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 22, 2022

In today’s Duke AI Health Friday Roundup: effectiveness and uptake of machine-learning application for sepsis detection; associations between community violence and cardiovascular risk; data leakage as challenge for machine learning replication; differences in aging have implications for dementia risk; assessing the state of telehealth in NC; bolstering diversity in clinical trials; tiny windup motor runs on DNA; much more:


Closeup photograph of drops of water leaking from an outdoor spigot. Image credit: Luis Quintero/Pexels
Image credit: Luis Quintero/Pexels
  • “A taxonomy of data leakage can enable a better understanding of why leakage occurs in ML-based science and inform potential solutions. We present a fine-grained taxonomy of 8 types of leakage that range from textbook errors to open research problems. Our taxonomy is comprehensive and addresses data leakage arising during the data collection, pre-processing, modeling and evaluation steps.” A draft paper from an upcoming Princeton conference on reproducibility in machine learning identifies the phenomenon of “data leakage” as the culprit in widespread failures to replicate results in machine learning studies (H/T @rusincovitch).
  • “…these technologies have great potential to improve care and rapidly identify patients who are deteriorating, especially those with sepsis — which could help prevent deaths and reduce costs for large numbers of patients. Achieving these benefits widely across hospitals and health systems will take time and will require that implementers follow the key precepts of using AI to improve care, such as verifying that the model works at multiple sites and that AI-based decision support is communicated to clinicians in ways they find acceptable…” An editorial at Nature Medicine by Bates and Syrowatka highlights two recent publications by a research team that describe the results following the deployment of a machine-learning tool for detecting sepsis in patients (including both application effectiveness and adoption experience in the clinical setting).
  • “Our systematic review demonstrates and corroborates evidence that issues of construct validity are sorely neglected in original and replicated research. We identify four measurement challenges replicators are likely to face: a lack of essential measurement information, a lack of validity evidence, measurement differences, and translation. Next, we offer solutions for addressing these challenges that will improve measurement practices in original and replication research.” A preprint by Flake and colleagues, available at PsyArXiv, offers a systematic review that scrutinizes measurement methodologies in psychology.
  • “Stellar nurseries aren’t just smooth distributions of gas — they’re very clumpy and filamentary. And when we’re looking at a flat picture, we often can’t tell how far a certain structure extends into the depth of the cloud. But when we have a tool like this 3D-printed object, it’s inherently interactive, and we can see a structure sort of winding its way through the cloud.” Talk about data visualization: this Quanta interview with astrophysicist Nia Imara showcases her uniquely artistic approach to representing enormously complex simulation data.


Photograph showing upright case clocks of different sizes, all showing different times on their faces. Image credit: Lucian Alexe/Unsplash
mage credit: Lucian Alexe/Unsplash
  • “Duke scientists recently discovered that the pace of aging may also serve as a powerful predictor of an individual’s risk of developing Alzheimer’s disease or other cognitive decline….To determine the pace of aging, researchers use what they call “epigenetic clocks.” Our experiences and environmental exposures leave little signatures on our DNA, and those signatures, or methylation marks, accumulate and can help scientists measure how fast or slow a person is aging.” Duke School of Medicine’s Alissa Kocer showcases recent work by Duke researchers that may shine new light on relationships between the differential rates at which people age and the likelihood of developing dementia.
  • “From 2000 to 2014, a greater decrease in violent crime at the community area level was associated with a greater decrease in cardiovascular and coronary artery disease mortality rates in Chicago. These findings add to the growing evidence of the impact of the built environment on health and implicate violent crime exposure as a potential social determinant of cardiovascular health. “ A longitudinal study by Eberly and colleagues, published this month in the Journal of the American Heart Association, traces the association between violent crime and cardiovascular death in Chicago communities.
  • “There are other innate challenges. People who contract monkeypox are infectious until all their lesions heal, a process that can take several weeks. During that time, they should be in isolation, but that’s a big ask, said Jay Varma, director of the Cornell Center for Pandemic Prevention and Response at Weill Cornell Medicine.” A STAT News article by Helen Branswell collects a largely pessimistic consensus on the likelihood of stopping the spread of monkeypox infections.
  • “The technique adds to a growing list of ‘DNA origami’ tricks that are being used to build structures on the molecular scale. The approach aims to find applications in fields such as chemical synthesis and drug delivery.” At Nature, Davide Castelvecchi describes findings from a recent report on the creation of a nano-scale wind-up “ratchet motor” made of DNA strands.

COMMUNICATION, Health Equity & Policy

Photograph of a jumble of metal moveable type letters and numbers. Image credit: Amador Loureiro/Unsplash
Image credit: Amador Loureiro/Unsplash
  • “Although telehealth predates the pandemic, the service has grown massively since the pandemic arrived in North Carolina in March 2020. The unprecedented expansion provided researchers nationwide with a natural experiment: could more telemedicine mean more people will see a doctor? Could the expansion of telehealth help eliminate some disparities in access — especially for rural residents or for people who struggled with getting transportation to and from their appointments?” At North Carolina Health News, Clarissa Donnelly-DeRoven weighs the effects of a pandemic-fueled increase in use of telehealth services in the state, and finds a complex picture.
  • “Advancing inclusive research is complex, involving genomic intricacies and intersecting social drivers of health. Achieving broader diversity, equity, and inclusion in clinical trials requires nothing less than a universal commitment to diverse, equitable and inclusive research that will lead to better medical treatments for more people.” A STAT News opinion article by Nicholas Kenny, Keri McDonough and Stephen Keith makes a case for stepping up efforts to increase the diversity among clinical trial participants in the wake of a recent draft guidance from the US Food and Drug Administration.
  • “A globally recognized naming process makes an otherwise confusing name game more manageable. It helps the medical community easily learn and categorize newly approved medications and reduce prescribing errors by providing a unique, standard name that reflects each active ingredient in the drug.” In an article for The Conversation, pharmacotherapy professor Jasmine Cutler explains that drug names, despite appearances, are more than just an arbitrary pile of syllables.
  • “To our knowledge, this analysis is the first to examine semantic shift in biomedical preprints and pre-publication peer-reviewed text, and it lays the foundation for future work to examine how terms acquire new meaning and the extent to which that process is encouraged or discouraged by peer review.” A preprint by Nicholson and colleagues, available from bioRxiv, examines shifts in the use and meaning of particular words within the biomedical literature as a marker of emerging diseases and medical technologies (H/T @RetractionWatch).
  • “The paper is a model for research communication. It describes, in accessible language, how Mendel established controls and protected the integrity of his experiments (such as taking steps to reduce the risk of wind-blown or insect pollination). He is generous in crediting others’ work on the subject. The final part of the manuscript includes a discussion of caveats and potential sources of error.” On the occasion of his 200th birthday, the editors of Nature celebrate pioneering proto-geneticist Gregor Mendel as a model worthy of emulation for the quality (and collegiality) of his scientific work.