788 - Revealing the Scarcity of AI/ML-Enabled Medical Devices for Children

Publication Number: 788.6477

Greg Zapotoczny, Ann & Robert H. Lurie Children's Hospital of Chicago, Cartersville, GA, United States; Ansh Goyal, UColorado Anschutz / Childrens Colorado, Aurora, CO, United States; Shahida Qazi, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Madison A. Christmas, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Juan C. Espinoza, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States

Background: When software is used for one or more medical purposes, and it performs these purposes without being part of a hardware medical device, it is regulated by the FDA as Software as a Medical Device (SaMD). Over the past 3 decades, SaMDs have incorporated artificial intelligence and machine learning (AI/ML) into their function. In August 2024, the FDA published a list of 950 AI/ML-Enabled SaMDs it had cleared or approved to date. Little is known about the development, availability, and applicability of AI/ML-enabled SaMD for children, and particularly about their testing on and clearance for use in pediatric populations.

Objective: To evaluate pediatric labeling and clinical evidence for marketed AI/ML-enabled medical devices.

Design/Methods: Using the publicly available FDA list to identify unique devices, we then extracted data from the FDA medical device database concerning the pediatric use of these devices focusing on their age labeling and clinical evidence. Descriptive statistics were used to summarize the data.

Results: The first AI/ML-enabled SaMD was cleared in 1995, but 2016 marked a dramatic increase in their entry to market (Figure 1). The majority of devices were developed in Radiology (75.9%), Cardiovascular (10.4%), and Neurology (3.57%). Among all SaMD records, 63% were silent on age in their Indications for Use statement, meaning, they do not specify the age group the device is for. Regarding clinical validation, ~45% of all SaMD records did not include any clinical validation, while ~30% was insufficient in detail to interpret age data.
Only 42 (4.5%) devices were specifically indicated for the use in children (0-17yo; Table 1). Of these, 33 (79%) presented data (10 - prospective patient recruitment study, 23 - retrospective data analysis). An additional 86 (9%) devices mentioned pediatrics in general, but only 5 (5.9%) presented clinical evidence with specific pediatrics ages; in fact 44 (51%) were devoid of any clinical data (pediatric or adult) (Table 2).

Conclusion(s): Pediatric AI/ML-enabled medical devices are few, and even fewer present pediatric clinical data to substantiate their indications for use. The rapid increase in SaMD has not met the needs of all populations equally, with children in particular being left behind. Regulatory and legislative efforts are needed to create appropriate incentives and requirements to drive pediatric-specific innovation.

Figure 1. Changes in the numbers of AI/ML-enabled devices over the years (1995-2024)


Table 1. Inclusion of Pediatric Subpopulations in the Indications for Use Statements


Table 2. Utilization of Pediatric Subpopulations for Clinical Validation