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.

October 13, 2023

In this week’s Duke AI Health Friday Roundup: the hidden influence of chronobiology; AI predicts immune escape; comparing COVID surveillance systems; yet another way to cheat at citations; updating models for ICU algorithm degrades performance; new “cooling” chemicals in cigarettes dodge menthol ban; AI image generator can’t be coaxed away from biased images; stroke deaths poised to rise in coming years; shedding light on AI’s dark corners; much more:


An ESCAPE (ESC) key from a computer keyboard, with the letters illuminated in red. Image credit: Jose G. Ortega Castro/Unsplash
Image credit: Jose G. Ortega Castro/Unsplash
  • “EVEscape provides surprisingly accurate early predictions of prevalent escape mutations but cannot anticipate all constraints unique to a new pandemic to determine the precise trajectory of viral evolution. This method will be best leveraged in synergy with experiments developed to measure immune evasion and enhanced with pandemic data as they become available.” In a research article published this week in Nature, Thadani and colleagues describe the use of a deep learning system to predict viral “immune escape” in the context of a pandemic.
  • “AI’s been around a really long time—since the ’60s, possibly longer, depending on how you define it. But we’ve had a number of what we call AI winters. Most of us have said AI’s kind of smoldering along in the image analysis space. But this year with the unveiling of what’s now being called generative AI, large language models, ChatGPT, everyone is seeing that the genie’s lamp has been opened and we’re having our wishes. No question, we are now in a place where AI is going to be a part of health care in a very visible, meaningful, and pervasive way.” In an interview with JAMA editor Kirsten Bibbs-Domingo, Charles Johnson talks about disappointments in the wake of promises that electronic health records would ease administrative burdens on physicians – and whether AI might be able to finally fulfill those promises.
  • Reuters rounds up multiple governmental efforts – including those of the US, UK, China, EU and UN – to impose some legal and regulatory controls on AI products.
  • “Try as they might, the team was unable to get Black doctors and white patients in one image. Out of 150 images of HIV patients, 148 were Black and two were white. Some results put African wildlife like giraffes and elephants next to Black physicians.” NPR’s Carmen Drahl reports on the results of an informal experiment by Arsenii Alenichev, in which the image-generating Midjourney AI offered up almost anything other than the results requested by prompts such as “Black African doctors providing care for white suffering children.”
  • “In simulated ICU settings, a universally effective model-updating approach for maintaining model performance does not seem to exist. Model use may have to be recorded to maintain viability of predictive modeling.” In a simulation study published in the Annals of Internal Medicine by Vaid and colleagues, the authors found degradations in performance of clinical prediction models following model retraining.


A woman in a green patterned blouse stands against a whitewashed brick wall. She holds a large red analog wall clock in front of her face with both hands. Image credit: Rodolfo Barreto/Unsplash
Image credit: Rodolfo Barreto/Unsplash
  • “…you are not, biochemically, the same person at 10 p.m. that you are at 10 a.m. It means that evenings are a more dangerous time to take large doses of the painkiller acetaminophen: Liver enzymes that protect against overdose become scarce then. It means that vaccines given in the morning and evening work differently, and that night-shift workers, who chronically disobey their clocks, have higher rates of heart disease and diabetes.” Quanta’s Veronique Greenwood profiles pioneering chronobiologist Carrie Partch.
  • “In 2023, unedited headlines in leading newspapers have read, ‘Rectal Cancer Patients Could Be Spared the Brutal Effects of Radiation’ and ‘More Women With Breast Cancer Could Skip Harsh Radiation, Study Says.’ By using adjectives such as brutal and harsh, these headlines sadly propagate and amplify fears and misconceptions regarding radiotherapy that research has shown to be unfounded and that negatively affect patient care.” A viewpoint article by Shaverdian and Jagsi, published in JAMA Oncology, takes exception to the ways radiation therapy for cancer is being described in the general press (H/T @CJTsaiMDPhD).
  • “During the COVID-19 pandemic, cases were tracked using multiple surveillance systems. Some systems were completely novel, and others incorporated multiple data streams to estimate case incidence and prevalence. How well these different surveillance systems worked as epidemic indicators is unclear, which has implications for future disease surveillance and outbreak management.” A research article published in Lancet Public Health by Brainard and colleagues compares different systems used for COVID surveillance in England.
  • “A chemical analysis of the “non-menthol” cigarettes detected a synthetic cooling agent, named WS-3, in four of the nine currently marketed products. WS-3 produces a cooling effect, but lacks the minty smell of menthol, allowing these products to bypass regulations. The researchers also detected vanilla and tropical flavor chemicals in “non-menthol” cigarettes, contained in flavor capsules in the filters.” Duke University’s Sarah Avery describes a newly published research study by authors from Duke and Yale that reveals chemical additives to cigarettes that appear to be an attempt to circumvent a coming ban on menthol cigarettes.
  • “Even though the rate of stroke deaths is expected to decline globally, the absolute number of deaths is set to increase in poorer countries, the Lancet report said. By 2050, stroke deaths in low- and middle-income countries are set to make up 91% of stroke deaths worldwide. In particular, deaths in the region of southeast Asia, east Asia and Oceania are projected to account for 51% of deaths globally.” STAT News’ Elaine Chen reports on a study, published this week in Lancet Neurology, that projects steep increases in the absolute number of strokes in the coming decades.

COMMUNICATION, Health Equity & Policy

Black and white high contrast photograph showing an illuminated desk lamp with articulated head against a dark background. Image credit: Robert Stemler/Unsplash
Image credit: Robert Stemler/Unsplash
  • “…the rapid proliferation of AI within healthcare has undeniably shaped a new realm of good possibilities, empowering comprehensive analyses and detailed solutions. But the integration of AI into clinical practice ought to entail a scope of ethics beyond medical diagnosis and treatment. As AI ethicists grapple with the interconnected challenges of labor exploitation, environmental degradation, and socio-economic injustice, it is essential to shed light on the obscured corners of AI networks.” A commentary in the American Journal of Bioethics by Duke AI Health’s Matthew Elmore probes the ethical implications of AI in medicine – implications that go beyond the immediate confines of the healthcare setting.
  • “But Cabanac noticed something weird: The study had been cited 107 times according to the ‘Altmetrics donut,’ an indicator of an article’s potential impact, yet it had been downloaded just 62 times….After a little probing, Cabanac and his sleuthing colleagues figured out where the extra citations were coming from: the metadata files submitted to Crossref, a repository for unique identifiers for scholarly metadata, as the group report in a preprint posted to the arXiv server October 4.” Retraction Watch describes the latest variation on publications-juking skullduggery: a stealthy technique that inflates citation counts.
  • “The inclusion of race and ethnicity in algorithms is problematic when it perpetuates the false notion that race is a biological construct. However, inclusion of race and ethnicity in algorithms to intentionally tackle inequities by accounting for environmental factors such as structural racism and social determinants of health can be beneficial. This race-aware approach acknowledges the role of structural racism in patient outcomes.” A commentary article published in Health Affairs by Hernandez-Broussard and colleagues address ways to incorporate data about race and ethnicity in clinical decision-making in ways that benefit patients while avoiding the perpetuation of bias.
  • “Imagine the absurdity of being able to apply for only one job at a time, waiting months for feedback before considering another opportunity. Such a scenario would undeniably hinder career progression. Yet, this exact practice persists in scientific publishing….The time is ripe to reassess the single-submission policy.” A Nature Careers column by Dritjon Gruda offers an eyebrow-raising suggestion for scientific journals: allow authors to submit a single paper to multiple journals simultaneously.