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.
June 6, 2025
In this week’s Duke AI Health Friday Roundup: parsing AI device data in MAUDE; tracking the academic migration to Bluesky; FDA accelerates internal AI rollout; trial finds benefit for exercise after cancer treatment; no-code AI company reported to rely on unseen human efforts; effects of work requirements for SNAP; funding secured for science sleuths; older expert-system AI goes head-to-head with LLMs; much more:
AI, STATISTICS & DATA SCIENCE
- “We focus on the Manufacturer and User Facility Device Experience database—the FDA’s central tool for tracking the safety of marketed AI/ML devices. In particular, we evaluate the data pertaining to adverse events associated with approximately 950 medical devices incorporating AI/ML functions for devices approved between 2010 through 2023, and we find that the existing system is insufficient for properly assessing the safety and effectiveness of AI/ML devices.” A paper published in NPJ Digital Medicine by Babic and colleagues takes stock of data on adverse events amassed in the FDA’s MAUDE database for postmarket evaluation of AI/ML-enabled medical devices (H/T @smcgrath.phd).
- “Two employees familiar with the situation, speaking on condition of anonymity for fear of retribution, told STAT that the tool is based on Anthropic’s Claude LLM and is being developed by consulting firm Deloitte. They also said that the tool has only been piloted in text summarization use cases, such as summarizing meeting minutes, and should only be used for daily administrative tasks, not scientific ones.” In an article for STAT News (subscription required), Brittany Trang reports on the accelerated roll-out of large-language model AI tools at FDA.
- “Bloomberg notes that these allegations initiated a cascade of investor apprehension, internal changes, and an eventual erosion of confidence. Adding to the troubles, Linas Beliūnas, Director of the financial company Zero Hash, recently exposed that Builder.ai lacked true AI, instead utilising a group of Indian developers who were merely pretending to be bots writing code.” In a somewhat startling story in the International Business Times, Vinay Patel reports that Builder.AI, a “no-code” AI startup that attracted significant investment and was valued at one point at more than $1B, is now filing for bankruptcy as its “AI” functionality is revealed to rest on hundreds of human programmers playing the role of coding bots.
- “In this diagnostic study comparing the performance of a traditional DDSS [diagnostic decision support systems] and current LLMs on unpublished clinical cases, in most cases, every system listed the case diagnosis in their top 25 diagnoses if laboratory test results were included. A hybrid approach that combines the parsing and expository linguistic capabilities of LLMs with the deterministic and explanatory capabilities of traditional DDSSs may produce synergistic benefits.” A research article published in JAMA Network Open by Feldman and colleagues presents findings from a study that compared the performance of an older-generation AI “expert system” vs large language model AIs for clinical decision support (H/T H/T Lola Ogunyemi@omololaogunyemi.bsky.social).
- “High study design standards are essential in AI research to train clinical prediction models that positively influence patients and healthcare. A key aspect of this process involves selecting appropriate sample sizes for model training and evaluation to ensure that predictions and model performance estimates are sufficiently precise to guide clinical decision making and patient–doctor discussions.” A viewpoint article by Riley and colleagues, published online ahead of press in Lancet Digital Health, highlights the importance of sample size in training predictive AI models used to guide clinical decision-making.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Exercise significantly reduced the relative risk of disease recurrence, new primary cancer, or death by 28%. The disease-free survival curves began to separate at about 1 year and continued to separate over the 10-year follow-up, with an absolute between-group difference of 6.4 percentage points at 5 years. Moreover, exercise reduced the relative risk of death by 37%. The overall survival curves began to separate at about 4 years and continued to separate over the 10-year follow-up, with an absolute between-group difference of 7.1 percentage points at 8 years.” A research article published by Courneya and colleagues in the New England Journal of Medicine presents results from a randomized trial that evaluated the effects of “structured” exercise on patients who had received adjuvant chemotherapy for colon cancer.
- “Our county-level dataset complements the state- and national-level CDC data, confirming a widespread decline in MMR vaccination rates in the US after the COVID-19 pandemic while revealing significant heterogeneity in vaccination patterns within and across states. This dataset can be used in spatial and statistical analyses to identify factors associated with low or declining MMR rates in US counties and help inform targeted vaccination strategies to reduce the risk of measles outbreaks.” A research letter published in JAMA by Dong and colleagues examines recent trends in measles, mumps, and rubella (MMR) vaccination rates for US children.
- “And now they’ve taken that a step further, showing that applying this liver immunotherapy technique to known respiratory allergens in sensitized mice dramatically protects them from asthma symptoms, even up to what would otherwise be life-threatening exposures….I don’t think there’s been anything like this before, in the way that this treatment just seems to abolish the allergic asthma response after just two treatments.” In a post at his In the Pipeline blog, Derek Lowe describes a new approach, now being tested in mouse models, that may hold promise from treating allergic asthma – albeit via an unusual route.
- “Harmful algal blooms develop when phytoplankton—microscopic, single-celled, photosynthetic algae that proliferate in warm, nutrient-rich waters—reproduce uncontrollably. These blooms disrupt ecosystems, consume and deplete oxygen, and produce potent toxins that can negatively affect human health. Harmful algal blooms are increasing in frequency and severity, and expanding toward the North and South Poles, posing challenges for clinicians, particularly during the onset of disease outbreaks from algal blooms, when surveillance and clinical testing options for such poisonings are lacking.” A JAMA Insights article by Semenza and colleagues explores potential threats to human health that may emerge or intensify due to warming oceans.
COMMUNICATION, Health Equity & Policy
- “….We uncover a striking asymmetry whereby information sources drive migration far more powerfully than audience, with this influence decaying exponentially within a week. We further develop an ego-level contagion classifier, revealing that simple contagion drives two-thirds of all exits, shock-driven bursts account for 16%, and complex contagion plays a marginal role. Finally, we show that scholars who rebuild a higher fraction of their former Twitter networks on Bluesky remain significantly more active and engaged.” A bibliometric analysis by Quelle and colleagues, available as a preprint from arXiv, examines patterns that have emerged as academics migrate between the Twitter/X social media platform and the newer Bluesky.
- “While work requirements in SNAP were associated with immediate coverage losses, removing work requirements was not associated with a change in enrollment. The enrollment patterns were similar in towns with and without a policy reversal. One potential explanation is that households automatically lost coverage when work requirements were imposed via procedural terminations but needed to be aware of a policy that eliminated work requirements and required a manual process of reapplying.” An analysis by Factor and colleagues, published in JAMA Health Forum, examines the effects of work requirements being added to qualifications for receiving Supplemental Nutrition Assistance Program (SNAP) benefits in nearly 90 US towns.
- “In the aftermath of the ‘AI boom,’ the report examines how the push to integrate AI products everywhere grants AI companies – and the tech oligarchs that run them – power that goes far beyond their deep pockets. We need to reckon with the ways in which today’s AI isn’t just being used by us, it’s being used on The report moves from diagnosis to action: offering concrete strategies for community organizers, policymakers, and the public to change this trajectory.” A 2025 landscape report, now available from New York University’s AI Now Institute, examines the effects of rapid AI adoption that are now diffusing through society.
- “The project, which has a US$900,000 grant from funder Open Philanthropy in San Francisco, California, to run for two years with a team of three to five people, aims specifically to root out flawed papers that have a serious impact on medical guidelines by skewing meta-analyses — reviews that combine the results of multiple similar studies to come to a statistically more powerful conclusion.” Nature’s Nicola Jones reports on a new scientific integrity project that will prioritize digging up dodgy science that may be exerting excessive sway on influential syntheses of data, such as clinical practice guidelines.
