Healthcare AI newswatch: Regulatory sparring partners, AI nutrition labels, agentic AI basics, more

Artificial intelligence AI in healthcare news

Buzzworthy developments of the past few days.

  • Medical device companies navigating regulatory pathways should know: Large language models are not reliable guides. Veeva Systems found this out when it put six of the top LLMs through their paces for this use case. The cloud application vendor tested the models’ chops on three discrete tasks. These were a.) giving strategic advice on the EU AI Act, b.) retrieving and summarizing information, and c.) classifying medical devices in order to illuminate marketing or regulatory pathways. The models were Perplexity, Mistral, Gemini, OpenAI, Llama and Claude. Veeva found that, on average, LLM performance ranged between 25% and 79% when benchmarked against human experts (100%). The author of the paper describing the project, Veeva director of med-tech regulatory strategy Diogo Geraldes, offers some advice. When analyzing and applying regulatory guidance, he writes, regulatory affairs professionals “may only want to use an LLM as a ‘sparring partner.’” 
     
  • Healthcare AI ‘nutrition labels’ have taken a big step toward fruition. Officially called CHAI Applied Model Cards, the acronym standing for the Coalition for Health AI, the labels are getting their own registry. CHAI is working with Microsoft experts at Avanade to set it up. The main idea behind the cards is to let health systems evaluate AI products before buying them. One major reason for the registry is to help AI vendors standardize their means of informing customers and prospects. CHAI hopes to see more sharing of info among and between both providers and developers. At UMass Memorial Health in Massachusetts, an early adopter of the card system, associate CMIO Elisabeth Garwood, MD, says the model cards will “play a crucial role in the AI governance process by consolidating information in an easily digestible format, facilitating product comparisons and providing standardized data that meets the needs of various stakeholders.” The information to which she refers includes model training data as well as any known risks for legal exposure. CHAI announcement here, Newsweek coverage here. 
     
  • The buzz over AI agents in healthcare is getting more insistent. If you need to play a little catchup on agentic AI—the general concept as well as some of the specific capabilities and products—you could do worse than to spend a few minutes with Harvey Castro, MD, MBA. Breaking down the basics at KevinMD, Castro says AI agents will soon be essential to modern healthcare systems. “These intelligent systems are increasingly applied in medical documentation, diagnostics and patient engagement, offering substantial potential for improved patient outcomes and reduced physician burnout,” he writes. However, as with any AI for healthcare, agentic AI “must be guided by rigorous clinical evidence, ethical oversight and regulatory compliance to ensure measurable and beneficial outcomes.” Read the piece. 
     
  • Clinicians have questions. Before beginning to adopt AI tools, those at some of the most tech-forward healthcare institutions in the country wanted to know some things. “How accurate is the model? Was it trained on local data? Does it represent the kind of patients we see?” Things like that. Plus “How transparent is it? Who developed it? How’d they do it? And how up-to-date is it?” Smart cookies, those clinicians. And patients-first advocates all. Healthtech Analytics found this out after speaking with AI technology leaders at Duke Health, Mayo Clinic and Vanderbilt University Medical Center. Sometimes anecdotal examples are no less enlightening than survey results. 
     
  • AI’s big sin in healthcare billing is supposed to be denying claims. But that may be on a good day. The technology is often still failing to prevent duplicate charges, standardize invoice formats and make sure bills reach intended recipients. It’s not always great, either, at synchronizing the timing of bill sends from providers and insurers. Or at extending grace periods for even smallish delinquencies. Did Benefits Pro leave out any other basic billing fumbles AI doesn’t reliably prevent? 
     
  • Young Brits are good with the use of AI in healthcare. And they don’t worry it will be used to replace doctors. Then again, maybe they’re good with the use of AI in healthcare because they don’t worry it will be used to replace doctors. Either way, their positive disposition bodes well for the future of healthcare AI in the U.K. See the full survey results in BJR Artificial Intelligence. 
     
  • The upbeat AI outlook is echoed by an erstwhile American radiologist who now works in industry. “Do I think doctors are going to be out of a job? Not at all,” says Shez Partovi, MD, who once practiced neuroradiology at Barrow Neurological Institute in Arizona and today works as chief innovation officer at Philips. Technology “can help deliver better care for more people,” he tells The Guardian, “and it also brings the joy of medicine back to radiographers and radiologists.”
     
  • Nurses too busy minding technology to tend to patients? Maybe your hospital needs to staff up with patient technology technicians. That’s what they’re doing at UCHealth in Colorado. The tech experts filling the new roles “are on the scene whenever there is an issue with wearables, telemetry, safety view cameras, medication scanners and other devices—both at the bedside and via the system’s Virtual Health Center,” SmartBrief reports. “With their help, nurses can get back to serving patients and working at the full scope of their licensure.”
     
  • Palm scans are said to be 100 times more accurate at confirming identity than iris scans. Healthcare people are putting the claim to the test at NYU Langone Health, where the newer technology is rolling out for patients who wish to use it for checking in. Before heading in to receive care, the patient needs to scan his or her own palm using the Amazon One application. Amazon explains it all.   
     
  • Recent research in the news: 
     
  • Notable FDA approval activity:
     
  • Funding news of note:
     
  • From AIin.Healthcare’s news partners:
     

 

Dave Pearson
Dave Pearson

Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.

Related Content

HIMSS survey report: Beware insider threats to cybersecurity involving AI
Cleerly and Bunkerhill Health unite in the name of AI-based cardiac imaging evaluations
Healthcare AI newswatch: Surgical symphonies, integrated intelligence, generalist AI, more
Primary care is primed for AI. But is AI aligned with primary care?
5 steps for evaluating radiology AI applications
FDA clears Google smartwatch tech that calls 911 when a user’s heart stops