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 18, 2026

In this week’s Duke AI Health Friday Roundup: reflecting on lessons from a randomized trial of AI; AI at heart of solutions, dispute involving fluid dynamics equations; recombinant antivenom protects mice against different snake species; synthetic validity for AI in social science research; weighing the benefits of ambient scribing in the emergency department; survey scopes out how medical school labs are using AI; much more:

AI, STATISTICS & DATA SCIENCE

Three human silhouettes are depicted mid-stride, each comprising of a collage of hardware used to develop AI including circuit boards and CPU chips. In the background there is a black and grey gradient with smoke. Beneath the runners from the left, there are two rectangle panels which run across the image, one is red and the other is blue. Image credit: Gloria Mendoza / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
Image credit: Gloria Mendoza / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
  • “Looking back, this journey has left me optimistic on the contribution of AI in medicine…. Humans and AI have complementary strengths and weaknesses. Humans contribute contextual understanding, adaptability and holistic perception, whereas AI provides consistency and vigilance. The challenge is therefore not to determine which is superior but to design systems in which each compensates for the other’s limitations.” In a commentary for Nature Medicine, Kristina Lång reflects on her experiences with an early randomized trial of an AI tool in a diagnostic application.
  • “There is clearly some evidence that AI-informed and delivered interventions can be effective, but effect sizes are relatively small and hard to predict, with limited ability to generalize beyond the context of the individual studies and uncertainty about what specifically is contributing to any effects. However, it is exactly in this area of strengthening the advance of a cumulative science of behavior change that we believe AI can truly make a difference…” In a perspective article published in NEJM AI, Michie and colleagues argue that the evidence base for AI applications in behavioral health requires shoring up to be truly useful.
  • “…we argue that AI presents a more insidious risk for science than hallucinations—namely, AI-generated content that resembles the concept under study rather than accurately representing it…if a communication scholar asks AI to “write a persuasive message tailored to introverts,” it will produce something in response. Whether that message reflects psychologists’ nuanced understanding of introversion—or just a stereotype about staying home on a Friday night—is another matter entirely.” An essay published in PNAS by Teeny, Lutrell, and Lee examines the issue of “synthetic validity” in the context of social science research involving AI.
  • “In three randomized controlled trials conducted in Indonesia, Kenya, and the Netherlands, we found that providing physicians with access to GPT-4o improved their clinical vignette scores by 10.7 percentage points in Indonesia, 18 percentage points in Kenya, and 7.2 percentage points in the Netherlands compared to physicians without access. This study provides the first multi-country experimental evidence that LLM augmentation may enhance physician performance across diverse economic and healthcare contexts.” A research article published in NPJ Digital Medicine by Rounding and colleagues presents results from a series of randomized, multi-national trial that assessed the degree to which physicians relied on an LLM tool for clinical decision support.
  • “Mathematician Tristan Buckmaster of New York University and Levent Alpöge, a researcher affiliated with Anthropic, one of OpenAI’s major competitors, had made significant progress solving the problem over the past month using AI tools adapted from OpenAI and Anthropic. But before they could publicize their findings, OpenAI mounted its own colossal effort—pouring millions of dollars of computational resources into the problem.” A Science feature article by Celina Zhao and Adrian Cho unpacks the emerging controversy over credit for a claimed solution to a long-intractable problem in mathematics – the Navier-Stokes equations that describe fluid dynamics.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

Close up photograph of a coiled-up Asian Monocled Cobra, showing distinctive black and white “spectacle” markings on its hood. Image credit: David Clode/Unsplash
Image credit: David Clode/Unsplash
  • “Current equine plasma–derived antivenoms suffer from variable efficacy, safety concerns, and limited cross-neutralization of geographically diverse venoms. Here, we report a rationally designed recombinant antivenom composed of five nanobodies (VHHs) that target the α-neurotoxins, cytotoxins, and phospholipases A2 prevalent in the venoms of cobra and king cobra species from India.” A research paper published in Science Translational Medicine by Samanta and colleagues report on the creation of a genetically engineered ‘broad-spectrum’ antivenom that in experiments protected mice from the venom of multiple cobra species.
  • “The new study homes in on a type of dendritic cells, immune cells that, in effect, give the killer T cells their orders. After cutting up the invader’s proteins, the dendritic cells present the antigens as bite-size pieces that T cells can recognize. T cells with compatible receptors can then become activated, which causes the cells to start to clone themselves, attack cells bearing the antigen, and cause inflammation.” In a feature article published in Science, Jennie Erin Smith explores recent evidence pointing to the possibility of links between the immune system and neurodegenerative diseases such as Alzheimer’s.
  • “The US declared measles eliminated in 2000 after blocking domestic transmission over a 12-month period via high vaccination rates. With vaccination rates slipping well below the target of 95 percent, measles is roaring back, and the US is on track to lose its elimination status.” Ars Technica’s Beth Mole reports on the growing toll of the ongoing Pennsylvania measles outbreak.

COMMUNICATIONS & Policy

The image shows a superimposition of colourful illustrations representing different objects in a secretariat: in the background, printed minutes. Above, the hands of a medical secretary typing on her keyboard, the hands of another stamping envelopes. The headset and foot pedal in the foreground are essential tools for typing. In the foreground, the red zig-zags typical of ambient scribe errors streak across the image. Image credit: Fanny Maurel & Digit / https://betterimagesofai.org / CC-BY 4.0
Image credit: Fanny Maurel & Digit / https://betterimagesofai.org / CC-BY 4.0
  • “Mass General Brigham’s community emergency medicine division chief Melisa Lai-Becker told STAT that there’s ‘something priceless’ to relieving the cognitive load for an emergency physician, especially when it can decrease interruptions and let physicians focus on a task during a shift when they might have dozens of patients to care for. But it’s unclear where the actual time saved with the AI scribe is going.” An in-depth report by STAT News’ Brittany Trang examines the lessons learned from the deployment of AI-assisted ambient listening in emergency departments for clinical documentation – and whether the technology has meaningfully improved outcomes for patients and physicians.
  • “AI education in medicine remains fragmented, narrowly technical, and disconnected from healthcare realities. Through an international collaboration, we developed a framework spanning seven domains and 24 learning objectives….AI education must be clinically grounded, ethically integrated, and implementation-aware across the training continuum.” A perspective paper published in NPJ Digital Medicine by Rosella and colleagues presents recommendations for incorporating AI into medical education.
  • “…one WISeR vendor warned the Centers for Medicare and Medicaid Services that it was unrealistic to expect a working product by the time the federal agency wanted to launch the program….The poorly implemented technology resulted in canceled and postponed surgeries and ‘patients calling our offices crying in pain because their procedures are being delayed while awaiting approvals or guidance,’ providers said.” STAT News’ Brittany Trang reports on assertions that a recent Medicare pilot designed to use AI to ease the burdens of prior authorization may not have been ready for prime time.
  • “Like people in many professions, however, medical scientists are testing the potential of AI with caution. The 2024 Ithaka S+R survey found that only 7% used AI regularly at that time, most often for such basic tasks as drafting text, editing, and searching scholarly literature. The biggest barriers to greater adoption of AI by scientists, according to the report, ‘are concerns about the accuracy of generative AI and a lack of clarity about best practices for using AI effectively and ethically.’” An article by Patrick Boyle, published on the website for the American Academy of Medical Colleges, examines the growing use of AI tools in biomedical research.

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