Forge Friday Roundup - May 25, 2018.

May 25, 2018

Welcome to the first edition of our Friday roundup of news stories, peer-reviewed articles, and preprints we encountered this week and found interesting or thought-provoking. This week: data trapped in electronic health records, Forge Director Robert Califf wins Alvarez prize for communication efforts, AI (and its discontents) in the patient care setting, the Robert Wood Johnson Foundation unveils the City Health data dashboard, and much more.

  • The New York Times’ Gina Kolata describes the challenges of unlocking potentially life-saving EHR data for cancer research.
  • “An idea has little value if it’s not communicated.” Forge Director Robert Califf receives Walter C. Alvarez prize for medical communication from the American Medical Writers Association.
  • The Robert Wood Johnson Foundation has released the City Health Dashboard, which provides data visualizations for a variety of key health outcomes and socioeconomic and environmental factors across the 500 most populous US cities.
  • The FDA has begun a pilot program for precertification of FDA-regulated software as a medical device (SAMD).
  • Stanford Medicine holds two-day conference on Big Data in Precision Health.
  • “Few subjects have afforded more room for doubt, or caused more harm through false certainty, than heredity.” Nick Lane reviews Carl Zimmer’s She Has Her Mother’s Laugh: The Power, Perversions, and Potential of Heredity for Nature.
  • Can new technologies, including EHRs, wearable sensors, and machine learning enable the personalized medicine paradigm in psychiatry?
  • The New Yorker’s Allison Pugh looks at some implications of the surprising findings that emerged from a pilot study of an AI-powered “virtual nurse.”
  • "In some sense this is like the age-old nature-nurture debate, now translated into engineering terms." Science takes a look at efforts to train AIs as if they were children in order to help them “learn the world.”
  • The Wall Street Journal’s Christopher Mims examines the intersection of AI, consumer electronics, and healthcare (subscription required).
  • “All the impressive achievements of deep learning amount to just curve fitting.” A Quanta interview with Judea Pearl touches on the AI expert’s growing skepticism of machine learning.
  • A pair of complementary posts by Frank Harrell and Drew Levy at Statistical Thinking offer a “roadmap” for choosing between statistical modeling and machine learning.
  • The joint NCI/CDC/ACS/NAACCR Annual Report to the Nation on the Status of Cancer contains good news on declining cancer mortality, but this good news is tempered by evidence of continuing racial disparities.
  • A paper by Bumgarner et al. in the Journal of the American College of Cardiology (subscription required) reports on the use of a smartwatch algorithm to detect atrial fibrillation.
  • New in JAMA: Change is coming for privacy regulation, but how far it will go and who it will cover is still uncertain. I. Glenn Cohen and Michelle Mello discuss HIPAA and the implications of the MyHealthEData initiative.
  • A new viewpoint article in the New England Journal of Medicine explores the largely untapped potential of the “electronic” in electronic health records.
  • High school students’ machine learning project discriminates suicides among drug overdose deaths and wins Addiction Science Award from NIDA.
  • A preprint available from ArXiv describes a new approach for applying more computationally efficient neural networks to image analysis.
  • “Highly predictive, transparent, and easily understood by humans.” When bias is suspected in the formulation of inscrutable “black box” machine learning methods, means to confirm or deny that bias is typically lacking. CORELS, a new algorithm intended to provide an alternative to CART and other modeling methods, is described in a preprint now available on ArΧiv.
  • The use of digital healthcare in areas of crisis and humanitarian need also raises ethical implications for the developed world. A recent article in BMC Conflict and Health by Forge advisor Eric Perakslis looks at what we can do with digital health and how we should do it.
  • Neural networks used for image analysis require large volumes of training data, and expert annotation is expensive and time-consuming. A recent ArXiv preprint explores a possible solution to this problem in the form of cinematic rendering techniques for generating photorealistic images of tissue.