The Consortium for Technology & Innovation in Pediatrics (CTIP) (Juan Espinoza)

Please find attached the full comments of the Consortium for Technology & Innovation in Pediatrics (CTIP), an FDA-funded pediatric medical device accelerator, on Docket FDA-2026-N-7874. Our recommendations support rigorous, risk-proportionate evaluation of GenAI-enabled medical devices and feasible development pathways that protect children while addressing longstanding gaps in pediatric device availability.

October 16, 2026

Michelle Tarver, MD, PhD
Director
Center for Devices and Radiological Health
Food and Drug Administration
10903 New Hampshire Ave
Silver Spring, MD 20993

RE: Docket No. FDA-2026-N-7874

Dear CDRH Leadership:

On behalf of the Consortium for Technology & Innovation in Pediatrics (CTIP), we appreciate the opportunity to
comment on FDA’s discussion paper on generative artificial intelligence (GenAI)-enabled medical devices. CTIP is a
pediatric medical device accelerator led by Ann & Robert H. Lurie Children’s Hospital of Chicago. Established in 2011
and first funded through FDA’s Pediatric Device Consortia (PDC) Program in 2013, CTIP advances the development,
commercialization, and clinical adoption of innovative medical technologies for children. Our collaborative model
brings together clinicians, engineers, regulatory experts, hospital administrators, patients, caregivers,
entrepreneurs, and industry partners throughout the innovation process.

Through a national network of 26 partner organizations across eight states, CTIP provides portfolio companies with
technical assistance, connections, funding support, and expertise spanning engineering and product testing,
intellectual property, regulatory strategy, clinical trial design, and reimbursement. Our perspective reflects this
direct experience helping companies bring pediatric technologies to market and a body of peer-reviewed research
to which our investigators and collaborators have contributed on pediatric device availability, clinical evidence,
labeling, development barriers, and AI-enabled technologies. This work informs our understanding of how
regulatory expectations affect both the evidence supporting a device and the feasibility of developing it for children.

CTIP supports FDA’s exploration of a risk-based approach that combines competency-based evaluation,
proportionate clinical confirmation, and ongoing monitoring. Pediatric considerations should be integrated
throughout this approach, with evidence and safeguards that reflect the intended clinical use, developmental
population, users, and care setting. Pediatric inclusion should not, by itself, trigger a higher risk category or a fully
separate evidence program. Additional requirements should address scientifically justified uncertainties and
potential harms through predictable, achievable evaluation pathways.

This balance is essential. Small markets, limited study populations, and fragmented evaluation infrastructure
already constrain pediatric innovation. Additional requirements that are costly, duplicative, or difficult to anticipate
may encourage developers to exclude children. FDA has an opportunity to support both safety and access through
appropriate reuse of evidence, clinically meaningful comparators, shared pediatric evaluation resources, and
feasible approaches to postmarket learning. A successful framework should protect children while making it
possible to develop and sustain the devices they need. The comments that follow identify areas of alignment with
FDA’s discussion paper, concerns about pediatric access and implementation, and specific recommendations for a
workable pediatric development pathway. We welcome continued engagement with FDA and other stakeholders to
advance these goals.
1. Areas of Alignment with FDA’s Proposed Approach

A. Risk assessment grounded in clinical use
CTIP supports assessing risk according to what a device does, who uses it, and the consequences of an
incorrect output. The complete deployed device—including its interface, prompts, retrieval sources,
connected tools, and safeguards—should be the focus of evaluation. Performance of the underlying
foundation model alone cannot establish safety and effectiveness in pediatric care. Regulatory
expectations should follow clinically meaningful differences in function and risk, rather than architectural
novelty alone. [1]

The value of this approach depends on how faithfully it reflects actual use. A designation such as
“informational” or “clinician-supervised” should not, by itself, establish that a function presents lower risk.
The assessment must consider foreseeable reliance on the output and whether the intended user can
recognize an error and intervene before harm occurs. For pediatric devices, these considerations vary
across children, caregivers, and clinicians and across home, outpatient, and hospital settings. FDA should
provide practical examples that make these distinctions predictable for developers.

B. Competency assessment and proportionate clinical confirmation
CTIP supports competency-based benchmarking followed by clinical confirmation proportionate to the
device’s intended use, risk, and remaining uncertainty. FDA’s recognition that confirmation may include
retrospective evaluation, shadow deployment (where the device operates in a live clinical workflow but its
outputs are withheld from clinicians and patients and do not affect care), and prospective clinical studies
is particularly important for pediatric development. These alternatives should be established as credible,
clearly described evaluation pathways. Requiring a prospective trial by default, without a specific
evidentiary justification, could make otherwise valuable pediatric applications infeasible. [1,2,5]

When a device is intended for children, relevant pediatric expertise should inform benchmark design,
reference standards, and adjudication of clinically consequential outputs. Adult clinical expertise alone
should not be presumed sufficient to judge pediatric performance. Evaluation should address
age-specific clinical reasoning, communication with children and caregivers, recognition of uncertainty,
and appropriate deferral and escalation. Both missed escalation and unnecessary escalation have
consequences that must be assessed in the intended care setting.

FDA should distinguish the expertise needed to judge an output from the comparator used to assess
clinical performance. Pediatric specialists may be essential to establishing a valid reference standard even
when the intended application supports a generalist or caregiver. Clear expectations for these choices,
together with reusable evaluation methods, would strengthen pediatric assessment and help developers
plan evidence generation with greater confidence.
C. Lifecycle oversight informed by existing frameworks
CTIP supports FDA’s emphasis on evaluation and monitoring throughout the product lifecycle. Existing
clinical AI governance approaches, including those described by Bedoya, van de Sande, and Bartels and
colleagues, along with the National Institute of Standards and Technology (NIST) AI Risk Management
Framework and World Health Organization (WHO) guidance, provide relevant foundations for staged
evaluation, multidisciplinary oversight, and ongoing quality management. [6–10] FDA should build on
these approaches and clarify how evidence generated through them can support regulatory review and
postmarket obligations.

The central principles are evaluation in the intended care environment, explicit criteria for progression to
clinical use, and monitoring tied to defined responsibilities and corrective action. Oversight must assess
whether safeguards work in practice, whether users can recognize and respond to errors, and whether
performance remains acceptable as models, workflows, and patient populations change. For pediatric
applications, this includes detecting clinically meaningful failures that may be obscured by satisfactory
aggregate performance.

Implementing these principles requires clinical expertise, access to data, technical infrastructure, and
sustained resources. FDA should not assume that each developer or adopting institution can
independently assemble and maintain these capabilities. That assumption is particularly consequential
for pediatric innovators, whose smaller markets limit their ability to absorb duplicated evaluation and
monitoring costs. FDA should explicitly identify opportunities to reuse evidence, align manufacturer and
institutional evaluation, and support shared resources while preserving manufacturer accountability. A
workable lifecycle framework must make rigorous oversight achievable for the companies developing
pediatric devices and the institutions caring for children.

2. Concerns about pediatric access and implementation

A. Pediatric inclusion must remain feasible
Pediatric device development is constrained by small markets, limited study populations, and
developmental differences that complicate evidence generation. For companies deciding whether to
pursue a pediatric indication, added cost, time, and regulatory uncertainty can determine whether
children are included at all. When developers cannot sustain a pediatric development program, children
may continue to lack access to devices evaluated for their clinical and developmental needs. FDA should
therefore consider how evidence expectations affect the feasibility of pursuing pediatric indications, while
maintaining rigorous standards for safety and effectiveness. Predictable, scientifically justified
requirements can support both adequate evaluation and pediatric access. [2,3]

The gap is already apparent in AI-enabled devices. An analysis of 952 FDA submissions reviewed between
November 1995 and June 2024 found that only 42 (4.4%) included pediatric age ranges of 0–17 years, and
only five technologies were exclusively pediatric. Pediatric submissions also had longer median review
times and more frequent references to registered clinical trials. These associations do not establish their
causes, but they reinforce the importance of predictable pediatric evidence expectations. [4]

FDA should explicitly state that pediatric inclusion alone does not warrant a higher risk category or a fully
separate evidence program. Additional evidence should address a scientifically justified concern about
the proposed use, developmental differences, or potential harm. Sponsors should be able to reuse
applicable evidence and address remaining pediatric questions through targeted evaluation.
Proportionate options could include testing on a limited set of pediatric safety-critical scenarios,
safeguards that recognize a pediatric patient and decline or flag the output, and labeling that states that
pediatric performance has not been established. These measures would protect children in current
practice while preserving a clear distinction between an adult device with foreseeable pediatric exposure
and a device seeking a pediatric indication. Clear expectations established early in development would
help companies plan studies, secure financing, and retain pediatric indications.

B. Assessing safeguard effectiveness in the intended setting
FDA should assess the risk reduction attributed to a safeguard based on evidence of its effectiveness for
the intended user, clinical function, and care setting, with the evaluation proportionate to the potential
harm.

Pediatric-specific evaluation and proportionate evidence requirements are compatible. Additional
evaluation is warranted when developmental differences, user capabilities, or the consequences of error
create clinically meaningful uncertainty. Pediatric status alone should not determine the risk category or
require duplication of otherwise applicable evidence. Relevant pediatric expertise should help identify the
questions that need to be answered and the most appropriate methods for answering them. This
distinction is essential to addressing pediatric safety needs without reinforcing the barriers that
contribute to gaps in device development and availability. [2,3]

FDA raises appropriate questions about reversibility, downstream safeguards, and user expertise. CTIP’s
concern is how those factors will affect the evidence expected of developers. A safeguard should reduce
assessed risk only when there is evidence that it can function in the intended setting. The presence of a
clinician in the workflow does not, by itself, establish that a consequential error will be recognized or
corrected before harm occurs. [1]

For example, recognizing an unsafe antiarrhythmic recommendation for a child may require pediatric
cardiology expertise that the intended clinician user does not have. A caregiver may be unable to
recognize deterioration in a young or nonverbal child soon enough to act. In both cases, nominal human
oversight may offer less protection than the device’s workflow suggests. Evaluation should address who
can detect the relevant error, how quickly harm could occur, and whether an effective intervention is
available.

FDA should translate these considerations into practical evaluation examples, including realistic scenario
testing of intended users’ ability to recognize and respond to consequential errors. Testing should also
address foreseeable reliance on informational outputs and multi-turn conversations that become
increasingly directive. Expectations should reflect the specific user, clinical function, and care setting;
these concerns should not become a blanket presumption that every pediatric application is high risk.

C. Evidence requirements must reflect pediatric populations and available
care
FDA should provide scientifically justified, risk-proportionate evidence pathways that account for limited
pediatric populations while maintaining standards for safety and effectiveness.

In a study of 482 pediatric device trials initiated during 2017–2022, 67.6% were single-center; among
completed studies, 51.6% enrolled fewer than 50 participants. A framework built around consistently
large pediatric datasets would be poorly matched to this development environment. Rigorous evaluation
will require coordinated evidence generation and methods suited to limited populations. [5]

FDA should describe when multisite pooling, prespecified small-sample methods, and justified borrowing
from adult or adjacent-age populations are acceptable. The justification should address developmental
differences and the uncertainty that remains, with direct pediatric evidence where transferability cannot
be established. Aggregate performance must not obscure clinically important failures in a small age
group. Shared evaluation resources, discussed in Section 3.B, would help make these expectations
achievable without requiring each developer to build the same infrastructure independently.

Clinical comparators should reflect the care reasonably available to the intended population so that
evaluation measures clinically meaningful benefits and risks. Pediatric specialists may be needed to
establish a valid reference standard even when the relevant comparator is a generalist-supported
workflow. Depending on intended use, justified comparators may include usual care, delayed specialist
review, or the care provided when no suitable pediatric device is available. Sponsors should prespecify
clinically justified acceptance criteria addressing consequential errors, such as unsafe treatment
recommendations, missed urgent escalation, or harmful delays, as applicable to the device. Favorable
performance relative to a constrained alternative should not, by itself, establish safety and effectiveness
for the proposed use. [1]

Synthetic data can help test uncommon scenarios and explore failure modes, but cannot by themselves
resolve gaps in knowledge about real pediatric performance. FDA should clarify acceptable uses,
including expectations for developmental and physiologic validity, independence from training data, and
validation against real pediatric observations. Synthetic cases should not be treated as independent
patients. The amount and type of real clinical evidence should follow the intended use, potential harm,
and uncertainty that synthetic testing cannot resolve.

D. A feasible balance between premarket evidence and postmarket learning
FDA should identify when staged evidence generation can support pediatric and rare-condition
indications, balancing the potential consequences of delayed access against the risks of unresolved
uncertainty. Premarket evidence should provide an adequate basis for the proposed initial use. Defined
residual questions may be addressed after authorization when a feasible monitoring program can detect
clinically important problems in time to act and generate the additional evidence needed. Greater
premarket evidence is warranted when errors could cause rapid, irreversible harm or when failures
cannot reliably be detected and addressed during use. [1]

A commitment to collect evidence later is credible only if the necessary patients, data access, expertise,
and resources will be available. Small pediatric populations may require pooled surveillance across
institutions to identify failures within a useful timeframe. Plans should specify the questions to be
answered, accountable parties, resources, review intervals, and thresholds for corrective action, including
restriction or suspension of use. FDA should make clear how the feasibility of these plans will be assessed
while preserving manufacturer accountability. Postmarket learning should be a workable evidence
pathway, with obligations that sponsors and clinical partners can actually carry out. [1,5]

3. Additional priorities for pediatric development and
implementation

A. Evaluation that reflects the child–caregiver–clinician relationship
Pediatric evaluation must account for the fact that the patient, person providing information, recipient of
an output, and person acting on it may be different people. A device communicating directly with an
adolescent presents a different interaction from one interpreting a caregiver’s description of an infant’s
symptoms or supporting a clinician. FDA should make these distinctions explicit and expect evaluation to
reflect the intended relationships, including transitions between users during an exchange. [1]

Expectations for child-facing functions should follow developmental capability and the clinical function
being performed. Communication and human-factors evaluation should address whether intended users
can understand the output, recognize its limitations, and take an appropriate next step. Children and
caregivers should inform this evaluation where appropriate, with attention to language, accessibility,
adolescent privacy, and caregiver access. Research on home devices for children with medical complexity
reinforces the importance of evaluating technology in the settings and relationships in which families use
it. [11]

FDA should also promote clear, consistent pediatric labeling that identifies evaluated ages, users, and
settings. Labeling should distinguish limitations in available evidence from evidence that use in a
population is inappropriate, while clearly defining the authorized intended use. Those distinctions matter
for clinical decisions and access; ambiguous pediatric labeling can leave clinicians and families unable to
determine what has actually been evaluated. [12]
B. Shared infrastructure to make pediatric evaluation feasible
Requiring each pediatric developer to independently assemble benchmarks, specialist adjudication, and
clinical evaluation networks would duplicate costs that small markets may not support. Shared evaluation
infrastructure should be a central part of implementing a competency-based approach. The PDC
Research Collaborative and the System of Hospitals for Innovation in Pediatrics – Medical Devices
(SHIP-MD) framework identify infrastructure and coordination needs that provide a foundation for this
work. Pediatric institutions, PDCs, and existing research networks could contribute expertise and access
without requiring a new evaluation system to be built from the ground up. [13,14]

CTIP recommends that FDA work with pediatric institutions, professional societies, families, developers,
and other partners to establish a pathway for reusable evaluation resources and clarify how evidence
generated through them could support regulatory review. Priorities include validated pediatric
benchmarks, protected test sets, independent adjudication, and access to clinical evaluation sites.
Developers should know what evidence can be reused and what device-specific evaluation remains
necessary. Resource development must include plans for maintenance, protection against
contamination, and affordable access.

Independent evaluation can strengthen confidence in evidence, but a mandatory or exclusive third-party
gatekeeper could become another barrier to pediatric innovation. FDA should preserve scientifically
justified alternatives and work with partners to address funding and access for small companies. Shared
resources should have transparent qualification criteria and conflict-of-interest protections, including
when participating organizations also support developers. Their value should be measured by improved
evidence quality and reduced duplication, not by adding another procedural requirement.

C. Making clinical evaluation and monitoring operationally achievable
Shadow deployment can provide a valuable route to clinical confirmation, but withholding outputs from
care does not eliminate the need for technical integration, data agreements, clinical expertise, and staff
time. FDA should clarify when this approach can provide sufficient evidence for an intended use, what
questions it cannot resolve, and what would justify progression to clinician-visible use. Clear expectations
would help developers and institutions determine whether a proposed evaluation is both informative and
feasible. [1,6]

FDA should work with clinical partners to support reusable evaluation protocols and reporting
approaches that reduce repeated negotiation across sites. These should address separation from clinical
care, access to reference outcomes, and documentation of performance, while helping sponsors and
sites determine applicable investigational-device, human-subjects, and institutional review requirements.
A company should not have to reconstruct the same evaluation process at each pediatric institution.

Postmarket monitoring likewise depends on practical arrangements for data access, clinical review, and
action. FDA should clarify the respective responsibilities of manufacturers, clinical sites, and shared
service providers, including how clinically meaningful local failures are communicated and addressed.
These arrangements should preserve manufacturer accountability and support coordinated surveillance
across small pediatric populations. Machine-based supervisory agents may assist with this work, but
reliance on them should be supported by evaluation of their reliability and shared failure modes with the
device being monitored. [1,6–9]

D. Proportionate change control and access to upstream information
Reassessment after a modification should follow its likely clinical consequences and permit reuse of
evidence for unaffected functions. FDA should provide practical examples distinguishing changes
manageable within a quality system, those requiring targeted reassessment, those suitable for an
authorized Predetermined Change Control Plan (PCCP), and those requiring a new submission. For
expansion to pediatric populations, sponsors need a predictable regulatory route and clear expectations
for the additional evidence required. Uncertainty about change control should not discourage otherwise
supportable pediatric indications. [1]

Manufacturer accountability must be supported by access to safety-relevant information from
foundation-model suppliers. Small device companies may have limited ability to negotiate advance notice
of changes, continued access to a validated model version, or information needed to assess an update.
FDA should describe workable mechanisms for version identification, change notification, safety
information sharing, and version pinning or rollback where feasible. Voluntary Foundation Model Master
Files could reduce duplicated requests, but the approach must also address situations in which suppliers
do not participate. Upstream information should support, rather than replace, evaluation of the complete
device for its pediatric use. [1]

For devices with pediatric indications, change control should include a defined set of pediatric
safety-critical test cases (for example, weight-based dosing, age-specific reference ranges, and red-flag
escalation scenarios) rerun after each model modification and after any upstream foundation-model
change against prespecified acceptance criteria. An update could potentially leave aggregate
performance unchanged while altering behavior in an uncommon pediatric scenario, where a fixed
regression set might be the most direct way to detect this. This would also give a PCCP a stable reference
point when future GenAI modifications cannot be fully specified in advance. Shared public test datasets,
as discussed above in Section 3.B, would keep this expectation achievable for small developers.

E. Oversight proportionate to agentic actions and their consequences
Agentic functions should be evaluated according to the actions they can take, their clinical consequences,
and the opportunity to prevent or reverse harm. An administrative action and an action that changes
care should not automatically receive the same oversight requirements. FDA should tie expectations for
authorization, human confirmation, interruption, and recovery to the specific action and its risk, including
consequences that emerge across a sequence of steps. In pediatric care, evaluation must establish who is
authorized and able to approve or interrupt an action—the child, caregiver, or clinician—and whether
those controls work in the intended workflow. [1,10]
Conclusion
CTIP appreciates the opportunity to provide these comments and welcomes continued collaboration with
FDA to ensure that GenAI-enabled medical devices are safe and effective for children. We encourage the
Agency to pair rigorous evaluation with feasible pathways to market so that regulatory barriers do not
perpetuate the longstanding gap in pediatric-specific innovation.

Sincerely,

Juan Espinoza, MD
Executive Director, The Consortium for Technology & Innovation in Pediatrics (CTIP)
Chief Research Informatics Officer
Stanley Manne Children’s Research Institute
Ann & Robert H. Lurie Children’s Hospital of Chicago
Associate Director, Center for Biomedical Informatics and Data Science​
Northwestern University Feinberg School of Medicine

References
1.​ US Food and Drug Administration. Considerations for the Regulation of Generative AI-Enabled
Medical Devices. Discussion Paper and Request for Feedback. 2026. Docket FDA-2026-N-7874.
https://www.fda.gov/media/194242/download
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
3.​ Bourgeois FT, Espinoza JC. Advancing equity in medical device development for children. JAMA
Pediatr. 2023;177(6):561–562. doi:10.1001/jamapediatrics.2023.0790
4.​ 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
5.​ Quazi S, Narang C, Espinoza JC, Bourgeois FT. Characteristics and results of pediatric medical
device studies: 2017–2022. Pediatrics. 2023;152(3):e2022059842. doi:10.1542/peds.2022-059842
6.​ Bedoya AD, Economou-Zavlanos NJ, Goldstein BA, et al. A framework for the oversight and local
deployment of safe and high-quality prediction models. J Am Med Inform Assoc.
2022;29(9):1631–1636. doi:10.1093/jamia/ocac078
7.​ van de Sande D, Van Genderen ME, Smit JM, et al. Developing, implementing and governing
artificial intelligence in medicine: a step-by-step approach to prevent an artificial intelligence
winter. BMJ Health Care Inform. 2022;29:e100495. doi:10.1136/bmjhci-2021-100495
8.​ Bartels R, Dudink J, Haitjema S, Oberski D, van ’t Veen A. A perspective on a quality management
system for AI/ML-based clinical decision support in hospital care. Front Digit Health.
2022;4:942588. doi:10.3389/fdgth.2022.942588
9.​ National Institute of Standards and Technology. Artificial Intelligence Risk Management
Framework (AI RMF 1.0). NIST AI 100-1. January 2023. doi:10.6028/NIST.AI.100-1
10.​World Health Organization. Ethics and Governance of Artificial Intelligence for Health: Guidance on
Large Multi-Modal Models. World Health Organization; 2024. ISBN 978-92-4-008475-9.
11.​Walsh P, Seewald S, Castillo L, et al. Medical device design flaws for children with medical
complexity at home. Pediatrics. 2026;157(3):e2025073091. doi:10.1542/peds.2025-073091
12.​Zapotoczny G, Snitman A, Shah P, et al. Describing inconsistencies in pediatric labeling of medical
devices. JAMA Netw Open. 2025;8(12):e2546517. doi:10.1001/jamanetworkopen.2025.46517
13.​Espinoza JC, Zapotoczny G, Eskandanian K, et al; Pediatric Device Consortia Research
Collaborative. Improving the infrastructure for pediatric medical device trials at children’s
hospitals. Pediatrics. 2025;155(4):e2024066974. doi:10.1542/peds.2024-066974
14.​Rosenthal GL, Peiris V, Torjusen E, et al; SHIP-MD Consortium. A national strategic framework for
developing medical devices for children. Pediatr Res. Published online 2025.
doi:10.1038/s41390-025-04548-2

www.ctipmedtech.org