NYU’s Customized ChatGPT; JAMA’s AI Goals; AI Projects to Ease Clinician Workload

Welcome to the first edition of MedAI Roundup, highlighting the latest news and research in healthcare-related artificial intelligence each month.

NYU Langone launched a customized version of OpenAI’s ChatGPT, then gave it to a team of employees to experiment with during an AI “prompt-a-thon.” (STAT News)

JAMA published an editorial outlining the journal’s goals for publishing research on the use of AI in healthcare, which will focus on essential areas including clinical care and outcomes, patient-centered care, healthcare quality, fairness in AI algorithms, medical education and clinician experience, and global solutions.

In the New England Journal of Medicine, researchers called for viewing biased clinical data as “artifacts“ in order to identify “values, practices, and patterns of inequity” in healthcare. This approach can provide new ways to use AI to detect inequities, they argued.

An AI model was more accurate than available alternatives at predicting incident adenocarcinoma of the esophagus or gastric cardia in Veterans Health Administration patients, according to a study published in Gastroenterology.

On the other hand, two studies in the Annals of Internal Medicine — one large randomized controlled trial, and one systematic review and meta-analysis — showed that incorporating AI computer-aided detection systems (CADe) into colonoscopy doesn’t improve rates of advanced neoplasia detection.

ChatGPT also failed to impress when it was prompted to recommend treatment regimens for cancer patients based on National Comprehensive Cancer Network guidelines. About a third of the AI chatbot’s responses did not fully align with the guidelines, according to a research letter in JAMA Oncology.

In the U.K., “smart” pants called NeuroSkin helped a stroke patient walk independently again by stimulating her paralyzed leg with electrodes controlled by an AI algorithm. (BBC)

HCA Healthcare will collaborate with Google Cloud to use generative AI technology to take on the most time-consuming tasks for physicians and nurses, including clinical documentation.

Abridge, a leader in generative AI for clinical documentation, will partner with Epic to implement a generative AI feature for note taking at Emory Health in Georgia.

FDA granted a “De Novo” approval to Viz.ai for its new AI algorithm, the Viz HCM Module, for detecting hypertrophic cardiomyopathy. The company said the approval creates a “new regulatory category” for cardiovascular AI programs.

Please enable JavaScript to view the

comments powered by Disqus.