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.

February 18, 2022

In today’s Roundup: why bigger is better for neural nets; REDUCE-IT eyes cost effectiveness for statin alternative; COVID’s burdens for immunocompromised; reinforcement learning yields AI that can beat humans at driving simulator; policy journal devotes issue to racial equity in healthcare; fighting smartphone addiction to boost scientific productivity; scientists not always equipped for social media furor; Califf returns to FDA leadership post; transgenic zebrafish on the loose in Brazil; much more:

Deep Breaths

Photograph of a toy orange, black, and white clownfish made of Lego blocks (or similar). Image credit: Grianghraf/Unsplash
Image credit: Grianghraf/Unsplash
  • “The biohybrid fish equipped with intrinsic control strategies demonstrated self-sustained body–caudal fin swimming, highlighting the role of feedback mechanisms in muscular pumps such as the heart and muscles.” A research article published in Science by Lee and colleagues describes the creation of a “biohybrid” – a mostly synthetic “fish” that can swim thanks to a layer of living cardiac muscle cells.


Photograph of a child (out of focus in background) playing with a pair of colorful toy racecars. Image credit: Sandy Millar/Unsplash
Image credit: Sandy Millar/Unsplash
  • “We demonstrate the capabilities of our agent, Gran Turismo Sophy, by winning a head-to-head competition against four of the world’s best Gran Turismo drivers. By describing how we trained championship-level racers, we demonstrate the possibilities and challenges of using these techniques to control complex dynamical systems in domains where agents must respect imprecisely defined human norms.” A research article published in Nature by Wurman and colleagues describes the use of reinforcement learning to train an AI capable of outcompeting human players at the racing video game Gran Turismo.
  • “As [neural networks have] gotten bigger, they have come to grasp more. This has been a surprise to onlookers. Fundamental mathematical results had suggested that networks should only need to be so big, but modern neural networks are commonly scaled up far beyond that predicted requirement — a situation known as overparameterization.” An article by Quanta’s Mordechai Rorvig delves into recent research by Sébastien Bubeck and Mark Selke to illuminate why, in spite of earlier expectations, neural networks perform better as they grow in size.
  • “…we introduce and study the disagreement problem in explainable machine learning. More specifically, we formalize the notion of disagreement between explanations, analyze how often such disagreements occur in practice, and how do practitioners resolve these disagreements.” A preprint by Krishna and colleagues, available at arXiv, tackles the problem of disagreements in explanation for the output of machine-learning models (H/T @arXiv_Daily).
  • “REDCap has built-in features that support pragmatic, internet-based studies, and REDCap is flexible enough to allow creative solutions for specific trials. We describe challenges, choices, and suggestions based on our experience with REDCap for our COVID-19 trials.” A paper published in Contemporary Clinical Trials by Bangdiwala and Boulware describes the use of Research Electronic Data Capture (REDCap) tools for conducting pragmatic, patient-reported clinical trials.
  • An article published in Methods of Information in Medicine by Schnellinger and colleagues examines approaches for detecting changes in model parameters that can affect the performance of predictive models.
  • “Metaphors are very often used to influence the audience’s opinion. This is hugely important because policymakers often use metaphors to frame and understand problems – the way you understand a problem has a big impact on how you respond to it and construct a solution.” In a blog post for Better Images of AI, Alice Thwaite cautions that the metaphors we collectively adopt to describe technologies such as AI can have far-reaching consequences.


Photograph from NIH Zebrafish Core showing striped Zebrafish swimming in illuminated aquariums. NIH/Flickr
  • “Fish genetically engineered to glow blue, green, or red under blacklight have been a big hit among aquarium lovers for years. But the fluorescent pet is not restricted to glass displays anymore. The red- and green-glowing versions of the modified zebrafish have escaped fish farms in southeastern Brazil and are multiplying in creeks in the Atlantic Forest, a new study shows.” In a feature for Science, Sofia Moutinho describes mounting scientific concern over the escape of transgenic zebrafish into Brazilian streams – and the potential for harm to the ecosystem.
  • “Méric’s team also explored which genetic variants might affect the abundance of certain microbes—and which of those variants were linked to 46 common diseases. When it came to depression, two bacteria that cause infections in hospitalized patients, Morganella and Klebsiella, seemed to play a causal role, the researchers say. One of them, Morganella, was significantly increased in a microbial survey of the 181 people in the study who later developed depression.” Science’s Elizabeth Pennisi reports on a recent study that may potentially link a species of bacteria found in some people’s gut microbiomes to depression (H/T @Duke_OSI).
  • A cost-effectiveness analysis from the REDUCE-IT trial published this week in JAMA Network Open by Weintraub and colleagues affirms the cost-effectiveness of icosapent ethyl for the treatment of high cholesterol levels that persist despite treatment with statins. These findings may be of particular interest due to the high cost of PCSK9 inhibitors, a class of drugs that have proven effective in similar scenarios but whose clinical uptake has lagged, potentially due to their expense.
  • “Ramps, accessibility buttons, screen readers, and many other measures have made life easier for disabled people, and a new wave of similar accommodations is now necessary to make immunosuppression less of a disability in the COVID era. Exactly none of the people I talked with wants a permanent lockdown….What they do want—work flexibility, better ways of controlling infectious diseases, and more equitable medical treatments—would also benefit everyone, not just now but for the rest of our lives.” At The Atlantic, Ed Yong explores the particularly acute – and prolonged – impact the COVID pandemic is having on immunocompromised persons.
  • “…beyond the first 30 [days] after infection, individuals with COVID-19 are at increased risk of incident cardiovascular disease spanning several categories, including cerebrovascular disorders, dysrhythmias, ischemic and non-ischemic heart disease, pericarditis, myocarditis, heart failure and thromboembolic disease.” A research article by Xie and colleagues published in Nature Medicine describes an elevated longer-term risk for cardiovascular problems in persons infected with COVID.


Illuminated screen of a smartphone showing various apps against a dark background. Image credit: Rami Al-zayat/Unsplash
Image credit: Rami Al-zayat/Unsplash
  • “Underscoring the addictive nature of smartphones, I experienced symptoms of withdrawal when I first cut the metaphorical cord, staring at my iPod throughout the day and hoping for a rush of dopamine that no longer came. With time, however, I began to make practical use of my quiet time.” In Nature’s Career section, Adam Weiss shares the benefits he discovered when he disconnected from his phone to concentrate on science.
  • “Ellis and Stevens’s research offers a historical overview of bounty programs and an analysis of contemporary bug bounty platforms​​—the new intermediaries that now structure the vast majority of bounty work. The report draws directly from interviews with hackers, who recount that bounty programs seem willing to integrate a diverse workforce in their practices, but only on terms that deny them the job security and access enjoyed by core security workforces.” A report from Data and Society published in January of this year examines current approaches used to find programming bugs and security flaws through bounty programs, and proposes some ways that the system could be improved.
  • “These three patients, and more than 350 other blind people around the world with Second Sight’s implants in their eyes, find themselves in a world in which the technology that transformed their lives is just another obsolete gadget. One technical hiccup, one broken wire, and they lose their artificial vision, possibly forever.” An astonishing and thought-provoking story in IEEE Spectrum by Eliza Strickland and Mark Harris unfolds the story of what happens when a company that makes a near-miraculous medical technology bails on product with its own future in doubt.
  • “Here, we investigate to what extent different epistemic authorities affect the perceived value of nonsensical information. To this end, we contrasted judgements of gobbledegook spoken by a spiritual leader with gobbledegook spoken by a scientist. In addition, we assessed whether the source effect is predicted by individual religiosity and varies cross-culturally, as a proxy for how scientists and spiritual authorities function as ‘gurus’ for different individuals and within different cultural contexts.” An article published in Nature Human Behavior by Hoogeveen and colleagues describes results from a study that evaluated how people across the world weigh the relative credibility of statements attributed to scientific vs religious or spiritual figures.
  • “The lawsuit alleges that Facebook scanned user photos without asking for permission and downplayed the fact that its automated tagging system amounted to mass biometric data collection. It also echoes an earlier lawsuit’s claim that Instagram scans images for facial recognition, allegedly contradicting statements in its user agreement.” The Verge’s Adi Robertson reports that the state of Texas is suing Meta (formerly Facebook) over its use of personal data in developing facial recognition technology.
  • “Alongside the support of family and friends, scientists rely on an army of technicians, librarians and other people who fill roles that contribute to research outputs — from the humble article to paradigm-shifting experiments….These essential, hidden roles are rarely celebrated alongside research achievements, which can make it hard to convince graduates in science, technology, engineering and mathematics to consider them as career options.” An article in Nature by Gemma Derrick and Simon Hettrick shines a spotlight on all-too-often overlooked contributors to science.


Photograph of Dr. Robert M. Califf speaking at scientific conference. Image credit: Nathan McKeel/Duke University
Dr. Robert M. Califf speaking at a 2018 scientific conference. Image credit: Nathan McKeel/Duke University
  • “Dr. Califf, who is 70, is expected to be sworn in this week. He faces a looming flurry of decisions — including intense scrutiny of a coronavirus vaccine for children under 5 and reviews of e-cigarette applications like Juul’s bid to stay on the market.” The New York Times’ Christina Jewett and Emily Cohcrane report on Tuesday’s Senate vote to confirm longtime Duke cardiologist and clinical trialist Rob Califf to a second tenure as FDA Commissioner.
  • “Some may wonder why all of this is needed, since the digital Hippocratic oath is essentially a form of consumer protection for data. The problem is the way some industry actors treat patient data today — and the way they’re rewarded for using data. There have been too many high-profile examples of third-party vendors using health data in ways that patients didn’t expect when they entered information into an app.” In an opinion article at STAT News, Ries Robinson and Aneesh Chopra propose a Hippocratic Oath for medicine’s emerging digital age.
  • “The overarching problem is that the algorithms used by the social media companies actively discourage authoritative information—disagreement and outlandish statements result in more engagement….Most scientists don’t appreciate how public resistance to facts can be amplified by inanimate algorithms, rather than by living conspiracy theorists, that reinforce what people choose to engage with. Science editor Holden Thorpe weighs the relative merits of scientists participating in social media, and finds a mixed bag of outcomes – although one amenable to improvement.
  • “…while the Sunshine Act has clearly helped expose important commercial influences on both prescribing and the scale of industry involvement with physicians, it has also, paradoxically, fuelled further commercial surveillance and marketing. The article casts new light on innovative pharmaceutical marketing approaches and the key role of data brokers and analytics companies in the identification, targeting, managing, and surveillance of physicians.” An article by Mulinari and Ozieranski published in Big Data and Society reports paradoxical effects from the Sunshine Act, a law that mandated transparent reporting of financial relationships between physicians and the medical products industry.
  • The February issue of Health Affairs is devoted to different facets of racial equity in healthcare, including articles examining racial bias in electronic health record notes and bias in machine learning in healthcare settings. (Notably, shortly after the theme issue was released, Health Affairs’ digital strategist Patti Sweet noted that attempts to promote the theme issue on Health Affairs’ social media channels were being blocked by bots that flagged their messages as potentially objectionable: “…now we’re just angry. The bots or the algorithms—whatever you want to call them—are blocking our content. Our ads have been scanned by a robot; the robot sees the word ‘racism’ and disapproves the material.”)
  • “Telehealth proponents expected the pandemic to net them a windfall of convincing evidence that virtual care could increase quality and cut spending. But two years after health systems went virtual almost overnight, industry watchers are still disputing a key aspect that could determine telehealth’s fate: whether the option for virtual visits means patients will see doctors more often than they would in-person.” A STAT News feature (log-in required) by Mohana Ravindranath examines ongoing efforts to get a handle on the actual impact of the “pivot” to telehealth services that was thrown into high gear during the COVID pandemic.