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.
March 17, 2023
In today’s Duke AI Health Friday Roundup: GPT4 AI debut wows users, raises questions; US maternal deaths continue to climb; early lower-respiratory tract infections have long-term consequences; transformer model predicts hundreds of millions of protein structures; revisiting the evidence for masking; portable MRIs possible; transparency practices in AI; benefits and burdens of open-access publishing; building healthy campuses; much more:
AI, STATISTICS & DATA SCIENCE
- “It is a step backwards for science, because it sets a new precedent for pretending to be scientific while revealing absolutely nothing. We don’t know how big it is; we don’t know what the architecture is, we don’t know how much energy was used; we don’t how many processors were used; we don’t know what it was trained on etc.” A new and significantly larger version of OpenAI’s large language model, dubbed GPT-4, has debuted this week, and Gary Marcus has some initial thoughts. The so-called multimodal AI – which can parse images as well as text – is the first iteration of the GPT model to be released since OpenAI became part of a for-profit venture, and details about its inner workings, including training data and weighting, have not been released to the public.
- “As language models of protein sequences are scaled up to 15 billion parameters, an atomic-resolution picture of protein structure emerges in the learned representations. This results in an order-of-magnitude acceleration of high-resolution structure prediction, which enables large-scale structural characterization of metagenomic proteins. We apply this capability to construct the ESM Metagenomic Atlas by predicting structures for >617 million metagenomic protein sequences, including >225 million that are predicted with high confidence…” A research article by Lin and colleagues, published this week in Science, describes the creation of a transformer language model capable of predicting hundreds of millions of protein structures.
- “…simply put, in order to understand an AI system, we need to understand what it is designed to do. In the case of an AI system, there are certain characteristics that are especially critical to understanding the system design. These include understanding the data on which the system was trained, the features of the data, and the objective of the system.” The Mozilla Foundation has released a report on the current state of transparency in artificial intelligence, which includes both short- and longer-term recommendations for creating a pathway to meaningful and useful transparency in AI.
- “…teams consistently describe responses to unexpected challenges encountered in the transition from conducting AI software research to translating a technology into practice. The success of AI in healthcare hinges on the ability to adapt and transition from research into routine clinical practice. Sharing challenges, failures and describing promising approaches that were implemented in real-world settings can inform teams around the globe looking to advance the use of AI software in healthcare.” An editorial published in Frontiers in Digital Health by Sendak and colleagues examines best practices for developing and deploying AI applications in healthcare settings.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Masks and mask mandates have been a hot controversy during the pandemic. The flawed summary — and further misinterpretation of it — set off a debate between those who said the study showed there was no basis for relying on masks or mask mandates and those who said it did nothing to diminish the need for them.” In an op-ed for the New York Times, UNC sociologist Zeynep Tufekci revisits the science around the use of masks for preventing COVID infection and transmission, in light of a series of corrections and clarifications around a recently released Cochrane study of the intervention.
- “In 2021, there were 32.9 deaths per 100,000 live births, compared with 23.8 per 100,000 in 2020 and 20.1 in 2019, the National Center for Health Statistics reports March 16. The U.S. rate greatly exceeds those of other high-income countries. The total number of U.S. maternal deaths rose from 861 in 2020 to 1,205 in 2021.” At Science News, Aimee Cunningham reports on recently released CDC data (note: public version of data not available until 3/19/2023) showing that the US maternal mortality rate, already at a concerning level, is continuing to climb, and the disparity in risk for Black women continues to worsen.
- “…this is the first prospective study to quantify the relationship between early childhood LRTI and premature adult death from respiratory disease, showing both the long-lasting sequalae of early childhood respiratory infection and the childhood origins of premature adult respiratory mortality…. By showing such a substantial link to early childhood LRTI, this study shows there is scope for much earlier preventive interventions and provides a more rounded explanation for existing inequities in survival.” A research article published this month in Lancet by Allinson and colleagues reports findings from a cohort study examining the long-term ramifications of lower respiratory tract infections acquired in early childhood.
- “For years, some researchers have been striving to build scanners that use much smaller permanent magnets, made of the alloy often found in desk toys. They produce fields roughly 1/25th as strong as a standard MRI magnet, which once would have been far too weak to glean a usable image. But, thanks to better electronics, more efficient data collection, and new signal processing techniques, multiple groups have imaged the brain in such low fields—albeit with lower resolution than standard MRI.” In a news article for Science, Adrian Cho reports on recent innovations and methodological refinements that may make smaller, cheaper, portable MRI machines a practical reality.
COMMUNICATION, Health Equity & Policy
- “Both ground up and top-down approaches are needed to create a cultural shift toward health as a shared value and movement toward healthier and more equitable campus communities. The findings from this research illustrate the importance of integrating health and well-being practices into educational and work settings, daily operations, and business practices.” In a research article published in the journal Health Education and Behavior, Cheney and colleagues examine academic approaches to cultivating healthy university campuses.
- “Open access is not only desirable, but necessary to ensure that the full benefits of research can translate into societal gains. Indeed, equal access to scientific output is recognized by UNESCO as essential for human development and critical for achievement of the Sustainable Development Goals. With this in mind, the transition to open access should involve special care to avoid exclusion of authors owing to the financial burden of APCs.” A guest essay at Scholarly Kitchen by Kowaltowski and colleagues explores the cost burdens imposed on some researchers by the page charges associated with open access publishing.
- “The agency said in a fact sheet that accompanied the announcement that the EPA decided to regulate PFNA (perfluorononanoic acid), GenX chemicals (hexafluoropropylene oxide dimer acid and its associated ammonium salt), PFHxS (perfluorohexane sulfonic acid), and PFBS (perfluorobutane sulfonic acid) together under a hazard index because recent peer-reviewed science ‘indicates that mixtures of PFAS can pose a health risk greater than each chemical on its own.’” STAT News’ Brittany Trang reports on the Environmental Protection Agency’s proposal to establish limits for concentrations of PFAS “forever” chemicals in drinking water.
- STAT News’ Ed Silverman reports that, following an initial announcement by Lilly, both Novo Nordisk and and Sanofi are following suit in lowering the prices of insulin products.
