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.

January 6, 2023

In today’s Duke AI Health Friday Roundup: using machine learning to create “synthetic” x-rays and answer medical questions; the continuing evolution of gender-affirming care; how glass frogs manage their disappearing act; validating survival models; going beyond Tuskegee when examining medical racism; how the scholarly community deals with paywalled papers; much more:


A photographic rendering of medical swabs and pills in a jar seen through a refractive glass grid, overlaid with a diagram of a neural network. Image credit: Alan Warburton/©BBC/Better Images of AI/CC-BY 4.0
Image credit: Alan Warburton/©BBC/Better Images of AI/CC-BY 4.0
  • “Until now, foundation models trained in natural images and language have not performed well when given domain-specific tasks. Professional fields such as medicine and finance have their own jargon, terminology, and rules, which are not accounted for in general training datasets. But one advantage presented itself for the team’s study: Radiologists always prepare a detailed text report that describes their findings in each image they analyze.” Stanford’s Human-centered AI reports on work by Stanford researchers that uses the Stable Diffusion AI image generator to create realistic “synthetic” chest x-rays.
  • “This study describes the largest to-date global FL [federated learning] effort to develop an accurate and generalizable ML [machine learning] model for detecting glioblastoma sub-compartment boundaries, based on data from 6314 glioblastoma patients from 71 geographically distinct sites, across six continents…this describes the largest and most diverse dataset of glioblastoma patients ever considered in the literature. It was the use of FL that successfully enabled our ML model to gain knowledge from such an unprecedented dataset.” A research article published in Nature Communications by Pati and colleagues describes the creation and use of a federated machine learning method to identify the margins of glioblastoma tumors.
  • “Risk prediction models need thorough validation to assess their performance. Validation of models for survival outcomes poses challenges due to the censoring of observations and the varying time horizon at which predictions can be made. This article describes measures to evaluate predictions and the potential improvement in decision making from survival models based on Cox proportional hazards regression.” An article published in Annals of Internal Medicine by McLernon and colleagues describes methodological approaches for validating predictive survival models.
  • “Using a combination of prompting strategies, Flan-PaLM achieves state-of-the-art accuracy on every MultiMedQA multiple-choice dataset (MedQA, MedMCQA, PubMedQA, MMLU clinical topics), including 67.6% accuracy on MedQA (US Medical License Exam questions), surpassing prior state-of-the-art by over 17%. However, human evaluation reveals key gaps in Flan-PaLM responses. To resolve this we introduce instruction prompt tuning, a parameter-efficient approach for aligning LLMs to new domains using a few exemplars. The resulting model, Med-PaLM, performs encouragingly, but remains inferior to clinicians.“ In a preprint available from arXiv, Singhal and colleagues present findings from a variation of the PaLM large language model that was tasked with answering questions from the US Medical License Exam.


  • “In a study published in the journal Science on Thursday, researchers report that when a glass frog falls asleep, almost all of its red blood cells retreat into its liver. They hide in the organ and allow the frog to achieve near invisibility while it rests. In addition to revealing another remarkable adaptation in nature, the discovery could lead to clues for how to prevent deadly blood clots.” Now, that’s a neat trick: The New York Times’ Veronique Greenwood reports on how a group of researchers, including Duke’s Carlos Taboada, figured out how the glass frog manages its near-disappearing act. (You can read more about this research, including the imaging technology used in the study, in this article by Duke’s Michaela Kane).
  • “The field of gender-affirming care has evolved substantially in the past century and consists of several key components, including psychological, social, legal, medical and surgical affirmation. Gender affirmation requires inclusive and culturally responsive healthcare environments, as well as broader policies and laws that support and protect gender diversity locally, nationally and globally.” A commentary recently published in Nature Medicine by Keuroghlian and colleagues surveys the current landscape of gender-affirming care in medicine and clinical research.
  • “In many emergency departments, expectant management has long been the only option made available. But now, amid the legal uncertainty unleashed by the fall of Roe, Prager and colleagues say they’ve been inundated with inquiries from emergency departments across the country.” An article at NPR’s Shots blog by Patrick Adams examines efforts to improve care for miscarriages amid uncertainty created by legislation in the wake of Roe v Wade’s overturning.
  • “In patients hospitalised for COVID-19 with clinical hypoxia but requiring either no oxygen or simple oxygen only, higher dose corticosteroids significantly increased the risk of death compared to usual care, which included low dose corticosteroids.” A new MedRxiv preprint available from the UK’s RECOVERY project examining various potential COVID therapeutics reports on a randomized trial examining the use of higher-dose corticosteroid treatments for COVID patients with hypoxia but not on ventilators.

COMMUNICATION, Health Equity & Policy

Photograph showing a steel gate in the foreground with buildings out of focus behind. The gate is closed, with a padlocked chain wound through the gate’s handles. Image credit: Masaaki Komori/Unsplash
Image credit: Masaaki Komori/Unsplash
  • “…researchers most frequently look for OA versions of the documents. However, more than 50% of the participants have used a scholarly piracy site at least once. This is less common in high-income countries, and among older and better-established scholars. Regarding disciplines, such services were less used in Life & Health Sciences and Social Sciences.” A preprint article by Segado-Boj and colleagues, available at arXiv, reports on results from a survey study that examined international scholarly practices with regard to accessing publications protected by paywalls.
  • “Understanding how any given group of organisms evolved from a common ancestor to their current forms is always tricky: Imagine assembling a gigantic puzzle where more than half the pieces were lost, a quarter haven’t been found yet and the remaining ones could disappear at any moment. The fact that lichens aren’t a single organism, but a symbiotic association, greatly increases the puzzle’s complexity, so every summer, Lutzoni, Miadlikowska and their graduate students canvas the globe looking for puzzle pieces.” An article by Duke’s Marie Clair Chelini reveals the results of a remarkable combination of science and art that yielded mural-sized results from tiny organisms found literally underfoot.
  • “Despite their popularity, the plans have been the subject of considerable scrutiny and criticism lately. A recent report by the inspector general of the U.S. Department of Health and Human Services found that several plans might be inappropriately denying care to patients.” The New York Times’ Reed Abelson and Margo Sanger-Katz report on increasing regulatory attention toward Medicare Advantage health plans, triggered by concerns about misleading advertising and sharp business practices.
  • “Half a century later, this racist experiment can look like a product of a long-gone era, even as it gets outsize credit for distrust today. In reality, the racism that fueled the syphilis study has existed for centuries and still permeates the U.S. health care system, causing racial disparities in access to medical care and measures of health. While there are a variety of efforts to address these disparities, including medical training to bring awareness of racial biases, there is far to go.” An article by Aimee Cunningham in Science News attempt to widen the historical focus on medical racism to encompass more than just the notorious Tuskegee experiments.