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.

July 1, 2022

In today’s Duke AI Health Friday Roundup: embodied AI reaches toward a new kind of problem-solving; considering the role of pragmatic and virtual clinical trials; Lancet surges to the top of journal impact factor ratings; mistrust of tech and the collapse of contact tracing efforts; FDA orders Juul to cease marketing vaping products; managing polypharmacy in heart failure; the ethics of large language models; the threat to privacy posed by inferential analytics; making inroads on childhood food insecurity; much more:

AI, STATISTICS & DATA SCIENCE

A girl wearing a backpack places her hand into the outstretched hand of a humanoid robot bedecked with flowers. Image credit: Andy Kelly/Unsplash
Image credit: Andy Kelly/Unsplash
  • “It’s the difference between observing a possible relationship between two objects and being the one to experiment and cause the relationship to happen yourself. Armed with this new understanding, the thinking goes, greater intelligence will follow. And with a suite of new virtual worlds up and running, embodied AI agents have already begun to deliver on this potential, making significant progress in their new environments.” At Quanta, Allison Whiten explores the strange new world of “embodied AI” – an approach to training AIs through interactions with the physical world that some researchers hypothesize may be a key step in true machine intelligence.
  • “A key reason for this failure is that people don’t trust the tech companies or the government to collect, use, and store their personal data, especially when that data involves their health and precise whereabouts. Although Apple and Google pledged to build privacy measures into the apps’ design—including opt-in choice, anonymity, use limitations, and storage of data only on a user’s device—Americans just weren’t persuaded.” The early days of the COVID pandemic saw a host of efforts aimed at contact tracing to better understand and prevent the spread of the disease, but despite the resources and technological ingenuity unleashed on these efforts, most posted disappointing results. A report by Brookings fellow Jessica Rich looks at why.
  • “When applied in real-life scenarios, the rule is typically applied as a first step in a longer process intended to understand why disparate impact has occurred and how to fix it. However, oftentimes engineers use fairness tools at the end of a development process, as a last box to check before a product or machine-learning model is shipped.” In an article for Protocol, Kate Kaye examines some of the limitation of the “80% rule” – a commonly used (but limited) tool for addressing issues of bias and fairness in AI packages.
  • “A new category of AI models, Large Language Models (LLMs) is rapidly gaining traction in systems ranging from internet search to automatic translation to online discourse moderation….However…the development of this technology happens primarily in large private labs that seldom share complete details of the specifics. This status quo excludes most of the direct and indirect stakeholders whose lives will be affected by these new systems, putting regulators and society in a position where they always have to respond to harms after they have already been created in real-world situations.” The Montreal AI Ethics Institute has compiled a series of recent articles on ethical approaches to constructing a type of artificial intelligence application known as a large language model (LLM).

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

Closeup photograph of a pile of different multicolored pills and capsules. Image credit: Myriam Zilles/Unsplash
Image credit: Myriam Zilles/Unsplash
  • “A shift in paradigm is therefore needed when evaluating polypharmacy in patients with HF. Instead of assuming all polypharmacy is “good” or “bad,” we propose a concerted move, using a multidisciplinary approach, to focus on the “appropriateness” of specific medications, in order to optimize HF medical therapy.” A review paper just published in Comorbidities by Sukumar and colleagues articulates an approach to managing polypharmacy – the prescription of multiple, and potentially interacting medications – in patients with heart failure.
  • “This body of work offers pragmatic approaches to overcoming the many reasons clinical trials are underperforming, such as high financial costs, long duration, administrative complexity, low recruitment and retention of participants, and difficulty translating knowledge to actual medical care.” A special theme issue in Contemporary Clinical Trials, with authors convened by the NIH Pragmatic Trials Collaboratory, has been devoted to the topic of pragmatic and virtual trials.
  • “…a recent study by a team of Stanford University researchers suggests that protein aggregation may be a universal phenomenon in aging cells and could be involved in many more diseases of aging than was suspected. Their discovery points to a new way of thinking about what goes wrong in cells as they age and, potentially, to new ways of staving off some consequences of the aging process.” An article by Quanta contributor Viviane Callier explores recent research that finds that proteins previously associated with neurodegenerative diseases such as Alzheimer’s may be a more ubiquitous concomitant of aging – one not just limited to the brain.
  • “As the pandemic has revealed, even powerful biomedical tools such as vaccines sputter in practice if disadvantaged people can’t access them, or if distrusting people refuse to use them. America’s recurring mistake is to create such technofixes at warp speed, while neglecting the systems that actually deploy those tools. Those systems—the country’s social infrastructure—are so porous that a multitude of smaller projects are necessary to patch each and every hole. Once lost, trust is hard to regain at scale. But it can slowly be rebuilt.” The Atlantic’s Ed Yong takes stock of the COVID pandemic and what has – and hasn’t – been done as the disease continues to waft through US society.
  • “After reviewing the company’s premarket tobacco product applications (PMTAs), the FDA determined that the applications lacked sufficient evidence regarding the toxicological profile of the products to demonstrate that marketing of the products would be appropriate for the protection of the public health.” The US Food and Drug Administration has issued orders for the makers of the Juul electronic cigarette to cease selling the popular vaping device, along with several types of “pods” used to provide flavored nicotine vapor.

COMMUNICATION, Health Equity & Policy

Husky dog in sled harness, looking directly at camera with tongue lolling out and a sardonic expression on its face. Image credit: Jérémy Stenuit/Unsplash
Image credit: Jérémy Stenuit/Unsplash
  • The latest report on journal impact factors is out from Clarivate analytics, and there’s been a shakeup at the top as The Lancet surges to the front of the pack, pushing aside the New England Journal of Medicine, the previous (and perennial) top of the listing. Context for this development, plus some more highlights from the Clarivate report, are at Retraction Watch.
  • “…health care systems screening for food insecurity and organizations best positioned to provide resources addressing food insecurity often operate in siloes, with very limited collaboration across these sectors. Additionally, the capacity of CBOs to respond to food insecurity—in the context of increased referrals from clinical screening, economic hardships caused by the COVID-19 pandemic, and soaring inflation—is insufficient or tenuous at best.” An article in Health Affairs by Duke’s Lilianna Suarez and Rushina Cholera looks at how inroads into addressing child hunger that were made during the initial response to COVID school closures can be sustained in the future.
  • “Privacy risks of the digital age do not stem solely from how personal data is collected, observed, or intentionally shared. Rather, the ability to draw intimate inferences about people from their data arguably poses the greatest risk. These inferences determine how they are viewed, evaluated, and ultimately treated by third parties.” A commentary in the American Journal of Bioethics by Brent Mittelstadt examines how simplistic privacy protections can break down in the face of “inferential analytics.”
  • “Eileen O’Grady, research manager at the nonprofit Private Equity Stakeholder Project, said private equity’s focus on strong, speedy returns makes it a risky business model for health care. ‘In rural hospitals,’ O’Grady said, ‘there are very few ways’ to boost revenue and cut expenses ‘without having an impact on patient care.’” An article by Kaiser Health News’ Sarah Jane Tribble examines the fallout after rural hospitals are bought by venture capital firms.