Beyond the label – Closing the gap in AI medical devices for children
Artificial intelligence algorithms that meet the definition of a medical device, including AI-enabled devices and Software as a Medical Device (SaMD), are regulated by the FDA in the United States and are increasingly prevalent in clinical care. Off-label use of medical devices and drugs already is widespread in pediatrics, where the majority of these products lack specific FDA authorization for use in children. AI-enabled medical devices now represent the latest and fastest-growing front of that same problem.
FDA authorization is the process by which the agency evaluates whether a medical device has sufficient evidence of safety and effectiveness to be marketed in the U.S. Without appropriate regulatory oversight, these algorithms may not perform reliably for pediatric patients. As one notable example, an adult-developed AI algorithm designed to evaluate thyroid nodules achieved a specificity of just 36% in pediatric patients, generating false positives nearly two-thirds of the time.
A recently published study in JAMA Network Open provides additional insight into this problem. The authors analyzed all AI-enabled medical devices authorized by the FDA through June 2024 and found that only 42 (4.4%) carried specific pediatric age labeling, and just five exclusively pediatric AI devices have been authorized. Nearly 60% of all devices included no patient age information at all, leaving clinicians with no way to know whether these tools were ever intended for children.
We were invited to write a commentary on these findings, and this question was frequently raised: What do we do about a system where the applicability and safety profile of AI devices for pediatric care are unknown? The intuitive answer is transparency. We support requiring manufacturers to state whether their device is intended for children and whether pediatric patients were included in validation. But transparency mandates carry a risk that is easy to overlook. Pediatric device development already is constrained by small market size, developmental physiological differences and ethical limitations on clinical trials.
Most of these barriers are ultimately financial in nature. And the data suggests there may be a regulatory penalty as well, with pediatric devices facing significantly longer FDA review times and higher evidentiary requirements than non-pediatric devices. If companies are required to declare whether their device is intended for pediatric use without any reduction in the cost of achieving it, the rational business decision is to print “not indicated for pediatric use” on the label and move on. What currently is passive omission of children could become active exclusion.
To prevent that outcome, transparency must be paired with support. That means streamlined validation pathways, acceptance of real-world evidence and smaller sample sizes, and incentive programs modeled on what already exists for orphan drugs and humanitarian use devices. As the leaders of the Southwest-Midwest National Pediatric Device Innovation Consortium (SWPDC), one of five FDA-supported consortia dedicated to advancing pediatric device development, we have spent years working to lower these barriers. To date, we have supported more than 250 device innovators across 31 states with regulatory, engineering, clinical and business development support as well as early stage non-dilutive funding, and those companies have gone on to raise more than $327 million in follow-on funding.
Children deserve AI tools that were designed and developed for them. Improved pediatric-specific testing and labeling are part of the answer, but they are not enough on their own. What we need is a regulatory and economic environment that realigns incentives and makes pediatrics an attractive space for innovators, not one that they avoid.
By Dr. R. Brandon Hunter, assistant professor of pediatrics, Baylor College of Medicine; Dr. Kolaleh Eskandanian, senior research scientist, Medstar-Georgetown AI Colab; Dr. Chester J. Koh, professor of urology, Baylor College of Medicine
