Context:

Medical AI devices for children remain scarce and concentrated in a few specialties, with pediatric-angled reviews taking longer and more often citing trial registrations, signaling cautious regulatory progress and uneven labeling practices; experts urge pediatric-specific performance evaluation when devices are marketed for all ages, especially in radiology where most AI devices are used. The study tracks devices reviewed by the FDA over a multi-decade span to reveal persistent gaps between adult-focused development and pediatric needs, implying a cautious near-term outlook for broader pediatric AI adoption.

Dive Deeper:

• The researchers analyzed 952 FDA entries for AI-enabled medical devices from 1995 to 2024, of which 948 included indications for use in children and teens. Among these, only 42 devices (4.4%) were labeled for pediatric patients, indicating a small pediatric share despite a long time horizon.

• The first pediatric-cleared AI device emerged in 2015, and between 2020 and 2024 five devices were exclusively for pediatrics, highlighting a gradual but limited growth in child-specific approvals.

• Age was omitted in 59.6% of indication statements, and when present, labeling formats varied widely. Pediatric labeling appeared across 8 of 18 clinical areas, with radiology (42.9%) and neurology (31%) being the most represented, while 10 clinical areas had no pediatric devices.

• Pediatric devices had longer FDA review times (median 162 days vs 134 days for non-pediatric devices) and were more likely to list registered clinical trials (14.3% vs 2.2%), suggesting higher regulatory and evidentiary demands.

• Experts advocate requiring pediatric performance reporting or strong justification when devices target all ages, emphasizing that algorithms trained on adult data may not generalize to children; concerns are especially acute in radiology where a majority of AI devices are used.

• The study, led by Grzegorz Zapotoczny, PhD, from Ann & Robert H. Lurie Children’s Hospital of Chicago, was published online March 20 in JAMA Network Open, and acknowledges limitations due to incomplete public FDA data and inconsistent age/trial reporting.

• Disclosures note support from the Health Resources and Services Administration and FDA, with one author reporting personal fees from industry; editors used AI tools as part of the drafting process and human review ensured accuracy.