Pediatric AI devices remain rare, face longer FDA review

Pediatric artificial intelligence (AI) devices are rare and have longer review times vs nonpediatric devices, according to a recent study published in JAMA Network Open.¹

Many publications have evaluated the application of AI in health care, highlighting the growing interest in this topic. The FDA regulates these devices under a risk-based framework from class 1 to class 3, though some class 1 and class 2 devices are exempt from premarket review.²

“AI use in medical devices presents unique regulatory challenges; for example, if an algorithm continues to learn and change over time, it could diverge from the initial, FDA-reviewed version, raising concerns about safety and continued effectiveness,” wrote investigators.¹

Identifying pediatric AI devices

The cross-sectional study was conducted to evaluate the extent of pediatric evaluation and indications in FDA-regulated AI-enabled devices. Publicly available data through November 2024 for 952 AI devices was obtained from the FDA’s AI-enabled medical devices database, 4 of which were excluded from the analysis.

Indications for use statements were extracted using a custom natural language processing program. Output was generated by an OpenAI program then human validated.

Zapotoczny et al criteria were used to analyze age labeling practices, allowing investigators to identify which devices are indicated for pediatric patients. Pediatric was defined as aged 0 to 17 years and young adult as aged 18 to 21 years.

Approval trends and application review times

The first FDA approval for an AI device was identified as occurring in 1995. Only 32 devices were approved across the following 20 years, but a record number of approvals has been reported every year since 2016. This culminated in a total of 952 devices, correlating with the increasing trend of AI sponsors from only 23 between 1995 and 2016 to 372 as of June 2024.

Of FDA marketing applications for AI devices, 97.2% were 510(k)s. These devices had a median review time of 134 days, which was significantly faster than other modes of application.

De novo applications had a median review time of 264 days and premarket approvals (PMAs) a median review time of 349 days. Significant variance in the review time was reported across 510(k) applications, ranging from 7 to 722 days.

Common clinical domains and applicant market share

Investigators also reported device groups, including 75.9% of devices being developed for use in radiology. The most common product codes for these devices were related to radiological image processing, x-ray tomography, and ultrasonic pulsed doppler imaging. Cardiovascular, neurology, hematology, and urology were also common domains.

Having only 1 device on the market was reported for 61.3% of applicants and fewer than 5 devices for 90.1%. The top 6 companies accounted for 24.5% of devices.

Use statements with age descriptors provided insights into the patient populations AI devices were developed for, though 59.6% of device for use statements were silent on age. Of those with age information, 58.5% described broad age ranges using descriptive language, while 134 followed a discrete number with a phrase and only 22 used discrete numbers.

Labeling findings and conclusion

Only 12.9% of devices for patients younger than 22 years followed FDA-recommended pediatric subpopulations age ranges. Labels were indicated for young adults in 63.8%, while 4.4% were labeled for pediatric patients.

Radiology, neurology, and cardiovascular were the most common areas represented by 42.9%, 31%, and 9.5% of devices, respectively. A median review period of 162 days was reported for pediatric devices vs 134 days for nonpediatric, non-PMA devices. Overall, the data highlighted pediatric labeling among FDA-marketed AI-enabled devices as rare, recent, and focused on a few clinical areas.

“Policy changes addressing device labeling, evidence requirements, and pediatric-specific considerations could reduce friction, alongside investments in pediatric clinical research infrastructure and dataset development,” wrote investigators.

References

1. Zapotoczny G, Goyal A, Christmas M, Qazi S, Carroll M, Espinoza J. FDA-regulated AI-enabled medical devices with pediatric indications. JAMA Netw Open. 2026;9(3):e262636. doi:10.1001/jamanetworkopen.2026.2636
2. Espinoza J, Shah P, Nagendra G, Bar-Cohen Y, Richmond F. Pediatric medical device development and regulation: current state, barriers, and opportunities. Pediatrics. 2022;149(5):e2021053390. doi:10.1542/peds.2021-053390