The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.
April 28, 2023
In today’s Duke AI Health Friday Roundup: study reveals bumpy path to integrating AI into clinical care; revisiting questions about credit for DNA discovery; asking ChatGPT to help with your research; graphene “tattoo” pacemaker tested in rats; model uses EHR data to predict hospitalization risks for kids with complicated health issues; transplanting pancreatic islet cells to control diabetes; testing a video explainer for patients who can benefit from ICDs; much more:
AI, STATISTICS & DATA SCIENCE
- “The challenges uncovered by the project point to a dawning realization about AI’s use in health care: building the algorithm is the easiest part of the work. The real difficulty lies in figuring out how to incorporate the technology into the daily routines of doctors and nurses, and the complicated care-delivery and technical systems that surround them.” STAT News’ Casey Ross reports on a study conducted by a team including researchers from the Duke Institute for Health Innovation that explores the on-the-ground reality of AI tools deployed in clinical settings.
- “Addressing bias in AI requires a holistic approach, involving diverse and representative datasets, enhanced transparency and accountability in AI systems, and the exploration of alternative AI paradigms that prioritize fairness and ethical considerations.” A review article by Emilio Ferrara, currently undergoing open peer review and available as a preprint from the Journal of Medical Internet Research, offers a look at the topics of AI fairness and bias as addressed in the literature.
- “…we perform an extensive evaluation of SAM’s ability to segment medical images on a collection of 11 medical imaging datasets from various modalities and anatomies. In our experiments, we generated point prompts using a standard method that simulates interactive segmentation…Performance appears to be high for tasks including well-circumscribed objects with unambiguous prompts and poorer in many other scenarios such as segmentation of tumors.” A new preprint by Mazurowski and colleagues, available from arXiv, provides an evaluation of the Segment Anything Model for image analysis across multiple applications and domains.
- “A novel EHR data-based predictive model was developed and validated as a CDS tool to identify CCHN at high risk for future hospitalization. The overall favorable performance of the predictive model as a CDS rule highlighted potential opportunities for an EHR data–based risk modeling approach to help inform future population-level clinical initiatives.” A research article by Ming and colleagues, published in the journal Hospital Pediatrics, presents results from a study that examined the performance of a predictive model that uses data from electronic health records to identify children with complex health needs who are at high risk for future hospitalization.
- Here comes the flood: shortly after Peter Gabriel announced a competition in partnership with generative AI company Stable Diffusion, the iconic musician began receiving internet blowback over potential infringement of copyright of artists whose work was used to train Stable Diffusion’s image generator.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “The need for immunosuppressive drugs is one major roadblock to using pancreatic islet transplantation to treat diabetes. Hu et al. used CRISPR to knock out the genes encoding class I and II MHC and overexpress CD47 in primary human pancreatic islet cells, making them immune-evasive. The hypoimmune cells were reaggregated into pseudoislets and transplanted into diabetic humanized mice. The grafted cells escaped immune-mediated killing and were able to control blood glucose without immunosuppressants.” In a research paper by Hu and colleagues, published in Science Translational Medicine, the authors describe experiments in allogeneic humanized mice to evaluate whether pancreatic islet cells could be transplanted to control diabetes without rejection or autoimmune reactions by the host.
- “Dr. Franklin’s early death also meant she missed out on the Nobel Prize, but the Nobel Assembly could have found other ways to acknowledge her contribution, said Nils Hansson, a historian of medicine at Heinrich Heine University Düsseldorf, in Germany. Neither Dr. Watson nor Dr. Crick mentioned her when they accepted their awards, Dr. Hansson noted, although Dr. Wilkins, who also received the prize, did.” An essay by Emily Anthes, published in the New York Times, makes the case that Rosalind Franklin – whose x-ray crystallography was critical to understanding the structure of DNA – deserves a greater share of credit for the discovery than she has hitherto been afforded.
- “The tattoos sandwich a transparent layer of graphene between sheets of stretchy silicone and ultrathin polymer. Gold tape connects the graphene to wires that run to a power source, which sends electricity through the device…Future versions of the tattoo will be wireless, Efimov says, perhaps using a tiny antenna to pick up electrical signals from an external device placed on a person’s chest.” Science News’ Meghan Rosen reports on a new study that tested a revolutionary pacemaker design made of ultra-thin graphene in rats.
- “Approximately 60% of the participants assented to ICD implantation, and there was no difference between the video and usual care groups…Compared with usual care, the decision support video increased patient knowledge and decreased the time clinicians spent with participants. A racially concordant video that featured Black patients and a Black clinician did not increase assent to ICD implantation compared with a racially discordant video that included White patients and a White clinician.” In a research article published in the Annals of Internal Medicine, Thomas and colleagues report findings from a randomized clinical trial that evaluated whether a video decision-support tools, including one designed specifically for Black patients who were candidates for receiving an implantable cardioverter/defibrillator would prove effective in reducing rates of refusal of the procedure.
COMMUNICATION, Health Equity & Policy
- “Despite the current hype surrounding tools like ChatGPT and GPT-4, further development of artificial intelligence capabilities is clearly required before one can embrace their use as reasonably trustworthy research tools. For now, the effort required is hardly worth the outcome. Human oversight clearly remains crucial in interpreting any results from the GPT family of large language models.” An essay at Scholarly Kitchen by Richard De Grijs documents a somewhat worrisome evolution of responses when the author attempted to use ChatGPT as a research assistant.
- “As the system flips to Gold – and thus to Green-via-Gold, will it still be useful to continue to invest in the cost of Green depositing? It is true that the repositories provide an agency-wide record of funded research outputs and serve as the basis for various search and discovery tools. Building these databases will only become easier with the Nelson memo mandates related to metadata and persistent identifiers, which will strengthen linking from the agency databases to the published scholarship. But, is capturing and maintaining files that duplicate the publisher-hosted files necessary to achieve these ends?” An essay at Scholarly Kitchen by Lisa Janicke Hinchliffe examines whether the likelihood of increased movement toward open access publishing prompted the OSTP Nelson Memo may have implications for “green” open-access publishing that includes online resources and databases maintained by academic centers.
- “Members of the Sackler family who were among the most heavily involved in running Purdue Pharma made their first donations to the National Academies in 2008, when Dr. Raymond Sackler, and his wife, Beverly Sackler, and the couple’s foundation, started contributing, according to Academy treasurer reports.” The New York Times’ Christina Jewett reports on news that the Sackler family – owners of OxyContin makers Purdue Pharma – were donating millions of dollars to the National Academies of Science, Engineering & Medicine during the opioid crisis.
- “Decentralised clinical trials involve many digital tools to facilitate research without physical contact between research teams and participants at various stages, such as recruitment, enrolment, informed consent, administering study interventions, obtaining patient-reported outcome measures, and safety monitoring. These tools can provide ways of ensuring participants’ safety and research integrity, while sometimes reducing participant burden and trial cost….The US Food and Drug Administration and other regulators worldwide have issued guidance on how to implement such adaptations. However, there has been little focus on the distinct ethical challenges these trials pose.” A perspective article published in Lancet Digital Health by Vayena and colleagues explores the ethical implications of decentralized clinical trials.