Healthcare AI today: Where’s the AI ROI?, AI messaging truth check, AI burnout busters, more

News and views you ought to know about:

• Provider organizations are using generative AI for all sorts of use cases. It’s delivering impressive returns in two. Those would be tech support and patient experience. The finding is from Google’s new survey of more than 600 senior leaders in healthcare and life sciences. The survey report shows just under 80% seeing better patient engagement metrics, 70% enjoying higher patient satisfaction scores and almost 30% riding high on patient experience boosts of more than 10% over previous checks. Meanwhile tech support is tied with patient experience at the top of use cases already showing ROI: Both were reported as doing so by 34% of respondents. 
   
  • The report also drills into the rise of agentic AI in the sector. There some 44% say their organizations are using one or more AI agents. These use cases “span a spectrum of complexity—from single-task agents to multi-agent systems that can take actions on your behalf and under your control,” write lead report authors Aashima Gupta and Shweta Maniar in their executive summary. “For the healthcare and life sciences industry, this marks a clear acceleration from planning to action.” The breadth of agentic AI adoption in the sector, they add, is “striking.” 
     
  • Unsurprisingly, the No. 1 concern on which healthcare executives want to grill LLM suppliers before buying their wares is data privacy and security. Just under half the field, 49%, reports using agentic AI to assist with security operations and cybersecurity. 
     
  • The ripest ROI opportunities for agentic AI in core operational functions present in inventory tracking and restocking, medical image interpretation and patient screening and on-demand personal care—all of which were named by 22% of respondents. “These functions—even those with longer implementation horizons—have the potential to revolutionize care delivery by directly addressing high-volume, repetitive administrative tasks,” the authors remark. “These applications are not experimental. They are already having a clear, tangible impact.” 
     
  • The full report is available in exchange for contact information (and a willingness to receive news, product updates, event information and special offers about Google Cloud). 
     
• Here’s some hard data on clinician use of LLMs for communicating with patients. Tracking the behavior across 55,000 portal messages, researchers found AI-equipped clinicians using the technology around 20% of the time. When they did, they cut the moments needed to compose by around 7%—331 seconds instead of 355. On the other hand, the time they saved went into reviewing or editing the bot-penned drafts. So the usage was a wash, at least if the main aim was saving clinicians’ time. The study was conducted at New York University and published in npj Digital Medicine. Going forward, unlocking LLM AI’s full potential will take “tailored implementation to ensure that AI tools meaningfully reduce clinician burden while enhancing care quality,” lead author Soumik Mandal, PhD, says. Journal study here, NYU’s own coverage here. 
   
• A number of AI models are designed specifically to relieve burnout in healthcare workers. Do they work? It’s complicated. Researchers in Sweden sought to peel back the hype and see how certain AI-based interventions were performing in real-world clinical settings. The AI they scrutinized had unique combinations of five components—engagement, control, journaling, psychoeducation and self-monitoring of demands on time and attention. “Our findings indicate that combining Demands and Control components yields the strongest benefits for mental health outcomes,” the authors report in JMIR Formative Research. “Similarly, the synergy between Demands and Engagement as well as between Journaling and Psychoeducation demonstrates improvements in participant health relative to other types of content.” As previously noted, it’s complicated. The study is interesting, though, and it’s available in full for free. 
   
• A lot of healthcare AI proponents suggest learning from other industries. One provider organization went beyond studying to going onsite. The org is HCA Healthcare Nashville. After reading up on how non-healthcare outfits were leveraging AI to improve safety, a cross-functional HCA team saw relevant use cases in person at a GE manufacturing plant, a DuPont Chemical facility and the U.S. Army’s 160th Special Operations Aviation Regiment. The mission of the tour was to “learn scalable, industry-agnostic safety strategies,” the American Hospital Association reports. “These learnings are now being incorporated into HCA Healthcare’s clinical operations and care delivery.” Good stuff. More here. 
   
• Where human thought moves from experience to understanding to meaning, ‘anti-intelligence’ moves from data to pattern to prediction. I hadn’t thought of AI as anti-intelligence till now, but the author of the thought has a point. Writing in Psychology Today, public intellectual John Nosta states his case in provocative—and sometimes poetic—terms. “Anti-intelligence is the glimmer of fluency mistaken for the light of understanding,” he writes. “Anti-intelligence will keep getting better at imitation, but our task is to get better at discernment.” Read the whole thing. 
   
• Microsoft is making a major push to turn nurses into Copilot aficionados. A big part of the initiative is working closely with industry partners of all sorts. “Dragon Copilot now equips nurses with an advanced suite of tailored, ambient-enabled AI capabilities that streamline documentation, surface clinical insights and automate routine tasks directly within their workflow,” Microsoft states in an announcement. “These innovations are a direct result of Microsoft’s multiyear collaboration with nurse leaders and frontline nursing staff across several healthcare organizations in the U.S. to deeply understand and support the complexities of the largest workforce in healthcare.” Get the rest. 
   
• We’ve been hearing for some time that AI might be able to scan an individual’s medical records to flag gathering health troubles long before the appearance of any signs or symptoms. The scenario is already taking shape. At the Icahn School of Medicine in New York City, researchers have designed a system called InfEHR that “links unconnected medical events over time, creating a diagnostic web that reveals hidden patterns.” The description is from the news operation at Mount Sinai. The research is published in Nature Communications. What’s more, unlike hallucination-prone AI models that sound authoritative while spewing nonsense, InfEHR “knows when to say ‘I don’t know’—a key safety feature for real-world clinical use.” Peer-reviewed study here, Mount Sinai news item here. 
   
• Worth a look if you have a minute:
   
  • Georgians say no to AI in healthcare jobs (Georgia Sun) 
     
  • Navigating the ‘AI-powered’ digital healthcare boom (TechTarget) 
     
  • America First Policy Institute launches AI team (AFPI) 
     
• Noteworthy research news:  
   
  • Counting bites with AI might one day help prevent childhood obesity (Penn State) 
     
  • AI analysis of world’s largest heart attack datasets opens way to new treatment strategies (University of Zurich)
     
  • Researchers advance orthodontics with AI-assisted growth prediction (Korea University)
     
• From AIin.Healthcare’s sibling outlets:
   
  • Cardiovascular Business: Advances in imaging could help predict, prevent heart attacks
     
  • Radiology Business: FDA clears RadNet imaging AI solution for detecting neurodegenerative disease
     
  • Cardiovascular Business: FDA clears AI-powered vascular imaging software