AI Health Roundup – September 5, 2025

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.

September 5, 2025

In this week’s Duke AI Health Friday Roundup: AI video analysis illuminates neuromuscular diseases; authors may be under-reporting their use of AI in research; a different angle on AI and physician deskilling; as thumbs get bigger, so do brains; pediatric cancer trials network loses funding, stop enrollment; analysis reveals shady publication network; trial evaluates effects of ultra-processed foods; much more:

AI, STATISTICS & DATA SCIENCE

An anonymous figure examines handwritten notes overlaid with a blue light projection resembling a virtual reality interface. The projection brings the notes to life, highlighting details such as a reference to the intricate ceiling of King’s College Chapel and a diagram of a man encircled by a series of numbers, blending historical imagery with futuristic technology. Image credit: Hanna Barakat & Cambridge Diversity Fund / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
Image credit: Hanna Barakat & Cambridge Diversity Fund/Better Images of AI/CC-BY 4.0
  • “My concern about deskilling in many parts of AI and healthcare today relates not so much to technology failures or the need to urgently toggle to the old way of doing things. Rather, it’s that AI penetration is likely to be spotty for the foreseeable future, and that today’s AI tools are often going to be far from perfect.” A recent entry at UCSF physician Bob Wachter’s Substack page includes an interesting and nuanced perspective on the recent paper, highlighted in a previous Roundup, that examined the issue of physician “deskilling” when AI-based tools were introduced into the clinical setting.
  • “In just 16 minutes per participant, they collected video recordings of nine functional movements in 129 individuals, including 86 patients with FSHD or DM. From these, they extracted 34 kinematic features with strong clinical relevance. Their approach showed excellent agreement with traditional TFTs (r>0.98) and superior ability to distinguish disease groups (balanced accuracy 82%).” In an editorial accompanying a research article by Ruth and colleagues published in NEJM AI, Nicoline B.M. Voet describes a new AI-assisted method for identifying neuromuscular disease via video analysis.
  • “The percentage of submitted articles with self-disclosed AI use was significantly lower in this study than what has been reported in recent surveys of researchers about their general use. Submitting authors may be underreporting their use of AI.” An abstract presented at the 10th International Congress on Peer Review and Scientific Publication by Al-Fayyad and colleagues provides an estimate on the extent to which the use of AI has permeated academic publishing, based on a sample of submissions to BMJ.
  • “…we constructed, tested, and implemented ML models to quantify variant penetrance for a diverse set of 10 dominant genetic conditions using large-scale EHR data. We applied these models to an independent set of individuals with linked EHR and exome sequence data to determine ML penetrance of rare variants in disease-predisposition genes….The implications of this ML-driven approach include improved variant classification, enhanced risk assessment, improved gene and variant prioritization, and augmented genetic discovery.” A research article published in Science by Forrest and colleagues debuts a machine-learning-based approach to evaluating genetic risk factors for a number of rare diseases.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

A pair of hands, one left and one right, clad in blue nitrile gloves giving a thumbs-up. Image credit: Anton/Unsplash
Image credit: Anton/Unsplash
  • “…on the basis of both our thumb length and workspace analyses, we interpret our results to indicate sustained historical coevolution between brain size and dexterity across the primate order, reflecting significant neural costs of manipulation behaviours and helping to explain the rapid increases in brain size observed in hominins…” A research article published in Nature Communications Biology by Baker and colleagues presents evidence for an evolutionary connection between dexterity (and thumb length in particular) and brain development.
  • “The Pediatric Brain Tumor Consortium, which has conducted dozens of trials since it was established in 1999 by the National Cancer Institute, has six active trials; five are related to treatment. These are early phase trials evaluating the safety and efficacy of novel therapies for pediatric brain cancers — mostly ‘the very highest-risk types’ of these cancers, said Dr. John Prensner, a pediatric neuro-oncologist at the University of Michigan, which is not a member of the consortium.” The New York Times’ Nina Agrawal reports that a pediatric cancer trials network has put new trial enrollments on hiatus after the group’s funding was withdrawn.
  • “…we found that UPF [ultra-processed food] consumption influences cardiometabolic and reproductive health. Diets also caused a trend in the differential accumulation of contaminants in blood and seminal fluid, with Li and Hg displaying a decreased accumulation and serum cxMINP displaying a trend of increased accumulation in the ultra-processed compared with the unprocessed diet. By providing a fixed number of calories to study participants during both diets, we were able to determine that the processed nature of the food itself, independent of the caloric and macronutrient intake, impacts numerous health markers.” In a research article published in Cell Metabolism, Preston and colleagues present findings from a crossover nutrition trial that explored the effects of ultra-processed foods on health.

COMMUNICATION, Health Equity & Policy

Illustration of a surreal office scene with neon birds interacting with digital elements around three people near servers and file cabinets; one bird writes on a digital mesh, another carries a paper. IceMing & Digit / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
Image credit: IceMing & Digit/Better Images of AI/CC-BY 4.0
  • “No matter what he said, the bot parroted canned answers to generic questions, not McGing’s obscure query. ‘If you do a key press, it didn’t do anything,’ he said. Eventually, the bot ‘glitched or whatever’ and got him to an agent…It was a small but revealing incident. Unbeknownst to McGing, a former Social Security employee in Maryland, he had encountered a technological tool recently introduced by the agency.” Kaiser Family Foundation Health News’ Darius Tahir reports on the rollout of a customer-service chatbot at the Social Security Administration – as well on concerns that the technology may not be ready for prime time.
  • “In a care-embedded trial across a large health care system in the United States enrolling over 3.6 million patients, a cardiovascular-focused communication strategy did not improve influenza vaccination rates or time to vaccination compared with usual care communication. Vaccination rates were statistically lower in those receiving any cardiovascular communications vs. those receiving usual care communication, though the magnitude of the differences were small.” A research article published in NEJM Evidence by Bhatt and colleagues evaluated whether messaging focused on the cardiovascular risks of contracting influenza could affect uptake of the seasonal flu vaccine in the US.
  • “Lawmakers, including committee chair Brett Guthrie (R-Ky.) and Frank Pallone (D-N.J.), worried about ensuring the safety of medical devices using AI while getting new innovations to patients quickly. Witnesses several times noted that the Food and Drug Administration’s regulatory framework is ill-suited for evaluating AI tools. In an FDA Digital Health Advisory Committee meeting last fall, regulators acknowledged as much.”  STAT News’ Brittany Trang reports on developments during the House Energy and Commerce Committee’s hearing on health AI this week.
  • “The story started with a single plagiarized paper in 2022. Pulling on that thread revealed something much bigger: more than 120 papers carrying the name of a fictional funder, Pharmakon Neuroscience Research Network, connecting over 300 authors and 230 organisations.” In a post at Forensic Scientometrics, Leslie McIntosh unpacks an investigation that revealed a shady network designed to influence scientific publishing.

Sites@Duke Express is powered by WordPress. Read the Sites@Duke Express policies and FAQs, or request help.