Discrepancies in Marketing, Clearance of AI-Enabled Medical Devices Found
Researchers have identified discrepancies between the 510(k) clearance and corresponding marketing materials of 12.6 percent of the FDA’s AI- or ML-enabled devices.
Source: Getty Images
By - In a recent study published in JAMA Network Open, researchers found discrepancies between the marketing and 510(k) clearance of artificial intelligence (AI)- or machine learning (ML)-enabled medical devices, including the revelation of some devices being marketed as having capabilities not approved by the United States Food and Drug Administration (FDA).
The study sought to investigate whether medical devices marketed as enabled for AI and ML were being appropriately approved for those capabilities in their FDA 510(k) clearances. A 510(k) clearance serves as a premarket submission to the FDA demonstrating that a device is equivalent to an already existing, legally marketed device in terms of safety and effectiveness.
These clearances play a key role in the regulation of medical devices, but a lack of uniform FDA guidelines to regulate AI and ML-enabled devices can lead to discrepancies between a device’s marketing and its approved indications for use, the researchers explained.
To explore potential discrepancies, the research team manually surveyed 510(k) approval summaries and accompanying marketing materials for medical devices approved by the FDA between November 2021 and March 2022.
A total of 119 devices were evaluated, with a focus on the prevalence of discrepancies between the marketing and certification material for each device. Devices were categorized as adherent, contentious, and discrepant based on the content of these materials.
The analysis found that 15 devices (12.61 percent) were discrepant, eight devices (6.72 percent) were contentious, and 96 devices (84.03 percent) were consistent between their marketing and FDA 510(k) clearance summaries.
Further, there was a statistically significant difference between the three categories across radiological and cardiovascular devices.
Of the 75 radiological devices, 62 were considered adherent, three were contentious, and ten were discrepant. For the 23 cardiovascular devices, 19 were adherent, two were contentious, and two were discrepant.
In the end, one-fifth of the devices surveyed were found to have discrepancies between their marketing materials and clearance documentation.
Concerns over the use of AI and ML in medical devices have grown in recent years, and the FDA has struggled with how to regulate these devices.
In April, new FDA draft guidance was released, outlining an approach to help ensure that AI- and ML-enabled medical devices can be safely and rapidly modified in response to new data. The approach aims to support continuous incremental improvements in ML-enabled device software functions and make sure that device users are aware of necessary information around planned modifications.
The guidance builds on earlier proposed regulatory frameworks for Software as a Medical Device (SaMD) modifications, including recommendations that some AI-powered clinical decision support tools be regulated as medical devices.
The FDA’s work in this area continues as criticisms from researchers and clinicians argue that current regulatory approaches are insufficient.
