|
The ideas and innovators shaping health care | | | | |  | | By Daniel Payne, Erin Schumaker and Ruth Reader | | | | | 
Murphy wants a narrow approach to AI regulation that protects doctors from liability. | Joe Raedle/Getty Images | | | “Inflexible constraints and one-size-fits-all policy will entrench this industry in bureaucracy and hinder innovation.” Rep. Greg Murphy, co-chair of the GOP Doctors Caucus | | Neither Congress nor the Biden administration has moved yet to set new rules for the use of advanced artificial intelligence in health care. Even so, big regulatory questions are coming to the fore. Rep. Greg Murphy, a North Carolina Republican who co-chairs the GOP Doctors Caucus, favors rules that are “narrow in scope,” protect physicians who use the technology from liability and avoid “overlapping regulatory governance” that would make it more difficult for patients to gain access to emerging therapies. Murphy spelled out his thinking in a letter today to Food and Drug Administration Commissioner Robert Califf, who figures to be a key player in the development of new rules. The FDA authorizes some AI products under its existing rules for medical devices and is considering whether it needs additional rules for new, mostly unregulated tools, such as chatbots. Still, the agency hasn’t yet explained how narrow — or broad — it thinks new rules should be. Murphy wants to knows: — How the FDA will handle the expected increase in AI-driven medical tools — Whether the agency plans to strengthen its third-party review program to free up its workforce — Whether the FDA supports alternative pathways for approving AI devices — or liability safe harbors for devices and doctors who use them — What role outside groups, such as medical or hospital credentialing boards, should play in creating standards for using the tech What’s next? How the administration sees AI regulation should become clearer soon. In October, President Joe Biden tasked the Department of Health and Human Services, of which the FDA is part, with launching an AI task force by the end of January and developing guidelines on AI-enabled health tools by the end of April.
| | | | 
Winter Park, Colo. | Shawn Zeller/POLITICO | This is where we explore the ideas and innovators shaping health care. Caring for a patient with Covid in the hospital got a lot more expensive — 26 percent more from 2020 to 2022 — as doctors got better at doing it, according to a RAND Corporation study published in JAMA. Why? After vaccines, a greater share of the patients had complicating health factors and needed pricier treatments. Share any thoughts, news, tips and feedback with Carmen Paun at cpaun@politico.com, Daniel Payne at dpayne@politico.com, Ruth Reader at rreader@politico.com or Erin Schumaker at eschumaker@politico.com. Send tips securely through SecureDrop, Signal, Telegram or WhatsApp. | | |  Labs to test AI products could reassure doctors thinking about using the new tech, according to a Mayo Clinic researcher. | Jim Mone/AP | As health care tools driven by artificial intelligence proliferate, the U.S. needs a national network to ensure they're fair, appropriate, valid, effective and safe, authors of a new JAMA paper argue. Why it matters: A public-private network of “assurance labs” would benefit developers, who could share resources for validating their AI models and innovate faster. Labs would also help protect the public by evaluating models on factors like technical performance, useability and bias. “Such a network could fill a critical gap in an ecosystem dominated by well-meaning but often overexuberant and inexperienced developers who lack the depth of understanding of health care delivery,” the authors, who hail from Stanford Medicine, Johns Hopkins, the Mayo Clinic and other players in health AI, write. The idea aligns with President Joe Biden’s October executive order on AI, which called for developing an AI assurance policy to assess AI-enabled health care tools before and after they go to market. What’s next? Assurance labs are also a next step in blueprints from groups like the Coalition for Health AI, which is concerned with implementing trustworthy AI in health care. (Nigam Shah, lead author of the JAMA paper and a Stanford Medicine professor, is a CHAI working group member.) CHAI's blueprint recommends labs have four infrastructure components: a shared definition of values, registries of tools, legal agreements and a sandbox for testing.
| | | | |  Standard methods of diagnosing tuberculosis require expensive equipment. | PUNIT PARANJPE/AFP via Getty Images | The sound of a cough could someday be enough to diagnose tuberculosis. An artificial intelligence program called TBScreen, developed at the University of Washington, could identify 149 patients with tuberculosis and distinguish their coughs from those of 46 patients with different respiratory illnesses at a Kenyan health care facility, according to a study in Science. The National Institute of Allergy and Infectious Diseases sponsored the study. Why it matters: The most common methods for detecting TB require expensive equipment. The researchers tested multiple recording devices, including Google Pixel smartphones, which are widely available tools and relatively inexpensive. Even so: TBScreen was able to correctly assess a positive case only around 70 percent of the time, with some variation among the recording devices. The World Health Organization has said that an ideal detection method for tuberculosis should have at least 90 percent sensitivity. “There is a lot more robustness that needs to be built into it,” one of the researchers, doctoral student Manuja Sharma, told Ruth. But she thinks there are ways to boost the accuracy of the technology, such as combining the cough recording with other data, like whether a patient has a fever. What’s next? More research is underway. Sharma said she and her fellow researchers are allowing a group of patients who might have TB to record their coughs at home for analysis. “We are trying to see if we can use this model to diagnose TB at home,” she said. | | | | Follow us on Twitter | | | | Follow us | | | | |  |
