AI newswatch: Lifecyle AI oversight, data matters, AI vs. animals, more

Buzzworthy developments of the past few days.

• Dear Makers of AI-Equipped Medical Devices: Please include a detailed plan for monitoring your products over the long haul as soon as you seek initial market approval. That was the gist in January when the FDA published draft guidance spelling out the particulars of the not-so-gentle request. The agency’s aim was, and is, cutting chances of recalls over time while supporting the agency’s ongoing evaluation of AI risk controls. The period for stakeholders to help revise the document—“Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations”—ended April 7. Covering the submitted comments, which numbered 46, Regulatory Focus spotlights input from two industry groups. “Many of the recommendations in the AI Lifecycle draft guidance are not relevant for less complex AI-machine learning (ML) models,” AdvaMed wrote. “The scope of the document should be narrowed to address more complex AI,” chimed in MDMA, “in accordance with FDA’s least burdensome principles and to ensure a risk-based approach.” AdvaMed’s full name is the Advanced Medical Technology Association. MDMA is the Medical Device Manufacturers Association. Get RF’s coverage of the pair’s respective takes on the draft guidance here. 
   
• It can be easy for outside observers to forget that doing healthcare AI right isn’t just about seeing to ‘soft’ concerns. Such as, for example, varying impacts across socioeconomic strata. It’s also about cold, hard data. An official at the Council for Affordable Quality Healthcare is speaking up to remind us. “AI can’t work well if it’s built on inconsistent or messy data,” CAQH’s chief policy and research officer Erin Weber tells the American Journal of Managed Care. “We need common formats, rules and processes to ensure the underlying data are usable across different systems.”
   
• Not that healthcare AI worries are ever either/or propositions. In fact, both/and tends to hold sway for all sorts of factors. The above are only two of these. Also in the mix are things like medical ethics, data security, patient privacy and potential communications spoilers. These and more are covered in a TechTarget article looking at seven challenges to AI integration in healthcare and their remedies. “You have to have data in a shape and form that AI can consume,” one subject matter expert points out for TT reporter John Moore, echoing the point made by CAQH’s Weber above. “Otherwise it will be junk in, junk out.” 
   
• A lot of people are loving the FDA’s plan to replace lab animals with AI algorithms. Count Elizabeth Baker, Esq., among them. The development will “usher in a new era for drug testing that will save human and animal lives by integrating better science to make better decisions for health,” says Baker, the director of research policy for the Physicians Committee for Responsible Medicine. “The announcement reflects that innovation, public health and animal protection go hand in hand.”
   
• AI may finally end the paper chase caused by prior authorization. Or it may worsen it. “Though innovators promise speed and better access, doctors say that insurers could easily use the technology to make approvals and appeals even more taxing,” explains healthcare journalist Donavyn Coffee in Medscape. If AI isn’t used to help make the prior-auth process more transparent and patient-centric, the technology will “simply make a flawed system work faster.” Read the rest. 
   
• Providers’ investments in healthcare AI have ‘surged past’ EHRs and digitalization plans. So notes KLAS Research in a report on global trends in healthcare IT released this week. Many organizations have “moved beyond utilizing standard business-intelligence capabilities into driving clinical and administrative efficiencies through AI,” the report’s authors write. “While interest in investment is high, respondents in most regions report that  AI adoption is still early and their first focus is to test solutions and refine governance structures.” 
   
• Individuals with serious vision impairments may soon be able to ditch their white canes in favor of high-tech systems combining goggles, cameras, earphones and AI. The breakthrough was made in China and is described in a study published April 14 by Nature Machine Intelligence. In a news item summarizing the research, Nature.com quotes one of the co-authors. “This system can partially replace the eyes,” says Leilei Gu of Shanghai Jiao Tong University. The system can detect and identify obstacles from a ways off, Gu explains. By comparison, a walking stick “can only touch the environment. It cannot know what the object is.” So: No question. If brought to market, the high-tech alternative will definitely represent an upgrade over the current standard. 
   
• A global pharma consultancy says it has developed an AI algorithm that can accurately forecast revenues for more than 90% of U.S. drug launches. That would handily best the predictive power of human Wall Street analysts making consensus projections, says the firm, Trinity Life Sciences. The details are in a new white paper, which Trinity is offering in exchange for contact information. 
   
• Recent research in the news:
   
  • University of Mississippi: Researchers Use Machine Learning to Predict Exercise Adherence
     
  • Stony Brook U: From Python to bioinformatics and deep learning: Preparing the next generation of AI-ready healthcare innovators
     
  • Stanford: AI in science and medicine: A deep dive from the AI Index Report
     
  • University of Delaware: AI-powered model predicts post-concussion injury risk in college athletes
     
  • University of Pennsylvania: Researchers use deep learning to pinpoint potential disease-causing variants in non-coding regions of the human genome
     
• Funding news of note: 
   
  • Assort Health secures $26M to expand specialty-specific generative AI platform for managing patient phone calls
     
  • Assort Health raises $22M Series A 
     
  • Brellium raises $16.7M to power AI-driven compliance for healthcare
     
  • Trellis Health emerges from stealth with $1.8M pre-seed funding to provide women with unrivaled access to health data
     
  • Healthcare leads AI adoption with record investments
     
• From AIin.Healthcare’s news partners:
   
  • Cardiovascular Business: Cardiology, radiology professionals unite to explore the challenges of diagnosing CAD in women
     
  • Health Imaging: AI bests radiologists at computed tomography angiography reads
     
  • Radiology Business: RadNet leaders discuss rationale behind blockbuster $103M iCAD AI acquisition
     
  • Health Imaging: AI is most beneficial as a prescreening tool, new research suggests