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.
November 11, 2022
In today’s Duke AI Health Friday Roundup: Meta language model predicts proteins; Paxlovid and long COVID risks; innate immune system offers possibilities for Alzheimer’s approaches; lawsuit sends ripples through the world of generative AI; harnessing value-based payment for health equity; teasing out the implications of OSTP publications access policy; pitfalls of using demographic data to promote algorithmic fairness; much more:
AI, STATISTICS & DATA SCIENCE
- “Normally, language models are trained on large volumes of text. To apply them to proteins, Rives and his colleagues instead fed the AI sequences of known proteins, which can be written down as a series of letters, each representing one of 20 possible amino acids. The network then learnt to fill in the sequences of proteins in which some of the amino acids were obscured.” Nature’s Ewen Callaway reports on a recent preprint by Meta researchers that describes the use of a large language model to predict the shapes of hundreds of millions of proteins. Over at his blog In the Pipeline, Derek Lowe has some additional analysis.
- “…demographic-based algorithmic fairness techniques look to overcome discrimination and social inequality with novel metrics operationalizing notions of fairness and by collecting the requisite data, often removing broader questions of governance and politics from the equation…we argue that collecting more data in support of fairness is not always the answer and can actually exacerbate or introduce harm for marginalized individuals and groups…” A research paper authored by researchers McKane Andrus and Sarah Villeneuve at the Montreal AI Ethics Institute explores paradoxical dangers in collecting and using demographic data in the interest of algorithmic fairness.
- “Microsoft, its subsidiary GitHub, and its business partner OpenAI have been targeted in a proposed class action lawsuit alleging that the companies’ creation of AI-powered coding assistant GitHub Copilot relies on ‘software piracy on an unprecedented scale.’ The case is only in its earliest stages but could have a huge effect on the broader world of AI, where companies are making fortunes training software on copyright-protected data.” The Verge’s James Vincent reports on a lawsuit that accuses Microsoft, GitHub, and OpenAI of breaching copyright in creating an AI-assisted computer code generator. The lawsuit could have significant implications for the larger world of generative AI, including now-ubiquitous image generators trained on scraped images.
- “The modus operandi of a clinical diagnosis may fail to consider the relative weighting of these disparate data inputs and potentially non-linear relationships highlighting the limitations of human decision-making capacity. The strength of algorithmic decision-making support is that it can be used to offload such tasks, ideally yielding a more successful result. This is the promise of precision medicine.” A scoping review by Kline and colleagues published in NPJ Digital Medicine examines the potential for machine learning approaches to incorporate multimodal data in clinical decision support applications.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “One physician’s stubborn pursuit for meaning with one foot out the door doesn’t make a neat sound bite. But the reality of medicine is messy. It’s a house constructed from seeming contradictions. These experiences don’t lend themselves to clean narrative arcs or research outcomes. I believe what makes them challenging to figure out or measure is what makes them worthy of our most vigilant attention.” In an opinion article for STAT News, Jay Baruch describes how his burnout as an emergency medicine physician took an unexpected turn.
- “In sum, our results show that in people with SARS-CoV-2 infection who had at least 1 risk factor for progression to severe COVID-19 illness, treatment with nirmatrelvir within 5 days of a positive SARS-CoV-2 test was associated with reduced risk of PASC regardless of vaccination status and history of prior infection. The totality of findings suggests that treatment with nirmatrelvir during the acute phase of COVID-19 reduces the risk of post-acute adverse health outcomes.” A preprint by Xie and colleagues, available at medRxiv, reports on associations between taking the antiviral medication nirmatrelvir (Paxlovid) and the subsequent risk of developing post-acute sequelae of SARS-CoV-2 (PASC) (i.e., “Long Covid” symptoms).
- “Despite substantial advances that have been made in our understanding of AD [Alzheimer disease] pathophysiology and the development of potential therapeutic interventions, particularly targeting Aβ [amyloid beta], the fundamental mechanisms underlying the clinically symptomatic phase of AD are still elusive. A better understanding of the specific roles of both the innate and adaptive immune system throughout the course of AD pathology, especially during the tau phase of AD, will likely lead to novel therapeutic intervention strategies for both the pre-clinical as well as clinically symptomatic phase of AD.” A review article by Chen and Holtzman, published in Immunity, examines the role of innate immunity in the development and progression of Alzheimer disease, and scrutinizes the potential for future therapeutic approaches.
- “With overdose deaths hovering around an all-time high, many addiction treatment advocates have shown a willingness to try aggressive new solutions. Methadone, in particular, has emerged as a flashpoint in the broader debate about how quickly the U.S. should expand access to addiction treatment and harm reduction services.” At STAT News, Lev Facher reports on advocacy for – and opposition to – easing access to the opioid-addiction treatment mainstay methadone.
COMMUNICATION, Health Equity & Policy
- “The effects of identity cues were significant and heterogeneous, accounting for between 28% and 61% of the variation in voting associated with commenters’ production, reputation and reciprocity. Our results also showed that identity cues cause people to vote on content faster (consistent with heuristic processing) and to vote according to content producers’ reputations, production history and reciprocal votes with content viewers.” A study by Taylor and colleagues, published this week in Nature Human Behavior, reports findings from a randomized study that found that people react differently to social media content based on whether they can discern the identity of the content’s author(s).
- “For years, families of people with disabilities have filed lawsuits and complained that North Carolina has dragged its feet in providing support services in community settings. There is a state Medicaid program…that allows people with disabilities to receive services at home or in the community setting of their choice. However, an ever-growing waitlist has more than 16,000 people who, under current circumstances, will wait up to a decade or more for a slot to receive services….Baddour’s ruling gives the state a 10-year timeline for serving all of the potential recipients and eliminating the massive waitlist.” At NC Health News, Taylor Knopf and Rose Hoban report on a recent Superior Court ruling that requires the state of North Carolina to provide funding for home care services for people with intellectual and developmental disabilities.
- “…we present here a brief framework for categorizing the spectrum of approaches we identified for improving equity through VBP [value-based payment] model implementation. And while critical to advancing health equity, these new equity-focused models are novel, and we need more evidence on what works, especially in early years of implementation. Accordingly, we end this article with a brief overview of where equity-focused evaluation is needed to assess the success of equity-focused design and implementation.” An issue brief by Bleser and colleagues, published in Health Affairs Forefront, lays out a schema for using value-based payment policies to support the goal of improving health equity.
- “…to understand where Nelson Memo implementation is going as far as access to research papers, it is essential to understand that this is not a publishing …What it does is to regulate the conditions required for a researcher to receive federal research funding…. It will set requirements on funded researchers, not pick a business model that publishers must follow.” At the Scholarly Kitchen, David Crotty muses on the likely effects of recent policy guidance put forward by the White House Office of Science and Technology Policy that will require expanded public access to publications reporting the results of federally funded research.