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.

March 8, 2024

In this week’s Duke AI Health Friday Roundup: toward generalist medical AI; limited benefit for rapid respiratory virus testing in ED; why successful health AI is more than algorithms; discussion paper examines AI impacts for Black community; epithelial organoids cultivated from stem cells in amniotic fluid; Coalition for Health AI debuts as nonprofit, announces leadership; Alzheimer disease biomarkers present long before clinical diagnosis; much more:

AI, STATISTICS & DATA SCIENCE

“Hypoglycemia could be detected noninvasively during real car driving with an ML approach that used only data on driving characteristics and gaze/head motion, thus improving driving safety and self-management for people with diabetes. Interpretable ML also provided novel insights into behavioral changes in people driving while hypoglycemic.” In an article published last month in NEJM AI, a group of Swiss and German researchers report on results from a study of a machine learning model designed to detect potentially dangerous hypoglycemia in drivers.
“At first glance, the Coalition for Health AI looks like any technology lobbying group. Its membership — including Microsoft, Google, and many of the nation’s largest academic hospitals — forms a formidable presence in Washington….But on Monday, the coalition forged a different deal with federal regulators. It will work with them to develop quality and safety standards for artificial intelligence, an experiment that will test whether industry and government can effectively partner in the regulation of a fast-moving technology.” STAT News’ Casey Ross reports on the Coalition for Health AI’s emergence as a nonprofit entity that will include representation from federal regulatory and public health agencies on its newly announced board of directors.
“Salient efforts from both academia and industry have validated the utility of retrospective data to enable data-driven decision-making for [assisted reproductive technology]…To ensure viable deployment, these models can benefit from larger, multi-center datasets that incorporate both heterogeneous patient populations and also capture the idiosyncratic nature of clinical practice worldwide. Achieving this is best achieved through a collaborative effort from all stakeholders representing multiple disciplines across the AI and healthcare landscape.” A review article published in NPJ Digital Medicine by Hanassab and colleagues explores the potential for AI applications to tailor therapy for reproductive medicine.
“Initially, nobody could explain why the newly discovered patterns exist. Since then, in a series of recent papers, mathematicians have begun to unlock the reasons behind the patterns, dubbed “murmurations” for their resemblance to the fluid shapes of flocking starlings, and have started to prove that they must occur not only in the particular examples examined in 2022, but in elliptic curves more generally.” Quanta’s Lyndie Chiou reports on machine-learning-driven analyses that revealed striking patterns in elliptic curves.
“AI algorithms tend to excel in controlled environments, where only specific predictive features may influence the clinical outcome. However, patients’ and providers’ inherently human nature introduces numerous challenges, causing even the most robust AI models to degrade over time. Diversity in patient characteristics, disease presentations, practice patterns, and evolving treatment paradigms contribute to the potential failure of algorithms post-deployment…” Also in NPJ Digital Medicine: an editorial by Kwong and colleagues uses the example of recent research by Boussina and colleagues to point that successful integration of AI into clinical care goes well beyond the algorithm in isolation.
“AI progress to date has largely been catalyzed by the development of high-quality benchmarks. Although several single-task biomedical AI datasets exist, there have been limited attempts to unify them and create benchmarks for the development of generalist biomedical AI systems. Our curation of MultiMedBench is a step toward addressing this unmet need.” A research article published in NEMJ AI by Tu and colleagues describes a new multimodal performance benchmark to encompass the kinds of tasks a “generalist” medical AI might be called on to perform.

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

“Here, we describe the discovery, optimization, and characterization of a broadly neutralizing 3FTx-L antibody that exhibits protective efficacy in mice against lethal challenge by venom from a range of medically relevant snakes. Our project workflow provides a generalizable strategy for finding antibodies that target conserved sites on antigenically variable proteins. The utilization of recombinantly produced toxins allowed for a high degree of control over antigenic variability and bypassed the need to purify the necessary toxins from multiple snake species.” A research article published in Science by Khalek and colleagues has some potentially good news for the cobra-adjacent.
“Overall, the results of this systematic review and meta-analysis suggest that the benefits of routine RV testing in the ED are limited. Such testing in EDs has no association with overall antibiotic use, length of ED stay, ED return visits, or hospitalization rates. Testing was associated with a minority of patients with influenza being prescribed antivirals and decreases in ordering of some ancillary tests. Patients with positive viral test results received less antibiotics compared with patients with negative test results, possibly improving the appropriateness of antibiotic treatment in this subgroup.” A meta-analysis of outcomes associated with the use of rapid testing for respiratory viruses in emergency medicine settings published this week in JAMA Internal Medicine by Schober and colleagues finds limited clinical benefit for the practice.
“In this study assessing change in CSF biomarkers in 648 persons who ultimately received a diagnosis of Alzheimer’s disease and the same number of matched persons who remained cognitively normal, the times before Alzheimer’s disease diagnosis at which biomarkers diverged between groups ranged from 18 years for CSF Aβ₄₂ concentration to 6 years for cognitive decline as measured on the CDR-SB, a scale that has been widely used in clinical trials.” A case-control study published in the New England Journal of Medicine by Jia and colleagues reports on changes in a constellation of biomarkers in participants in a longitudinal study of cognition and aging up to 20 years before a clinical diagnosis of Alzheimer disease.
“This work demonstrates that the AF contains tissue-specific fetal epithelial progenitor cells originating from various developing organs. We show that, under defined culture conditions, these cells form epithelial organoids resembling their tissues of origin (small intestine, kidney and lung). Finally, we provide evidence that lung organoids derived from AF and TF of fetuses affected by CDH exhibit features of the disease.” A research article published in Nature Medicine by Gerli and colleagues describes the successful derivation of organoids from epithelial stem cells harvested from amniotic fluid.

COMMUNICATION, Health Equity & Policy

“While AI holds the potential to deepen racial inequalities, it can also benefit Black communities. If deployed carefully, AI has the power to improve access to healthcare and education, as well as create new economic opportunities. For example, AI can help doctors make more accurate diagnoses and provide personalized treatment plans, particularly in underserved communities where access to healthcare is limited. AI can also assist educators in tailoring lessons to individual student needs, increasing the chances of academic success for all students, including those from low-income and minority communities…. Our vision for human-centered AI is rooted in the belief that AI should be assistive, augmenting, and complementing human capabilities but never replacing human judgment.” A new discussion paper, the product of a collaboration between Black in AI and Stanford’s Human-Centered Artificial Intelligence institute examines AI’s potential for both harm and benefit for communities.
“…academia has an enormous stake in imperatives like ensuring the trustworthiness of the scholarly record, providing for the platforms through which humans and machines will engage with scholarship, and addressing the atomization of the scholarly article. These topics demand collaboration by academia and research publishers. The current investments in developing AI-powered tools that support scholarly communication — and in resisting some of the challenges posed by the use of AI — makes these kinds of partnerships only more important and urgent.” An essay at Scholarly Kitchen by Roger C. Schonfeld examines the alignment of the publishing marketplace and emerging priorities and incentives in academia. (With some additional commentary from Richard Sever.)
“We contend that the ideal instructional sequence to reduce genetic essentialism is to introduce students to the models of Mendelian genetics…and then move beyond these models and highlight their limitations using a humane genomics curriculum…. This prediction is most likely to be accurate when highly trained biology teachers implement this instructional sequence in a high school located in one of the states where our results have high generalizability…” A Policy Forum article published in Science by Donovan and colleagues argues that going “beyond Mendel” in basic education about genetic concepts can reduce racist and essentialist biases.