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.
June 28, 2024
In this week’s Duke AI Health Friday Roundup: AI for decoding dog barks; Surgeon General designates firearm violence as public health crisis; where next for AlphaFold?; coming to grips with AI’s water use; randomized trial evaluates extended-release ketamine for depression; parsing the implications of vaccine exemption rates; forking paths in statistics; Coalition for Health AI releases assurance standards guide; LLMs for analyzing radiology reports; how AI will impact workforces; much more:
AI, STATISTICS & DATA SCIENCE
- “Because of this dearth of usable data, techniques for analyzing dog vocalizations have proven difficult to develop, and the ones that do exist are limited by a lack of training material. The researchers overcame these challenges by repurposing an existing model that was originally designed to analyze human speech….This approach enabled the researchers to tap into robust models that form the backbone of the various voice-enabled technologies we use today, including voice-to-text and language translation.” A web article by the University of Michigan’s Emily France describes work underway at UM that uses AI in an attempt to decode human-understandable information from a dog’s bark.
- “AlphaFold2 has undeniably shifted the way biologists study proteins. However, while AlphaFold2 is a powerful prediction tool, it’s not an omniscient machine. It has solved one part of the protein folding problem very cleverly, but not the way a scientist would. It has not replaced biological experiments but rather emphasized the need for them.” In an article for Quanta, Yasemin Sapakoglu explores the questions that have arisen in the fields of biochemistry and biology in the wake of AlphaFold’s debut as a powerful tool for predicting protein folding.
- “The point of the ‘forking paths’ metaphor in statistics is that multiple comparisons can be a problem, even when there is no ‘fishing expedition’ or ‘p-hacking’ and the research hypothesis was posited ahead of time. Indeed, often we can look at existing literature or even a single published article containing multiple studies to get a sense of the ‘multiverse’ spanned by possible choices of data coding and analysis….It is good to analyze data in different ways! The mistake is to choose just one.” An essay posted by Columbia’s Andrew Gelman at the blog Statistical Modeling, Causal Inference, and Social Science examines “forking paths” in statistical practice and communication.
- “More than a century after the first radiology report was written, we are now making significant progress in developing tools to leverage large collections of radiology reports, to extract relevant information, and to identify potential connections between imaging findings and symptoms. Open-source LLMs currently appear to be the best way to deploy such systems in health care while safeguarding patient privacy and data security, despite all the possible confabulations they may suffer like the rest of the LLM family.” A commentary article in the journal Radiology: Artificial Intelligence by D’Antonoli and Bluethgen examines recently published research by Le Guellec and colleagues that explored the use of the Vicuna LLM to extract and analyze text from patient reports.
- “Our exploratory analysis estimates only 1.86% of tasks could be fully automated by LLMs plus additional software integrations without human oversight. Still, more than 71% of tasks have at least some component that an LLM plus additional software could plausibly complete with high quality…High LLM exposure in occupations correlates with higher automation potential, with automation scores explaining 55.6% of the task-level exposure variance…This suggests that occupations with greater automation risk may also have higher augmentation potential…”A Science Policy Forum article by Eloundou, Mishkin, and Rock scrutinizes the impact of LLM applications on the labor market.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “A new analysis of the island’s landscape suggests Polynesian seafarers who reached Rapa Nui around 800 years ago maintained a modest farming system and a small but stable population of no more than around 3,900 individuals until Europeans showed up in 1722 and renamed the early settlers’ homeland Easter Island…That finding challenges a popular idea that expanding Rapa Nui communities grew so large that they exhausted available resources, decimating the island society by the time Europeans arrived in what’s been called an “ecocide” event…” In an article published in Science News, Bruce Bower reports on recent research that casts doubt on popular speculation that overexploitation of the local environment by Rapa Nui islanders caused a catastrophic ecological collapse.
- “This study provides some of the first empirical evidence on the association of restrictive abortion policies with infant deaths by using population-based data and a rigorous causal inference technique. Our findings suggest increases in infant death, particularly due to congenital anomalies, among infants who would have been in early gestation when SB8 went into effect in Texas, which banned abortions after embryonic cardiac activity. Although replication and further analyses are needed to understand the mechanisms behind these findings, our results indicate that restrictive abortion policies may have important unintended consequences in terms of trauma to families and medical cost.” A research article published in JAMA Pediatrics by Gemmill and colleagues reports findings from a cohort study examining rates of infant deaths in Texas following the 2021 enactment of restrictions on abortion in that state.
- “Such high levels of exposure to firearm violence for both children and adults give rise to a cycle of trauma and fear within our communities, contributing to the nation’s mental health crisis….This Advisory describes the public health crisis of firearm violence in America and describes strategies for firearm injury and violence prevention, with a focus on the health and well‑being of children, families, and communities.” The S. Surgeon General has released an advisory declaring that U.S. firearm violence constitutes a public-health crisis.
- “…extended-release R-107 tablets were effective, safe and well tolerated in an enriched patient population with TRD [treatment-resistant depression]. Use of an extended-release oral dosage ketamine formulation may be advantageous compared with intranasal or intravenous dosing, in terms of reduced intensity of dissociation, lower risk of abuse, reduced frequency and intensity of sedative and cardiovascular side effects, and improved convenience for administration in the community.” A research article published in Nature Medicine by Glue and colleagues reports findings from a randomized trial of extended-release ketamine for treatment-resistant depression.
COMMUNICATION, Health Equity & Policy
- “…the vast majority of data centers still use potable water and cooling towers. For example, even tech giants such as Google heavily rely on cooling towers and consume billions of liters of potable water each year. Such huge water consumption has produced a stress on the local water infrastructure…Moreover, many data centers are also located in drought-prone areas such as California.” The Markup’s Nabiha Syed interviews University of California Riverside professor Shaolei Ren about the demands that AI technologies are placing on water supplies.
- “The Assurance Standards Guide serves as a playbook for the development and deployment of AI in healthcare, providing actionable guidance on ethics and quality assurance. It stems from the consensus-based approach of the CHAI community, drawing upon the collaborative work of patient advocates, technology developers, clinicians, data scientists, civil servants, bioethicists, and others. The Guide is written for an equally broad audience, encompassing everyone with rights and responsibilities in the process of designing, developing, deploying, and using AI technologies.” The Coalition for Health AI (CHAI) has released the first public version of its framework for the development and implementation of trustworthy health AI.
- “The advent of the open-access movement in science publishing in the past 20 years was meant to make the results of scientific work more widely accessible, and to a substantial degree that goal has been achieved. But many funding agencies, particularly in Europe, made the mistake of requiring that research funded by their grants be published in “gold” open-access journals — or in free journals that charge neither subscription nor publication fees but, understandably, remain quite rare….That decision gave the veneer of respectability to the paper mills that present themselves as gold open-access journals, but for all practical purposes operate as profit-oriented self-publishing venues.” The Washington Post has published a collection of rejoinders to a recent article Retraction Watch co-founders Ivan Oransky and Adam Marcus that tackled the growing problem of dubious or even fraudulent publications swamping the scientific literature.
- “One path forward would be to further invest in projects that work toward greater transparency in communicating with patients and the public about medical care and the implications of findings from medical research. At a minimum, this requires a commitment to offer clear, accessible explanations about how scientists conduct and report research, as well as to candidly discuss the strengths and limitations of scientific research.” A blog post at the American Scientist’s Macroscope by RTI’s Brian Southwell, Duke School of Medicine Dean Mary Klotman, and Tuckson Health Connections Director Reed Tuckson parses the health implications of growing rates of vaccine exemptions.
