Radiology AI medical devices have led in development and clearances, but they need a more dedicated regulatory approach for pediatric indications, according to a cross-sectional study published March 20 in JAMA Network Open.
Researchers not only analyzed AI-enabled devices marketed in the U.S. for pediatric indications but also exposed clinical area gaps and laid out a practical policy agenda for the U.S. Food and Drug Administration (FDA).
Overall, pediatrics-focused AI technologies remain scarce amid an ambiguous regulatory environment, according to lead author Grzegorz Zapotoczny, PhD, of Stanley Manne Children’s Research Institute in Chicago, and colleagues.
“The FDA does not have distinct statutory evidentiary standards for pediatric devices; rather, review teams have latitude to make case-by-case benefit-risk determinations while adhering to least burdensome principles,” they wrote.
Using publicly available data, the group examined 952 regulatory submissions filed between 1995 and 2024. The availability of AI for pediatric patients could not be readily ascertained from any database, given the lack of pediatric-specific information gathered and published during the regulatory process, they noted.
The analysis revealed that only five exclusively pediatric AI technologies were introduced between 2020 and 2024, with the first being Medo Aria, which is designed to help radiologists evaluate and report hip ultrasound images.
Radiology comprised 723 of all devices, 18 of which were specifically labeled for pediatrics. The group counted 14 cardiovascular pediatric devices and 13 neurology pediatric devices, according to the findings.
Unique challenges
The study highlights the unique challenges of adapting adult AI technologies to pediatric patients, including off-label use of adult devices in children, access to robust pediatric datasets, persistent technical and clinical challenges, and even inconsistencies with the definition of “pediatric” across FDA centers such as the Center for Devices and Radiological Health and the Center for Drug Evaluation and Research.
“Given that medical device regulations do not offer age-specific labeling standards, the interpretation of the intended patient population remains challenging,” the authors wrote. “These patterns, along with inconsistent publicly available data, limit clinician ability to assess pediatric applicability and may slow or inhibit access to AI-supported care for pediatric patients.”
Ultimately, the group proposed a practical policy agenda for the FDA that includes the following:
- Standardize age labeling and validation requirements for AI-enabled technologies.
- Require or strongly encourage pediatric performance reporting, or at a minimum, a strong justification of why it is not necessary when devices are marketed for all ages.
- Consider pediatric-focused predetermined change control plans that allow iterative model updates with targeted postmarket commitments to support labeling expansion when premarket pediatric data are limited.
- Invest in shared pediatric datasets through federal and consortium mechanisms that support privacy-preserving linkage and access to lower development costs.
- Address financial and operational barriers to observational studies for faster dataset creation.
- Harmonize pediatric definitions across FDA centers to reduce ambiguity for sponsors and reviewers.
“In this study, pediatric devices were rare, emerged recently, and had longer review times and a higher proportion of registered clinical trials compared with nonpediatric devices, suggesting expectations for more pediatric-specific evidence despite unchanged statutory standards,” Zapotoczny and colleagues concluded.
A critical need exists to improve and adapt clinical trial infrastructure to the increasing demands for robust training data, they added. This could include establishing pediatric medical device clinical trial units at children’s hospitals through the FDA Pediatric Device Consortia Program or a nationwide clinical trials network, such as one proposed through the Pediatric Medical Device Public-Private Partnership.
Radiology is crucial
Pediatric-specific device labeling is especially pressing in radiology, where more than 75% of FDA-authorized AI devices are used, stated R. Brandon Hunter, MD, a pediatric critical care physician at Texas Children’s Hospital in Houston, TX, with others in an invited commentary.
“Ultimately, clinicians need clear information about which AI tools are appropriate for pediatric use and a regulatory environment that makes developing those tools worthwhile,” Hunter and colleagues said. “[Zapotoczny et al’s] analysis should serve as a call to action for all stakeholders invested in pediatric health in the age of AI.”
Read the complete paper here.