Forge Friday Roundup - October 4, 2019

October 4, 2019

In today’s Roundup: unpacking algorithms behind YouTube recommendations; PAWS for machine learning; why data standards matter; crank science infects concussion research; misinformation complicates polio-eradication efforts; self-driving cars hit NIMBY roadblock in Silicon Valley; causal Bayesian networks for fairer AI; herd immunity, visualized; controversy erupts over recent “GM mosquito” paper; stumping for a kinder approach to academic research; much more:

AI, Statistics & Data Science
  • “Attackers share in the power of machine learning and also its vulnerabilities. We can turn this vulnerability into a weapon to defend our privacy." Wired’s Andy Greenberg has some rare good news for privacy in the form of machine learning adversarial examples – an “inherent bug” or blind spot where a few modest alterations to data or images can trick the machine learning system into drawing false conclusions such as “mistak[ing] a rifle for a helicopter” (H/T @DukeResearch).
  • In a new post for the Google AI Blog, Google’s Yuan Zhang and Yinfei Yang introduce two new datasets sets for the larger research community – Paraphrase Adversaries from Word Scrambling (PAWS) in English, and PAWS-X, an extension of the PAWS dataset to six different languages – to equip machine learning models to identify paraphrase and non-paraphrase pairs – sentences or phrases that deliver the same meaning using different wording, or different meaning using the same wording.
A dog's paw on a human hand
Image via Pixabay
  • In a post for the biostatistical blog DataMethods, statistician Andrew Althouse of the University of Pittsburgh breaks down a recently published trial (featured in last Wednesday’s Roundup) assessing the impact of Vitamin C for patients with severe sepsis. He dives deep into some of the intricacies of interpreting findings from the trial, which found no difference in the treatment arms in the predetermined primary endpoints, yet did suggest fewer patients receiving Vitamin C died compared with patients on the control arm.
  • A blog post by DeepMind’s Silvia Chiappa and William Isaac provides a digestible overview of a pair of recent papers exploring the benefits of causal Bayesian networks in creating fairer machine learning systems.
  • “Yet even with all of this data entry going on, it is a struggle for patients to get copies of their records, and an even bigger one to get them in useful, digital formats. Even more alarming, despite the vast amount of data collected by electronic medical record systems, little of it is used to help clinicians make decisions about their patients’ care. That’s unacceptable.” A STAT News editorial by Kenneth Mandl and Isaac Kohane offers a spirited justification for paying close attention to the otherwise “wonky” world of data standards.

Basic Science, Clinical Research & Public Health
  • “This is intriguing and exciting research. But it’s probably too soon to add broad spectrum antifungals, many of which have lots of side effects, to cancer treatment regimens, even in experimental settings.” New York Times’ Knvul Sheikh quotes Stanford’s Ami Bhatt while reporting on a new Nature study by NYU researchers whose work illuminates ways that fungal infections in the pancreas may drive the growth of tumors.
  • “When enough of the population has been vaccinated, a disease is no longer able to spread effectively, protecting even the unvaccinated.” In a recent tweetorial, Berkeley Earth lead scientist Robert Rohde shares a simple yet alarming animation he created that explains how herd immunity and works.
  • “When your head gets yanked around, your brain does too, and it’s nearly impossible to decouple the two. You can’t put a seat belt around the brain.” Wired’s Christie Aschwanden surveys on a plethora of scientific research on helmets and head-protective technology including a 2019 Lancet study that underscores the degree to which pseudoscience has infiltrated the response to the “concussion crisis” in contact sports in the United States. 
    A football helmet on a football field
    Image credit: Andzelika Tokarska via Pixabay
  • “Obviously there’s an interest in monetizing biomarkers collected from diverse populations and underrepresented populations, [but] without adequate protections, the concern becomes about exploitation.” Undark’s Adrian Pecotic examines the need for greater diversity in genetic research and the ethical challenges of consent and possible interruption of indigenous populations that can complicate the achieving of it.  
  • “The effects of e-cigarettes have similarities to and important differences from those of cigarettes. [Yet] studies show measurable adverse biologic effects [of e-cigarettes] on organ and cellular health in humans, in animals, and in vitro.” In a State of the Art Review for BMJ, Duke’s Sven-Eric Jordt and colleagues investigate the health effects of vaping and e-cigarettes and find present knowledge to be insufficient in dubbing them any less harmful than tobacco products.
  • “Persistent virus transmission is happening across [Pakistan], even in areas that were previously not reporting circulation of the virus,” in most parts because of a fake video that went viral in the country showing a man’s loud uproar claiming that the polio vaccine was harmful for children, “with the camera cutting to a group of young boys who acted to faint on cue.” Tooba Masood reporting for Pakistani English-language daily Dawn, details the challenges facing government officials in trying to curb misinformation about polio as well as virus as cases mount across the country. Pakistan, Afghanistan and Nigeria are the only remaining countries in the world where polio is still categorized as an endemic viral infection.
  • “…in line with what has been said before, Farrar concedes that a focus on excellence also contributes to “destructive hyper-competition, toxic power dynamics and poor leadership behaviour” — the latter echoing the findings from Nature’s comprehensive survey on research-group culture last year.” In an editorial published in Nature, the journal praises Wellcome’s recent emphasis on creating a kinder, less cutthroat research culture, but notes that the entire ecosystem will need to be brought on board.
  • Enhancing innate immunity has become a key mechanism for fighting cancer, from immunotherapy to CAR-T cells. Yet, only a minority of patients are eligible for benefit from these therapies. In a review for Nature, researchers details some opportunities to expand the pool of patients who could be helped by these therapies (H/T @EricTopol). 
  • An intense scientific squabble has broken out after the publication of a recent paper suggesting the introduction of genetically modified mosquitoes, intended to reduce the population of Zika-transmitting mosquitoes, resulted in the creation of hybrid mosquitoes, which led to “a more robust population than the pre-release population.” Five co-authors, some of them reporting they didn’t see the final published version (!), object to at least some aspects of the paper’s interpretation and conclusions and are asking the editors to retract the article. Kelly Servick reports on the controversy for Science Magazine.
  • “Bravery isn’t universally successful in any venture—scientific or architectural—but buildings that survived to the modern era were remade to fix their flaws, like a repeated experiment that benefits from prior failures.” At Citylab, Matthew Francis shares an excerpt from the upcoming Midwest Architecture Journeys, a study of some of the state’s most architecturally inventive buildings, with the Silicon Detector (SiDet) Building under the spotlight. The covert structure behind the security perimeter of the Fermi National Accelerator Laboratory (Fermilab), the center of Cold War physics research, is said to reflect the vision of the lab’s former director, physicist Robert R. Wilson.

Communications & Digital Society
  • “Silicon Valley types can be most skeptical of advanced technology because they know how it works and what its risks are. Parents with experience at large tech firms have famously cracked down on screen time for their children. Some tech executives will not let female family members ride alone at night with ride-sharing cars. Others keep their kids off social media indefinitely.” Washington Post’s Faiz Siddiqui reports on Silicon Valley residents who are wary of the self-driving revolution and the prospect of its early iterations being tested in their neighborhood (H/T @missy_cummings).
  • “YouTube should spend more energy in understanding which actors their algorithms favors and amplifies than how to keep users on the platform.” MIT Technology Review’s Karen Hao quotes Berkman Klein Center for Internet & Society’s Jonas Kaiser while reporting on a new paper by Google researchers that contradicts YouTube’s public claims of addressing the many controversies surrounding its recommendation engine and the algorithms that fuel it. 
  • A research study led by Baylor’s Bryan Shaw published in Science Advances introduces and reviews the free CRADLE app – a computer-assisted detector of leukocoria, an abnormal white reflection from the retina of the eye, in casual photographs of children. Leukocoria is a key warning sign of retinoblastoma and multiple other eye pathologies.
  • “Secondary endpoints should not be redacted like the Mueller Report.” In a recent tweetorial, oncologist Vinay Prasad expresses takes issue with the various blacked-out redactions – including secondary endpoints and eligibility criteria - in New England Journal of Medicine clinical trial protocols included with published reports of results from clinical trials.
  • “Burnout is like a cake with frosting. The frosting is the yoga classes, therapy dogs, free meals, baseball tickets: the stuff that lets people know that we value them. But making a more robust cake so we don’t need a lot of frosting on it. That would be a better strategy to make people enthusiastic for long careers in medicine.” In an article for Duke School of Medicine’s Magnify Magazine, medical student Celia Reynolds quotes Duke’s Catherine Kuhn amongst others in making a case against “burnout shaming” in medicine while sharing stories from the DukeMed Monologues storytelling event, an annual show about identity written and performed by members of the Duke community.  

  • STAT News’ Ed Silverman reports on the FDA’s questioning of testing methods used by Valisure Laboratory in results that found the carcinogen NDMA in the heartburn medication ranitidine. The FDA’s statement comes three weeks after the agency, together with the European Medicines Agency, announced their investigation into traces of the carcinogen found in some heartburn drugs following a petition filed by Valisure, stirring widespread concern.
  • “The recordings show that the transplant team was fixated on keeping him alive, rather than his quality of life or his family’s wishes, because of worries about the transplant program’s survival rate, the proportion of people undergoing transplants who are still alive a year after their operations.” A disturbing report from ProPublica’s Caroline Chen shows a family being deliberately left in the dark as a patient was kept on life support over concern about the hospital’s transplant program performance statistics.
  • In a First Opinion for STAT, Duke pediatricians Michael Cohen-Wolkowiez and Kanecia Zimmerman highlight the work of pediatric teams in contributing to the U.S. FDA research guidance for industry on clinical pharmacological studies in neonates, which they describe as “a big step in the right direction” in eliminating “guesswork” from drug dosing in newborns (H/T @Moorejh).
  • “It's probably not a good time to own a therapy company now. But if you do, you need to adapt and change the business model tremendously.” At Modern Healthcare, Alex Kacik reports on the Centers for Medicare and Medicaid Services’ (CMS) new patient-driven payment model for skilled-nursing facilities implemented earlier this week and thousands of physical, occupational and speech therapists losing their jobs or taking a substantial pay cut in its wake.
  • At the Washington Post, Ben Guarino reports how the U.S. Department of Agriculture (USDA)’s move from Washington, DC to Kansas has left tens of millions of dollars in approved study and research grants in limbo and key completed research reports - including those on veterans’ nutritional needs, the opioid epidemic, obesity trends and more – unpublished (H/T @bylenasun).
  • “Even with the best reimbursement reforms, these market entry rewards will be needed to restore the antibacterial R&D ecosystem to health without driving inappropriate overuse of novel antibiotics.” In a detailed letter published in Nature Biotechnology, Kevin Outterson teases out the complex ramifications of new CMS incentive structures for the development and use of much-needed new antibiotic therapies.