In this week’s Duke AI Health Friday Roundup: nurses’ judgment vs the algorithm; mapping mutations in primate and human genomes; GPT4 tackles differential diagnosis; the maternal death crisis threatening Black women; AI for drug design; ventilation as public-health priority; cut-and-paste errors proliferate in EHRs; aspirin and anemia in older adults; applying Ubuntu to AI ethics; informed consent in psychedelic research; cognitive impairments bring financial peril for elderly; much more:
AI, STATISTICS & DATA SCIENCE
- “What this would look like in practice, Hopkins says, is performing deep, multi-omics (single-cell proteomics, transcriptomics, and genomics) analyses of participants before the trial starts to identify multi-gene signature biomarkers. This will help the researchers to determine which participants are most likely to respond — and why.” A feature article by Nature Medicine’s Carrie Arnold offers a look at the emerging world of AI-enabled drug design.
- “A generative AI model provided the correct diagnosis in its differential in 64% of challenging cases and as its top diagnosis in 39%. The finding compares favorably with existing differential diagnosis generators… Generative AI is a promising adjunct to human cognition in diagnosis. The model evaluated in this study, similar to some other modern differential diagnosis generators, is a diagnostic “black box”; future research should investigate potential biases and diagnostic blind spots of generative AI models.” A research letter by Kanjee and colleagues, published this month in JAMA, examines the accuracy of the GPT4 large language model when it is prompted to provide differential diagnoses for challenging cases
- “While Beebe can override the AI model if she gets doctor approval, she said she faces disciplinary action if she’s wrong. So she followed orders and drew blood from the patient, even though that could expose him to infection and run up his bill. “When an algorithm says, ‘Your patient looks septic,’ I can’t know why. I just have to do it,” said Beebe…” An article by the Wall Street Journal’s Lisa Bannon explores what happens when nurses’ ability to immediately overrule an AI-based clinical algorithm is restricted.
- “Electronic copying and pasting has become a major problem in health care. A recent study of 100 million notes, consisting of 33 billion words, in the electronic medical record at the University of Pennsylvania Medical Center found that more than half of the text was duplicated. This copied-and-pasted content was prevalent in notes written by everyone from nurses and therapists to physicians at all levels of training.” An opinion article at STAT News by Sandeep Jauhar examines some new wrinkles on an old problem – the tendency for cut-and-paste errors to proliferate in medical records.
- “The UK has a stated ambition to be a global leader in the democratic development of AI. This can only be realised if the voice and interests of the public, communities, civil society and independent and interdisciplinary experts are given the same esteem currently granted to a small segment of businesses. Indeed, any claim the UK currently has to global leadership on AI is a product of its strong academic and civil society ecosystem.” Connected by Data offers a critique and rebuttal to a recent UK government prospectus outlining future approaches to AI regulation in Great Britain.
- “Existing ethical principles in healthcare are based on, and mostly influenced by, Western epistemology, which emphasizes individual rights, often at the expense of collective well-being. The African philosophy of Ubuntu, which emphasises the interconnectedness and interdependence of all people, is an attractive framework for addressing ethical concerns in AI for healthcare because healthcare is intrinsically a community-wide issue.” A paper presented at a recent ACM conference by Amugongo and colleagues presents an alternative framework for thinking about AI ethics.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Our results demonstrate the successful pairing of primate population sequencing with state-of-the-art deep learning models to make meaningful progress toward solving variants of uncertain significance. Primate population sequencing and large-scale human sequencing are likely to fill complementary roles in advancing clinical understanding of human genetic variants.” A research article by Gao and colleagues published in Science provides a newly comprehensive survey of benign “missense” mutations across human and primate genomes made possible by machine-learning approaches.
- “One of the paramount lessons of the Covid-19 pandemic is that fresh air matters. Although officials were initially reluctant to acknowledge that the coronavirus was airborne, it soon became clear that the virus spread easily through the air indoors. As the pandemic raged on, experts began urging building operators to crank up their ventilation systems and Americans to keep their windows open. The message: A well-ventilated building could be a bulwark against disease.” The New York Times’ Emily Anthes offers a historical perspective on the public-health importance of good ventilation – a long-neglected issue brought back into sharp focus by the COVID pandemic.
- “Schultz’s team and others have used this approach to genetically encode more than 200 non-standard amino acids into proteins, providing a powerful tool with which to study protein structure and function. Scientists could, for instance, introduce fluorescent markers or other labels into proteins, or conduct photocaging experiments, in which proteins are rendered inactive (‘caged’) by chemical groups that can be removed with light.” A Nature technology feature by Diana Kwon examines new frontiers in genetic engineering as scientists use sophisticated techniques to expand the range of the possible in customized proteins.
- “Whereas only a minority of participants are likely to experience major bleeding, this study demonstrates an increased risk for anemia likely due to lesser degrees of bleeding (including occult blood loss) induced by aspirin. With increasing years of cumulative exposure, such ongoing blood loss may cause a substantially larger fraction of older persons who are treated with low-dose aspirin to develop anemia.” A research article by McQuilten and colleagues, published in the Annals of Internal Medicine, finds that regular use of low-dose aspirin may result in anemia in older adults.
COMMUNICATION, Health Equity & Policy
- “In most cases, an older adult’s dementia is progressive. The first signs are often memory slips and changes in high-level cognitive skills related to organization, impulse control, and the ability to plan — all critical for money management. And because the causes of dementia vary, so do the financial woes it can create, said Hilsabeck.” A Kaiser Health News/NPR story by Sarah Boden delves into the financial perils threatening older Americans with dementia (or even mild cognitive impairments), many of whom are besieged by relentless scams and fraud attempts.
- “Growing attention to promising clinical results of psychedelic treatments are bringing important ethical and policy challenges to the field of mental healthcare. Currently, the field does not have a unified and empirically informed administration protocol and setting for each of these substances.” A commentary in Nature Medicine by Seybert and colleagues examines the potential challenges posed to informed consent processes by mental health research involving psychedelic drugs.
- “…despite COPE and ICMJE guidelines and the existence of reference management software services such as Zotero or EndNote, retracted articles are not universally marked as such across multiple websites hosting references or full text articles. Retracted articles continue to be cited long after their retraction, propagating errors and possibly affecting the accuracy of subsequent analyses. Guidelines regarding retracted articles have existed for almost 25 years but have failed to improve the situation.” An analysis by Boudry and colleagues published in BMJ addresses a long-standing problem in the scientific literature – the fact that retracted research articles continue to be cited after they are retracted.
- “…yes, it is true that someone with less access to a healthy environment is often less healthy. But that isn’t the whole story. It seems that providers are simply more comfortable talking about social determinants of health than they are with doing the hard work of tackling bias.” The horrific death of Olympic athlete Tori Bowie – only the latest in a string of high-profile incidents in which Black women, all elite athletes, have experienced life-threatening or lethal complications in childbirth – prompts a STAT News essay by Omare Jimmerson, who points out the elevated risks of maternal death that too many Black and Brown women face.