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 18, 2025
In this week’s Duke AI Health Friday Roundup: why some LLMs start off confidently – and then retreat; organoids grow vascularized tissue; modeling the feedback loops that are straining scientific publishing; how AI tools can change the experience of disability; how to approach continuous variables; “exposome” elements that affect aging; ML model detects structural heart disease from ECG data; much more:
AI, STATISTICS & DATA SCIENCE
- “…we introduce a deep learning model, EchoNext, trained on more than 1 million heart rhythm and imaging records across a large and diverse health system to detect many forms of structural heart disease. The model demonstrated high diagnostic accuracy in internal and external validation, outperforming cardiologists in a controlled evaluation and showing consistent performance across different care settings and racial and/or ethnic groups.” In a research article published in Nature, Poterucha and colleagues present findings from a study that examined the use of an AI model to detect structural heart disease using data from ECGs.
- “In the rapidly growing literature on explanation algorithms, it often remains unclear what precisely these algorithms are for and how they should be used. In this position paper, we argue for a novel and pragmatic perspective: Explainable machine learning needs to recognize its parallels with applied statistics….we must think carefully about the matter of interpretation, or how the explanations relate to intuitive questions that humans have about the world.” In a preprint available from arXiv, Bordt and colleagues make a case for improving explainable machine learning by drawing upon lessons from domain of applied statistics.
- “The flawed handling of continuous variables is a longstanding problem evident from reviews done in various fields, including prediction modelling, prognostic factor research, and causal inference. Categorisation (especially dichotomisation) is common but linear relationships are also often assumed without thought.” A research methods paper published by Lopez-Ayala and colleagues in BMJ offers a primer on appropriate statistical approaches to continuous variables.
- “We show that LLMs — Gemma 3, GPT4o and o1-preview — exhibit a pronounced choice-supportive bias that reinforces and boosts their estimate of confidence in their answer, resulting in a marked resistance to change their mind. We further demonstrate that LLMs markedly overweight inconsistent compared to consistent advice, in a fashion that deviates qualitatively from normative Bayesian updating. Finally, we demonstrate that these two mechanisms — a drive to maintain consistency with prior commitments and hypersensitivity to contradictory feedback — parsimoniously capture LLM behavior in a different domain.” A group of authors from Google have released a paper (available as a preprint on arXiv) that traces why some LLMs when queried start out confident, then retreat in seeming confusion.
- “Access to information is not the same as expertise in interpreting it. Healthcare providers might be wary of LLMs’ enabling patients to challenge their professional opinion on the basis of shallow appraisals of the relevant literature. Yet well-designed AI systems may be able to help individuals understand, in great detail, why experts reach certain conclusions. Done well, this could blunt the ‘epistemic fragmentation’ that fuels polarization.” A viewpoint article published in Nature Medicine by Thomas Costello suggests potential for large language models to help blunt the onslaught of medical misinformation and resulting mistrust.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Global disparities revealed more accelerated aging in participants from African countries (Egypt, South Africa) and LACs than in participants from Europe and Asia (China, South Korea, Israel and India). Individuals from lower-income countries showed larger BBAGs [biobehavioral age gaps] than those from high-income countries, suggesting the adverse health effect of socioeconomic inequalities. Adverse physical, social and sociopolitical exposomal factors were associated with accelerated aging.” A research article published in Nature Medicine by Hernandez and colleagues corrals a list of “exposome” elements that affect healthy aging.
- “The resulting lung organoids, when transplanted into mouse models, matured into a range of cell types, including unique cells found in the alveoli, sacs in which gas exchange takes place. When the researchers transplanted the cells onto a 3D scaffold, they spontaneously formed structures that looked like alveolar sacs, suggesting that the presence of blood-vessel cells had allowed the organoids to form complex tissue with diverse cell types.” Nature’s Smriti Mallapathy reports on recent efforts capitalizing on a serendipitous finding that have produced organoids with vascularized tissue.
- “A gene variant known to increase the risk of Alzheimer’s disease also makes people vulnerable to a host of other age-related brain disorders, from Parkinson’s disease to motor neuron disease. The gene variant, a version of apolipoprotein E called APOE ε4, produces a distinct set of proteins that contribute to chronic inflammation…” Another Nature news feature by Mallapathy highlights recent work revealing that one of the genes implicated in Alzheimer disease may be responsible for inflammation that also raises risk for a variety of other neurological diseases.
- “One of the things that you have to get used to in science (and especially biomedical science) is the constant possibility that something that Everybody Knows will turn out to be wrong. Today’s installment is the way that Everybody Knows that the brain is an obligate user of glucose for fuel.” At In the Pipeline, Derek Lowe highlights a new study demonstrating that, contrary to previous understandings of brain physiology, some kinds of neurons use triglycerides to store energy.
COMMUNICATION, Health Equity & Policy
- “Removing medical debt from consumer credit reports was expected to increase the credit scores of millions of families by an average of 20 points, the bureau said. The CFPB states that its research has shown outstanding healthcare claims to be a poor predictor of an individual’s ability to repay a loan, yet they are often used to deny mortgage applications.” STAT News relays an AP article reporting on the judicial reversal of rule enacted by the Consumer Financial Protection Bureau that removed medical debt from credit reporting.
- “Shifting illocutionary force means changing the conventions that give it power – moving away from seeing every cancer as a battle to be won at all costs, and toward treating it like other diseases, where decisions carefully consider a patient’s preferences, along with the risks and benefits of different options. That shift requires action across many fronts.” In an essay for Aeon, Ben Chin-Yee considers the enormous freight carried by the way we think of and talk about cancer diagnoses and treatments.
- “…we develop mathematical models to reveal the intricate interactions among incentives faced by authors, reviewers, and readers in their endeavors to identify the best science. Two facets are particularly salient. First, peer review partially reveals authors’ private sense of their work’s quality through their decisions of where to send their manuscripts. Second, journals’ reliance on traditionally unpaid and largely unrewarded review labor deprives them of a standard market mechanism — wages — to recruit additional reviewers when review labor is in short supply.” A preprint by Carl Bergstrom and Kevin Gross, available from arXiv, presents mathematical models that illuminate the feedback loops placing stress on the quality of scientific publishing.
- “…we’ve been framing this in terms of assistive technologies that could help people in various disability communities, but what you’re describing is technology. that it has the potential of changing everybody’s lives, right? It has the possibility of resetting what we all see, whether we’re in a disability community or not. As the kind of ways that we want to interact and engage with the world.” An episode of the WBUR public radio podcast program On Point considers how AI tools are opening up new avenues in assistive and adaptive technology for persons with disabilities.
