Days before leaving office, President Trump’s top health official is proposing to permanently exempt from regulatory review a wide swath of artificial intelligence products used to help physicians detect serious conditions such as cancer and respiratory disease on medical images.
The proposal, posted to the Federal Register on Friday and signed by Health and Human Services Secretary Alex Azar, asserts that FDA review of several categories products, including AI imaging devices, is not necessary because they are seldom associated with reports of adverse events.
But notice of the proposed policy change raised alarms among health technology experts who said careful review of these products is essential to ensuring they meet basic performance standards before they are adopted in clinical practice.
“It seems like very slippery slope territory to remove regulation,” said Karandeep Singh, a physician and professor at University of Michigan Medical School who researches the use of artificial intelligence in health care. He said adverse events in such products can be difficult to detect and are seldom reported, because they are used to help guide clinicians, rather than directly diagnose or treat a patient.
“With these kinds of devices, you’re not necessarily going to experience the side effect right away,” he said. “It might take time to uncover issues with these products, so the absence of an adverse event being reported isn’t justification to completely waive all requirements.”
A senior HHS official said the administration was deliberative in its approach and sought to create regulatory flexibility for developers of devices that stand to improve care. The official said the proposal allows for public comment and review of the exemptions for the AI devices and other products within that same risk category.
In total, the proposal would eliminate FDA review of 91 products for which the normal regulatory process was waived during the public health emergency. It immediately lifts regulatory requirements for seven products — all different kinds of gloves — and calls for a 60-day comment period on plans to permanently exempt other types of products for review.
That means the incoming Biden administration would need to pursue and approve the policy for it to take effect for those devices. It is unclear whether it will do so.
In addition to personal protective equipment, the list of devices that would be exempt includes ventilators and digital devices to manage the safety of drug infusions, monitor fetal heart rates, and deliver behavioral therapy to patients with psychiatric disorders. About a dozen of the products perform tasks using artificial intelligence, including computer aided detection software to help identify lesions suspicious for cancer, respiratory illnesses, bone fractures, and other medical problems.
Those devices are not considered high-risk in the FDA’s regulatory scheme because similar products have already been approved and are being used in health care. The agency’s review process, known as 510k, requires that new products in those categories demonstrate “substantial equivalence” to already-approved devices, to provide some assurance of their performance capabilities, effectiveness, and safety.
However, the type of computer-aided detection software the proposal seeks to exempt has been associated with uneven performance in crucial areas such as breast imaging. Initially approved by the FDA in 1998, the software became widely adopted and was approved for increased reimbursement by the Centers for Medicare and Medicaid Services. But subsequent research found the software, the use of which increased costs by about $400 million a year, did not result in improved cancer detection and at some hospitals, even led to poorer performance.
The proposal to entirely eliminate review prior to allowing products to enter the market also seems at odds with an FDA action plan published earlier this week that specifically calls for more careful review of products that rely on artificial intelligence. The document proposes to establish standards for the training, evaluation, interpretability, and documentation of AI devices and calls for transparency in labeling and the development of methods to audit AI systems for bias. It also seeks to establish a review framework in which products would be examined prior to commercial release and monitored once deployed in clinical practice.
“This framework would enable FDA to provide a reasonable assurance of safety and effectiveness while embracing the iterative improvement power of artificial intelligence and machine learning-based software as a medical device,” the document states.
A spokesperson for the FDA declined to comment.
The document proposing the exemptions does not specifically raise the topic of artificial intelligence, but notes that all of the products listed were associated with few adverse events. It also states that requirements for FDA review of such products often costs developers millions of dollars during an often years-long process.
“The exemptions provided for and proposed under this notice for these 91 device classes could eliminate anywhere from $9.1 to $364 million in startup costs if there were one new entrant into each device market,” the proposal states “Savings could further accrue based on each new market entrant. Instead of being costs passed along to patients and taxpayers, these savings could be invested in other areas such as research and development and manufacturing.”