The U.S. Department of Health and Human Services on Thursday announced new funding to help improve the use and maintenance of artificial intelligence-powered medical devices.
Research has shown that machine learning models used in clinical settings can degrade over time. The new HHS money, earmarked for the Advanced Research Projects Agency for Health, or ARPA-H, will be used to innovate efforts to make AI tools more reliable for doctors and more beneficial for patients.
That initiative is called the Performance and Reliability Evaluation for Continuous Improvement and Usability of Artificial Intelligence, or PRECISE-AI.
WHY IT MATTERS
According to ARPA-H, more than 950 medical devices with integrated AI functionality have been cleared by the US Food and Drug Administration — a tenfold increase from 2018.
While these new AI-powered tools can transform a physician’s ability to deliver care, ML can degrade over time due to changes in clinical input, data collection, patient populations, or IT infrastructure.
“The promise of AI-powered tools for healthcare is only as powerful as the relevant real-world data that informs them,” Berkman Sahiner, director of the PRECISE-AI program, said in a statement.
Current techniques challenge users to monitor and maintain AI-driven device performance in real-world clinical settings.
“The inability to automatically monitor and maintain the performance of AI-enabled medical devices based on real-world activity poses risks to clinicians and patients,” Sahiner added.
HHS said that through PRECISE-AI, it wants to establish a repository of open-source tools that sustain AI clinical decision support by automatically identifying and remediating performance impairments without human oversight.
Root cause analysis tools also need to communicate clear and actionable information about the source of deterioration to enable healthcare users to better interpret model uncertainty.
By “proposing optimal approaches to detect and mitigate performance degradation of the underlying AI model,” clinicians can provide better patient care, Sahiner said.
The agency said the upcoming final call will focus on five technical areas, and teams will start by developing tools to establish the best estimate of a patient’s diagnosis based on available evidence, or “ground truth.”
Teams will then develop automated methods to monitor the AI engine’s performance against real-world data, identify the root cause of any declines, and make necessary adjustments.
They will also create underlying data infrastructure and mechanisms to notify clinicians, AI tool developers, hospital administrators, and regulators when performance is impaired.
The response date for the current solicitation is January 15, 2025.
THE BIGGER TREND
As an independent organization of the National Institutes of Health, ARPA-H seeks to invest in ways to build stronger, healthier, and more resilient health care systems.
The project focuses on creating scalable platforms and strategies from advanced mobile hospitals providing acute care in rural areas to securing open source software used in critical infrastructure.
As part of the Digital Health Security Initiative, DIGIHEALS, ARPA-H has partnered with the Defense Advanced Research Projects Agency on the Artificial Intelligence Cybersecurity Challenge, and in May, HHS announced $50 million in funding under ARPA-H to help vendors patch ransomware vulnerabilities on medical networks and devices.
ON THE RECORD
“PRECISE-AI addresses a growing gap in ensuring that AI tools used in clinical decision-making are accurate, safe, and robust in real-world settings,” ARPA-H Director Renee Wegrzyn said in the call announcement. “In doing so, ARPA-H is creating a foundation of trust between clinicians and these AI tools, which will further expand the potential of AI to improve health outcomes for all Americans.”
Andrea Fox is senior editor of Healthcare IT News.
Email: [email protected]
Healthcare IT News is a publication of HIMSS Media.
The HIMSS Healthcare AI Forum is scheduled for September 5-6 in Boston. Learn more and register.
