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.
February 21, 2025
In this week’s Duke AI Health Friday Roundup: curated source for automated machine learning papers; using AI in the fight against epidemics; gene-spliced mice offer clue to human language evolution; NIH study sections remain in stasis; assessing spironolactone in myocardial infarction; when to use trial emulation; unpacking the implications of reductions in indirect costs; the role of environmental exposures in health burdens of aging; much more:
AI, STATISTICS & DATA SCIENCE
- “We cannot think that our AI systems will not be biased. That is the wrong way of going about things. We need to assume that there’s a risk of bias. Even if we think it’s zero, there is a risk of bias. There will always be a risk of bias. As long as we are aware that this risk exists, then we can take measures to try to measure it, to try to detect it, to try to correct it, and to try to avoid damage. Or in the case that damage has been taking place, then do something about it.” In an interview with JAMA +AI, Maria Villalobos-Quesada talks about the context and broader picture surrounding findings from a recent review article evaluating the evidence underpinning machine-learning algorithms used in primary care settings.
- “As the complexity of these tasks is often beyond non-ML-experts, the rapid growth of machine learning applications has created a demand for off-the-shelf machine learning methods that can be used easily and without expert knowledge. We call the resulting research area that targets progressive automation of machine learning AutoML. As a new sub-area in machine learning, AutoML has got more attention not only in machine learning but also in computer vision, natural language processing and graph computing.” A resource that provides a curated source for articles and other materials related to automated machine learning is now publicly available at GitHub.
- “…the benefits of AI for public health will be critically dependent on the availability, accessibility and representativeness of data and a firm ethical framework. Even though more data have become available during health emergencies, especially for COVID-19, routine surveillance data for infectious diseases often remains siloed and inaccessible to the broader community, prohibiting the use of these data for improved disease modelling. However, the substantial volume of existing data means that advances aided by AI will probably be generated from existing openly available datasets.” A review article published in Nature by Kraemer and colleagues explores the potential role of AI technologies in combating disease epidemics (H/T @erictopol.bsky.social).
- “The increasing popularity of target trial emulation as an approach to causal inference from observational data raises questions about the utility and scope of this approach. Two key questions are “Why is it helpful?” and “When is it helpful?”…In this article, we answer both questions. To do so, we propose an updated structure for target trial emulations, clarify what problems target trial emulation does and does not solve, and describe some settings where adopting this approach is advantageous.” In a methods article published in the Annals of Internal Medicine, Hernán and colleagues discuss the relative merits and appropriate applications of using observational data to emulate a hypothetical clinical trial.
- “Most readers are not equipped to judge the plausibility of a proposed disease mechanism, the merits of some particular drug candidate aimed at it, or the intricacies of the clinical trial data that it might generate. Ideally, one would not take advantage of this informational asymmetry, but it seems pretty clear that we do not live in such a world.” In an article at his “In the Pipeline” blog for Science, Derek Lowe throws some cold water on hype surrounding the likelihood of substantially accelerated drug development courtesy of AI-driven protein design.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Mice typically produce pulses of ultrasonic squeaks that resemble syllables in human language. But mice carrying the human version of NOVA1 made peculiar squeaks, the scientists found. The difference was especially noticeable when males sang courtship songs to females. Their songs contained more complex sounds, and the mice switched between those sounds in more intricate patterns.” A New York Times article by Carl Zimmer examines genetic research that may shed new light on the evolutionary development of language.
- “Our study provides the first assessment of the relative contributions of environmental and genetic influences on aging.….We find that the major drivers of premature death and aging in our sample are smoking, socioeconomic status and deprivation, ethnicity, physical activity, living with a partner, sleep and mental and physical wellness including tiredness, as well as early life exposures including height and body size at 10 years and maternal smoking around birth.” An article published in Nature Medicine by Argentieri and colleagues compares the relative contributions of genetics and environmental exposures to the health burdens of aging.
- “After myocardial infarction, treatment with spironolactone, as compared with placebo, did not reduce the incidence of death from cardiovascular causes or new or worsening heart failure or the incidence of composite-outcome events (death from cardiovascular causes, recurrent myocardial infarction, stroke, or new or worsening heart failure) over a median follow-up of 3 years.” In a research article published in the New England Journal of Medicine, Jolly and colleagues report the results from a randomized trial that evaluated the therapy spironolactone, widely used in patients with congestive heart failure who experience a myocardial infarction, would benefit all patients experiencing an MI.
- “In his work as an assistant professor in the department of population medicine at Harvard Medical School and Harvard Pilgrim Health Care Institute, Kanjilal analyzes real-world data using machine learning to improve the diagnosis and treatment of infectious diseases. To put one such AI tool to the test, the researchers developed 2 sets of baseline covariates that functioned as potential confounders in their analysis—1 set manually curated by Kanjilal and another infectious disease specialist, and the other derived by machine learning.” In an interview with JAMA +AI, Sanjat Kanjilal discusses recent research that used a machine learning model to assess the accuracy of 14-year-old treatment guidelines for urinary tract infections.
COMMUNICATION, Health Equity & Policy
- “The shortage of physicians in the United States isn’t a new phenomenon. Unfortunately, it is also one that isn’t expected to disappear anytime soon….In fact, demand is expected to outstrip supply for the foreseeable future. The latest prediction: The shortfall of physicians could reach as high as 86,000 by the year 2036, according to a 2024 report from the AAMC.” A Medscape article by Jennifer Larson reports on data that paints a worrisome picture of current and looming physician shortages.
- “A freeze on meetings of expert panels that peer review grant proposals at the National Institutes of Health is kicking in this week. Until now these meetings, known as study sections, were generally proceeding as long as a required notice had been posted in the government bulletin known as the Federal Register. But soon after taking office on 20 January, Trump barred NIH from posting new Federal Register notices, which means any study section not advertised before then cannot take place.” Science reports that due to a freeze on posting updates to the Federal Register, new NIH study section meetings are also effectively frozen for the time being.
- “How should we consider the value of academic health research, and perhaps, implicitly, of health itself? We would argue that the value of academic health research does not lie only in its health benefit or its contribution to the economy. The true value of academic health research lies in the ineffable foundational role it plays in enabling a country to thrive and grow. At the heart of this argument is the observation that health is a means, not an end. The purpose of health is to create the opportunity for all to live long, rich lives.” An editorial published in JAMA by JAMA Health Forum editor Sandro Galea and JAMA editor Kirsten Bibbins-Domingo use recent cuts to so-called “indirect costs” associated with grants from the National Institutes of Health as a springboard for discussing the larger issue of the value of publicly funded research.
