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 17, 2026

In this week’s Duke AI Health Friday Roundup: Stanford 2026 AI Index report shows public, expert view of AI diverging; why we’re drawn to scary stories about AI; rethinking cancer care; dubious data sets may have been used to train predictive models; rates of youth-onset T2DM increased over a decade; ancient origins for immune system’s antiviral tools; are preprints recapitulating disappointments from the open-access movement?; much more:

AI, STATISTICS & DATA SCIENCE

“Assessing AI’s impact on jobs, 73% of U.S. experts are positive, compared with only 23% of the public, a 50 percentage point gap. Similar divides emerge with respect to the economy and medical care. Globally, trust in governments to regulate AI varies. Among surveyed countries, the United States reported the lowest level of trust in its own government to regulate AI, at 31%.” Stanford’s Human-centered AI institute has released its 2026 state-of-the-AI-industry report, and among the many findings of note are data that indicate a growing gulf between public and expert attitudes about AI technologies in the US. Some additional commentary by Nature’s Nicola Jones homes in on the report’s findings about AI’s role in bench science settings.
“Our findings show that positively- and negatively-framed pairs are significantly more likely to produce contradictory conclusions than same-framing pairs. This framing effect is further amplified in multi-turn conversations, where sustained persuasion increases inconsistency….Our results demonstrate that LLM responses in medical QA can be systematically influenced through query phrasing alone, even when grounded in the same evidence, highlighting the importance of phrasing robustness as an evaluation criterion for RAG-based systems in high-stakes settings.” A preprint article by Yun and colleagues, available from arXiv, reports results from a study that finds LLM outputs in response to queries about health topics are sensitive to variations in phrasing (H/T @smcgrath.phd).
“Adrian Barnett, a statistician at the Queensland University of Technology in Brisbane, Australia, and his colleagues identified 124 peer-reviewed papers that report using one of two open-access health data sets to train machine-learning models that provide little information about where the data came from….An analysis revealed multiple oddities that would not be expected for data from real people, leading Barnett and his colleagues to suspect that the data could have been fabricated.” Nature’s Mohana Basu reports on a recent study, available as a preprint from medRXiv, that found suspicious patterns in data used to train clinical prediction models.
“In multi-center evaluations, Brainfound demonstrates state-of-the-art performance across seven tasks, including brain disease diagnosis, lesion segmentation, MRI enhancement, cross-modality translation, automatic report generation, zero-shot disease classification, and human-AI dialogue. It substantially outperforms leading models in automated report generation and clinical question answering for brain imaging, and its performance approaches that of expert physicians.” A research article published in Cell Patterns by Zhang and colleagues presents a new multimodal AI model for interpreting CT and MRI brain scans.
“To address these growing inequities and systemic pressures, Mayo Clinic piloted an innovative model of home-based chemotherapy through its Cancer CARE (Connected Access and Remote Expertise) Beyond Walls program — CCBW. This program reimagines oncology care by integrating virtual oversight, remote patient monitoring (RPM), mobile health services, and a unified software platform connected to the electronic medical record.” An article by Dronca and colleagues appearing in NEJM Catalyst describes a novel pilot approach to home-based cancer care that utilizes digital health tools and remote monitoring.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

“With modern trial designs, strengthened governance, and fair liability frameworks, inclusion can be both responsible and rigorous. The choice is no longer between protection and participation, but between managed risk with evidence and unmanaged risk without it.” An editorial by Ali and colleagues, published in BMJ, makes the case for expanding the participation of pregnant and breastfeeding women in clinical trials.
“In recent years researchers have come upon a surprising finding: Some of the machinery that bacteria use to defend against phages exists, almost unchanged, in our own cells. According to dozens of discoveries made over the past decade, the rules of engagement between cells and viruses were written billions of years ago and still largely define how our innate immune system, the first responder to infection, defends us against viruses and bacteria today.” Quanta’s Viviane Callier reports on recent research that sheds new light on the paths by which our innate immune system evolved its defenses against viral infections.
“…we show that young APOE4 knockin (E4-KI) mice exhibit hippocampal region-specific network hyperexcitability that predicts later cognitive deficits. This early phenotype arises from cell-type-specific subpopulations of smaller, hyperexcitable neurons and is eliminated by selective removal of neuronal APOE4.” A research article published in Nature Aging by Tabuena and colleagues implicates gene-induced neuronal “hyperexcitability” as a possible culprit in the development of Alzheimer disease.
“Our findings from a national sample of youths indicate that the burden of youth-onset type 2 diabetes has increased substantially during the past decade. Enhancement of early identification and prevention in high-risk children and adolescents may help slow this growing public health challenge.” A research letter published in the New England Journal of Medicine by Fang and colleagues reports on registry data showing a marked increase in rates of type 2 diabetes among US youth.

COMMUNICATIONS & Policy

“…why would a company make their product sound scarier than it is? Perhaps because this is the best advertising money can’t buy. People like Harari and others repeat these accounts like ghost stories around a campfire. The public, awed and afraid, marvels at the capabilities of AI.” In an essay for Quanta’s Qualia series, Amanda Gefter dissects some of the reasons behind our fascination with scary visions of an AI-dominated future.
“Right now, the preprint movement sits on the precipice, and the decisions made in the next few years will prove pivotal in determining whether preprints become a stable, community-governed part of the scholarly ecosystem or drift into the same patterns of fragmentation, inequity, and co-option that have complicated the open access transition.” A guest post by Jonny Coates at Scholarly Kitchen examines some shortcomings of the open access publishing revolution (also discussed here in a recent unrelated article) and asks whether preprints are headed for a similar set of problems.
“Artificial intelligence tools are accelerating manuscript production far faster than peer review capacity can expand. Applying the theory of constraints from manufacturing science, we formalize this asymmetry through a minimal two-variable ordinary differential equation model coupling review queue evolution and verification quality degradation via an endogenous, queue-pressure-driven review AI adoption mechanism.” A research article by Kwon, available at arXiv, offers a mathematical description for the growing mismatch between AI-driven publication volume and the capacity to meaningfully peer-review it.