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.
August 26, 2022
In today’s Duke AI Health Friday Roundup: the limits of language in AI; a worsening child mental health crisis in North Carolina; big open-access policy change from OSTP; using machine learning to predict carcinogenic compounds; choosing whether to attempt to eradicate diseases; new method may offer cheap path to unravelling “forever” PFAS chemicals; reporting of NIH-funded clinical trials still lags; lip service vs meaningful action in publication integrity; equity, justice, and disability data; much more:
AI, STATISTICS & DATA SCIENCE
- “…as these [large language models] become more common and powerful, there seems to be less and less agreement over how we should understand them. These systems have bested many ‘common sense’ linguistic reasoning benchmarks over the years, many which promised to be conquerable only by a machine that ‘is thinking in the full-bodied sense we usually reserve for people.’ Yet these systems rarely seem to have the common sense promised when they defeat the test and are usually still prone to blatant nonsense, non sequiturs and dangerous advice.” An essay in Noēma by Jacob Browning and Yann Lecun argues that recent enthusiasm (if not hype) about the capabilities of recent large language models may reflect a fundamental misapprehension about the nature and complexities of true human-like intelligence.
- “Accurate prediction of carcinogens is an ever-challenging task owing to the limited information about bona fide (non-)carcinogens. We developed Metabokiller, an ensemble classifier that accurately recognizes carcinogens by quantitatively assessing their electrophilicity, their potential to induce proliferation, oxidative stress, genomic instability, epigenome alterations, and anti-apoptotic response.” A research article by Mittal and colleagues, published in Nature Chemical Biology, describes “Metabokiller,” an AI classifier designed to predict potentially carcinogenic human metabolites.
- “The genomics world has no shortage of visualization tools. But as new methods and data types emerge, existing techniques can struggle to cope. Now, a tool known as Gosling allows bioinformaticians to build apps that can display genomic information with the same level of flexibility that developers have come to expect from other graphics programming tools.” A feature article by Nature’s Jeffrey M. Perkel describes a visual “grammar” for visualizing complex suites of genomic data.
- “It’s well known that ML models are supremely adept at recognizing patterns. Often, though, the patterns that they learn can incorporate features that the systems’ authors never intended. We’ve laughed—and despaired—over the systems that diagnosed pneumonia based on a radiographic marker…or detected pneumothorax based on confounding image features, such as an inserted chest tube…” An editorial in Radiology: Artificial Intelligence by Charles E. Kahn, Jr. introduces a set of articles addressing strategies for reducing or countering bias in AI systems in radiology.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Methods to dispose of PFASs typically rely on expensive and harsh treatments, some of which require high pressures and temperatures above 1,000°C. What’s more, there’s evidence that incinerating products containing PFASs can lead to the spread of these compounds into the environment, says Brittany Trang, an environmental chemist at Northwestern University in Evanston, Illinois, who co-led the study describing the new approach…The latest method…showed promise in breaking down one of the largest groups of PFASs using inexpensive reagents and temperatures of about 100°C.” At Scientific American, Giorgia Guglielmi reports on recent research that may offer a practicable approach to breaking down PFAS compounds, often called “forever chemicals,” whose extreme persistence in the environment has heightened concerns in the wake of revelations of potential health threats posed by the chemicals.
- “Hospital officials across the state say there are children in mental health distress living in their emergency departments. Atrium Health has seen a 65% increase in emergency department patients needing psychiatric care, according to leaders at the Charlotte-based hospital group. For children in need of psychiatric care, the demand tripled over the course of the pandemic.” An article by North Carolina Health News’ Taylor Knopf (co-published by WRAL) exposes the severity of NC’s current child and adolescent mental health crisis.
- “For years, roving clinics like the Family Van have hit roads across the country to reach millions of people who might not otherwise have access to medical care….The work of mobile clinics grew all the more important during the Covid-19 pandemic, when health care saw widespread disruptions. The Family Van still operated during this slowdown by calling people to check their wellbeing, delivering masks, and handing out gift cards.” STAT News’ Edward Chen profiles the work of mobile clinics that have been providing vital health care for people in communities and regions that would otherwise have significant difficulty in accessing care and services.
- “Decisions about whether and how to wipe out an infectious disease can be controversial, weighed up on fine lines in terminology and risk–benefit calculations. A basic distinction between elimination and eradication often gets lost, which causes confusion. Most experts agree that to eradicate means to make sure no cases of an infection are seen anywhere in the world, a goal that hinges on interrupting transmission permanently; to eliminate is to do this in a specific geographic area, usually a country. A feature in Nature by Anita Makri surveys the debate within the global health community over whether to attempt the complete eradication (as opposed to management) of certain disease.
COMMUNICATION, Health Equity & Policy
- “No one questions the critical importance of a reliable biomedical literature… Yet the cases described above are but a few of the many examples of the slow, opaque, inconsistent, frustrating, and unsatisfactory outcomes of tumbling into the rabbit hole of publication integrity. Watching paint dry is ultimately more fulfilling: at least the paint will be dry eventually. Why is achieving and maintaining publication integrity so fraught? Could it be that the main protagonists don’t actually care?” A guest post at Scholarly Kitchen by Andrew Grey, Alison Avenell, and Mark Bolland pointedly questions why addressing issues of integrity in scientific publication is such a slow, cumbersome, and frustrating process.
- Some big news from the White House’s Office of Science and Technology Policy this week with major implications for access to publications and data produced with help from federal funding: “In a memorandum to federal departments and agencies, Dr. Alondra Nelson, the head of OSTP, delivered guidance for agencies to update their public access policies as soon as possible to make publications and research funded by taxpayers publicly accessible, without an embargo or cost.”
- “The U.S. National Institutes of Health failed to ensure that results of roughly half of clinical trials funded by the agency — both those conducted by its own scientists and outside researchers — were reported to a federal database during a recent two-year period, a new government review has found.” An article at Pharmalot (log-in required) by Ed Silverman reports that National Institutes of Health, despite years of attention on the issue, is still falling short on standards for reporting on the clinical research the agency funds and conducts.
- “The COVID-19 pandemic has forced us to take a hard look at public health data systems. Efforts are underway to reimagine public health infrastructures to support equity and justice. But too often people with disabilities are not included on teams leading this restructuring and remain at risk of being de-prioritized in public health data systems….Public health data systems and infrastructure must be built to collect disability data and use this information to combat ableism and support equity and social justice.” In a perspective for Health Affairs Forefront, Bonnielin Swenor presses the case for the routine and comprehensive collection of public health data on disabilities.
- “After measuring the rural/urban differences in EHR use and hospital costs, the study aims to answer the question of whether EHRs are helping both rural and urban hospitals reduce costs. For urban hospitals, the answer is yes. EHR use helps urban hospitals reduce costs, particularly ancillary and outpatient costs. For rural hospitals, however, the results are not so promising, as EHR use is never significantly associated with any changes in costs.” A study conducted by North Carolina Health News’ Claudia Rhoades, Brian Whitacre and Alison Davis shows that rural hospitals in NC have not been sharing in the cost-saving benefits of electronic health records that have been enjoyed by larger health systems in more populous regions.