technology.” A meta-analysis published in Nature Human Behaviour by Benge and Scullin examines the evidence for relationships between digital technology use and cognitive aging.
COMMUNICATION, Health Equity & Policy
- “In a nutshell, we propose that if there is inadequate safety information available about a digital product — and processes don’t exist to produce that information at speed — researchers and policymakers should lower the evidence thresholds needed to start testing interventions to address potential harms. This should, in turn, accelerate both our understanding of harms and the provision of regulations.” In an essay for Nature, digital researcher Amy Orben proposes a framework for developing a more robust evidence base for the effects of digital technologies on consumers.
- “Data from the Nature Careers global science jobs platform show that US scientists submitted 32% more applications for jobs abroad between January and March 2025 than during the same period in 2024. At the same time, the number of US-based users browsing jobs abroad increased by 35%.” Nature’s Laurie Udesky and Jack Leeming explore a possible incipient “brain drain” as U.S. scientists begin to look for options abroad amid disruptions to research funding.
- “While disparities in digital health care use were observed in association with traditional individual-level indicators of disadvantage, increased digital health care use among traditionally underserved participants found in this study is promising. Individual-level characteristics showed different associations with digital health care use based on county-level social vulnerability.” A study published in JAMA Network Open by Wakeman and colleagues investigates disparities in the use of telemedicine, telehealth, and telemonitoring services.
- “An AI–CBME [competency-based medical education] life-cycle framework would address the unique regulatory challenges posed by AI by embedding safeguards for real-time validation and patient safety. Aligning regulation with proven educational frameworks promotes a system of accountability and trust while maximizing the potential of AI to improve health care. We can implement such a framework today by inviting medical educators…into the room where regulation happens.” A perspective article by Vokinger and colleagues published in NEJM AI proposes a framework for AI regulation that borrows from medical education approaches.
