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 7, 2024

In this week’s Duke AI Health Friday Roundup: countering hype around AI for discovery science; CO2 levels associated with risk from airborne disease; disproportionate effects of Medicaid disenrollment; a path forward for AI in nursing; references to nonexistent cell lines reveal tracks of paper mill publications; medicine stares down challenges of heart disease in coming years; deciding whether a “frictionless” experience in instruction is actually desirable; much more:

AI, STATISTICS & DATA SCIENCE

“AI can’t replace nurses, nor should any administrator or policymaker contemplate such a disastrous mistake. On the other hand, AI can help nurses and other health professionals deliver better, more efficient, more personalized care. By leveraging AI, we can not only overcome long-standing challenges, such as increasing workloads and burnout, but also elevate our practice in ways that were unimaginable a decade ago.” In a perspective available at the American Nurse website, Duke nursing professor Michael Cary argues that the nursing profession stands to benefit from a thoughtful approach to incorporating AI tools into practice.
“With the expanding integration of artificial intelligence (AI) and machine learning (ML) into the structural heart domain, numerous ML models have emerged for the prediction of adverse outcomes following transcatheter aortic valve implantation (TAVI). We aim to identify, describe, and critically appraise ML prediction models for adverse outcomes after TAVI.” In a review article published in the Journal of Thoracic and Cardiovascular Surgery, Jacquemyn and colleagues examine the literature reporting on use of machine learning approaches for predicting risk in the procedure known as transcatheter aortic valve implantation (TAVI).
“…we think the biggest reason for the poor quality of research is pervasive hype, resulting in the lack of a skeptical mindset among researchers, which is a cornerstone of good scientific practice. We’ve observed that when researchers have overoptimistic expectations, and their ML model performs poorly, they assume that they did something wrong and tweak the model, when in fact they should strongly consider the possibility that they have run up against inherent limits to predictability. Conversely, they tend to be credulous when their model performs well, when in fact they should be on high alert for leakage or other flaws. And if the model performs better than expected, they assume that it has discovered patterns in the data that no human could have thought of…” In a recent entry at their AI Snake Oil blog, Arvind Narayanan and Sayash Kapoor address reproducibility, leakage, and hype surrounding breathless reports of the utility of AI application for discovery science.
“Here we introduce Aurora, a large-scale foundation model of the atmosphere trained on over a million hours of diverse weather and climate data. Aurora leverages the strengths of the foundation modelling approach to produce operational forecasts for a wide variety of atmospheric prediction problems, including those with limited training data, heterogeneous variables, and extreme events. In under a minute, Aurora produces 5-day global air pollution predictions and 10-day high-resolution weather forecasts that outperform state-of-the-art classical simulation tools and the best specialized deep learning models.” A preprint paper by Bodnar and colleagues posted at Microsoft Research Blog (and also available from arXiv) describes Microsoft’s Aurora, a foundational model designed for meteorological prediction (H/T @erichorvitz).

BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH

“Understanding community mixing patterns can inform infectious disease risk, support analyses to predict epidemic size, or be used to design campaigns such as vaccination strategies so that community members who have vulnerable contacts are prioritized.” A research article by Pasquale and colleagues, published in PLOS ONE, examines how patterns of sociodemographic associations interact with the risk of epidemic diseases.
“No sensor can monitor how many infectious aerosols are swirling around us in real time. But carbon dioxide, or CO2, can act as a convenient proxy. People exhale it when they breathe, and in spaces that aren’t well ventilated, the gas accumulates. High CO2 concentrations can provide a warning sign that a lot of the air you’re inhaling is coming out of other people’s respiratory tracts.” In an article for STAT News, Megan Molteni explores recent research that suggests previously unrecognized links between CO₂ concentrations in room air and the amount of risk exposure for infectious airborne pathogens.
“Research integrity sleuths may have found a new red flag for identifying fraudulent papers, at least in cancer research: Findings about human cell lines that apparently do not exist. That’s the conclusion of a recent study investigating eight cell lines that are consistently misspelled across 420 papers published from 2004 to 2023, including in highly ranked journals in cancer research. Some of the misspellings may have been inadvertent errors, but a subset of 235 papers provided details about seven of the eight lines that indicate the reported experiments weren’t actually conducted, the sleuths say.” In an article for Science, Jeffrey Brainard examines recent research that has uncovered a large tranche of papers whose citation of nonexistent cell lines points to orchestrated research fraud.
“On the basis of US Census projections and the forecasted prevalence of cardiovascular risk factors and conditions through 2050, we project that inflation-adjusted total cost related to cardiovascular risk factors will nearly triple and total cost related to CVDS conditions will almost quadruple between 2020 and 2050. This result is driven primarily by large increases in health care spending for coronary heart disease, stroke, atrial fibrillation, and heart failure due to the expected growth in the burden of these CVDS conditions resulting from an aging population.” A Presidential Advisory from the American Heart Association, just published in Circulation by Kazi and colleagues, suggests that the health and economic impacts of cardiovascular disease are poised to exact a heavy toll in coming years unless prompt and effective action is taken.

COMMUNICATION, Health Equity & Policy

“We would all benefit from exploring this longer history of tech, including ed tech, and who it serves. We would benefit from taking the time to ask hard questions about what constitutes effective teaching and learning in our disciplines and how tech might or might not play into it. And our students would benefit from us slowing down, providing some friction, and asking the tough questions about the value and harms of technologies before deploying them.” At her “Pandora’s Bot” Substack, Katie Conrad sounds a skeptical note about the proliferation of AI platforms and tools touted as solutions that reduce “friction” in educational settings.
“Adopting a more formulaic approach to medicine, it seems, is part of a natural evolution. Clinical scoring systems that I, and many physicians, regularly employ help to predict aspects of our patients’ health we cannot reasonably foresee. Feeding in certain parameters such as a heart rate, age, or a measure of liver function allows us to retrieve various likelihoods such as the chance of having a blood clot in the lungs, a heart attack in the next decade, or a favorable outcome from steroids for someone whose liver is inflamed by alcohol….Yet the more familiar we become with these methods, the less certain we know those truths to be. A person’s health follows a multipronged course set as much by the mysteries of their biology as by the realities of where they happen to live, grow, and work.” An essay by Arjun V.K. Sharma, published in Undark, ponders the potential – and the fundamental, non-technological limitations – of AI-based tools for healthcare (H/T @AI_4_Healthcare).
“The end of the continuous enrollment provision may have reversed some coverage gains for Black and Hispanic individuals during the COVID-19 PHE [public health emergency]…This finding is consistent with state-collected data suggesting that early coverage losses are secondary to renewal processes rather than changes in Medicaid eligibility…With states resuming eligibility redeterminations, policy attention is needed to improve enrollment processes, particularly for populations more likely to experience challenges.” A research letter published in JAMA Internal Medicine by Rumalla and colleagues examines disparities revealed in Medicaid disenrollment patterns follow the COVID-19 pandemic (H/T @UREssien).