Oncology Nursing Society

See attached

October 16, 2026

Kyle Diamantas, JD
Acting Commissioner of Food and Drugs
Food and Drug Administration
Department of Health and Human Services
Submitted electronically via Regulations.gov

Re: Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and
Request for Feedback

Dear Acting Commissioner Diamantas,

Oncology nurses are privileged to support people with cancer through their entire cancer experience, from
diagnosis and treatment into survivorship or as they face their end-of-life journey. Today, the Oncology
Nursing Society (ONS) provides responses to key questions addressed in the aforementioned discussion
paper. Across these questions, ONS encourages FDA to ensure that evaluation and oversight approaches
reflect the realities of oncology care, where decisions frequently involve complex symptom management,
high-risk treatment toxicities, and vulnerable patient populations. AI systems should augment oncology
nursing expertise and clinical judgment, support safe and equitable care, and preserve appropriate human
oversight, particularly when delayed intervention or inappropriate recommendations could result in serious
patient harm.

Question 3: CDRH seeks input on whether and when a GenAI-enabled function that results in delivery of
clinical information to patients, as opposed to HCPs, could present different or higher risks, while also
recognizing the potential benefits associated with improved patient empowerment, engagement, and
access to clinical information. What device characteristics, output features, or safeguards might mitigate
risks that could arise when a user lacks the domain knowledge to independently evaluate an output,
without unnecessarily underestimating patient capability?
ONS agrees that patient-facing GenAI-enabled functions may present distinct safety considerations compared
with healthcare professional-facing functions, particularly in oncology care. Cancer patients and caregivers
are increasingly likely to use AI-enabled tools to seek information about symptoms, treatment side effects,
medication questions, survivorship concerns, and supportive care needs. While these tools may improve
access to information and support patient engagement, patients often lack the clinical knowledge necessary
to independently evaluate the accuracy, completeness, or appropriateness of AI-generated
recommendations.

ONS is concerned that the consequences of incorrect or incomplete information in oncology can be
substantial. For example, failure to recognize symptoms such as fever, infection, immunotherapy-related
adverse events, uncontrolled pain, dehydration, or other treatment toxicities may result in delayed
intervention and serious patient harm. For this reason, ONS encourages FDA to consider specialty-specific
risk factors when evaluating patient-facing clinical functions.

ONS recommends that patient-facing GenAI-enabled devices include safeguards such as transparent
communication of uncertainty, clearly defined limits of intended use, evidence-based symptom escalation
protocols, and mechanisms that direct patients to seek professional evaluation when symptoms indicate
potential urgency. In addition, evaluation frameworks should assess performance across varying levels of
health literacy, digital literacy, language proficiency, and patient populations that have historically
experienced disparities in cancer care.

To mitigate these risks without underestimating patient capability, manufacturers should evaluate how
devices respond when patients omit important clinical context, ask leading questions, or express emotional
distress. Safeguards should include appropriate follow-up questions, recognition of insufficient information,
and referral to the cancer care team when an individualized answer cannot be safely provided. User
instructions should help patients understand what information to provide and how to interpret the response,
while complementing safeguards built into the device. Evaluation should also assess whether outputs
appropriately consider the individual’s cultural context, spiritual needs, and preferences when relevant to the
intended use. A technically accurate answer may still be inappropriate for the patient’s circumstances or fail
to support quality of life.

Question 6: For GenAI-enabled care escalation functions, CDRH is considering how the evaluation may
account for both under-escalation and over-escalation. How could manufacturers characterize and weigh
these two directions of error, given that they may not be commensurable and that acceptable trade-offs
may vary by clinical context?
ONS strongly supports FDA's consideration of both under-escalation and over-escalation when evaluating
GenAI-enabled care escalation functions. In oncology, symptom triage and escalation decisions are central
patient safety activities and are frequently managed by oncology nurses and nurse practitioners with
specialized training and/or certification.

ONS believes under-escalation can have severe or life-threatening consequences, particularly when oncologic
emergencies initially present with subtle symptoms but progress rapidly. ONS notes that failure to
appropriately direct patients to urgent evaluation for febrile neutropenia or sepsis, immune-related adverse
events, Superior Vena Cava Syndrome, spinal cord compression, or other oncology emergencies can lead to
significant morbidity or mortality. Conversely, persistent over-escalation may increase patient anxiety,
contribute to unnecessary emergency department utilization, increase healthcare costs, and place additional
strain on already limited healthcare resources.

ONS recommends that FDA evaluate escalation functions using clinical scenarios that reflect real-world
oncology practice, including complex symptom presentations, comorbidities, evolving clinical conditions,
novel treatments, and the social and structural factors that influence symptom recognition and access to
care. Evaluation should account for clinically meaningful variation in how symptoms and treatment-related
adverse effects may present based on sex and across diverse racial and ethnic populations, without treating
these characteristics as biological proxies. Further, because GenAI systems trained on incomplete,
unrepresentative, or historically biased data may perpetuate or amplify inequities – and may fail to recognize
or integrate these crucial differences – FDA should require representative datasets, transparent reporting of
data limitations, subgroup performance analyses, and ongoing monitoring for disparate under-escalation or
over-escalation.

ONS also recommends that oncology nurses and other relevant members of the cancer care delivery team
should be meaningfully involved in the design, validation, implementation, and oversight of these functions.
Risk assessment should place particular emphasis on conditions where delayed intervention is associated
with serious patient harm, and acceptable performance thresholds may need to vary by clinical context,
disease severity, and patient population. These recommendations are consistent with the principles outlined
in the ONS position statement, Artificial Intelligence in Oncology Nursing, including the need for equitable AI
systems, meaningful nursing involvement, transparency, and continued human oversight. These systems
should augment, rather than replace, clinical judgment and should maintain clear pathways for human
review and intervention.
Specifically, manufacturers should characterize under-escalation and over-escalation separately, including
the severity of potential harm, time to appropriate escalation, and whether the recommended level of care is
clinically warranted. These errors should not be combined into an overall accuracy measure that could
obscure missed emergencies. For time-sensitive oncology emergencies, acceptance criteria should give
greater weight to avoiding missed or delayed escalation; for lower-acuity presentations, criteria should also
assess unnecessary escalation. Oncology nurses and other relevant oncology specialists should help define
and justify these thresholds using evidence-based standards of care.

Question 14: For open-ended device outputs, how should performance comparators and acceptance
criteria be selected? When a panel of qualified clinicians serves as the comparator, how should the
applicable standard (for example, the standard of care versus the performance of a median clinician in
practice) be defined and justified? Should generalist or specialist physicians be used as a performance
standard? When should human-AI team performance, rather than the device operating alone, serve as the
basis for evaluation?
ONS appreciates FDA's recognition that traditional performance standards may not be sufficient for
evaluating open-ended GenAI outputs. However, ONS is concerned that comparisons limited to physician
performance may not adequately reflect how many healthcare AI systems are implemented and used in
practice.

Many GenAI-enabled functions are likely to support symptom management, patient education, care
coordination, navigation, survivorship care, clinical trial participation, and triage workflows routinely led by
nurses and nurse practitioners. Accordingly, nurses should help develop evaluation frameworks, including
performance comparators, acceptance criteria, and measures of clinical utility, to ensure these reflect
nursing expertise. Therefore, ONS recommends that FDA consider multidisciplinary comparator models that
include nurses and other relevant members of the cancer care delivery team based on the device's intended
use. For oncology applications specifically, comparisons against specialty oncology clinicians may be more
clinically meaningful than comparisons against generalist practitioners. Additionally, FDA should evaluate
both device-alone performance and human-AI team performance, recognizing that many clinical use cases
involve AI augmenting, rather than replacing, professional judgment.

The applicable standard should be defined and justified using evidence-based clinical guidelines and the
standard of care for the function being evaluated, with input from qualified oncology nurses and other
relevant oncology specialists. Median clinician performance alone should not justify acceptance of responses
below an established safety standard. Human-AI team performance should serve as the basis for assessing
clinical utility when the intended use depends on a professional reviewing and acting on device outputs,
while device-alone testing remains important for identifying errors and limitations. Such evaluation should
reflect the patient population and workflow in which the device will be used. Local validation can help
determine whether performance holds for the specific community and care team, and should complement
manufacturer validation rather than transfer that responsibility to healthcare institutions.

Question 21: What roles might clinicians, healthcare institutions, professional societies, standards-setting
bodies, and other stakeholders appropriately play in postmarket monitoring, and how can these roles be
structured without diffusing manufacturer accountability?
ONS agrees that effective postmarket monitoring of GenAI-enabled devices requires participation from
multiple stakeholders, with clearly defined roles that preserve manufacturer accountability. Professional
societies’ standards of practice and evidence-based clinical guidelines should inform the evaluation and
updating of these devices.

As the professional organization representing oncology nurses, ONS believes professional societies can
contribute by:
• Identifying specialty-specific safety concerns and use cases.
 • Developing clinical best practices that inform postmarket evaluation.
• Educating clinicians on safe and effective AI utilization.
• Supporting awareness and reporting of emerging performance concerns.
• Contributing specialty expertise to benchmarking frameworks, performance measures, and
postmarket evaluation activities.
• Helping assess the impact of GenAI tools on care quality, patient experience, workforce
implications, and health equity.

Oncology nurses and other members of the cancer care delivery team can identify and report safety concerns
observed in practice, while healthcare institutions can assess performance in their local patient populations
and workflows. To ensure these concerns are addressed, ONS encourages FDA to establish clear reporting
pathways and expectations for manufacturers to investigate reported concerns, communicate findings and
limitations, and take timely corrective action. In doing so, FDA should make clear that manufacturers remain
responsible for the ongoing safety and effectiveness of their devices.

Question 26: Are there additional considerations that inform the premarket and postmarket evaluation of
agentic GenAI-enabled devices, beyond those applicable to non-agentic GenAI-enabled devices? How
should the elevated risk associated with autonomous multi-step action, tool use, and reduced opportunity
for human review be reflected in acceptance criteria and oversight?
ONS agrees that agentic AI systems warrant additional scrutiny because of their ability to autonomously plan,
sequence, and execute actions with reduced opportunities for direct human review, which may introduce
risks beyond those associated with traditional decision-support functions.

Within oncology care, future agentic systems may be used to support care coordination, symptom triage,
patient outreach, treatment scheduling, care navigation, documentation, and other operational or clinical
workflows. As the degree of autonomy increases, so does the potential impact of errors, inappropriate
actions, or failures to recognize changing clinical circumstances. As such, ONS recommends that FDA require
heightened evidence expectations for agentic systems whose outputs or actions may influence clinical
decision-making, treatment pathways, or patient access to care. Evaluation should assess not only the
accuracy of individual outputs but also the safety of multi-step workflows, performance under unexpected
conditions, handling of uncertainty, maintenance of human oversight, and protection against automation
bias. Acceptance criteria should require human review before irreversible or high-consequence actions and
reliable mechanisms to stop the system or transfer control when needed. In addition, agentic systems should
maintain transparent audit trails that allow members of the cancer care delivery team and healthcare
institutions to understand, evaluate, and review actions taken by the system.

Finally, given the cybersecurity risks associated with agents that access patient records or connected tools,
ONS urges FDA to clarify that manufacturers are responsible for effective safeguards to prevent
unauthorized access or actions, detect malicious instructions or abnormal behavior, and stop unsafe
activity. As part of its evaluation, FDA should assess whether these safeguards withstand attempts to
manipulate the agent, limit data access and permitted actions, and allow it to handle tool failures safely.
Before deployment, healthcare institutions’ clinical, information technology, and cybersecurity teams should
be involved in reviewing these safeguards and establishing oversight and response procedures. Once
deployed, manufacturers should monitor the safeguards and promptly address failures.

***

We appreciate your consideration of these important issues as FDA considers approaches to the regulation of
GenAI-enabled medical devices. If you have any questions or would like to discuss our responses in more
detail, please contact us at healthpolicy@ons.org.
 Sincerely,

The Oncology Nursing Society

The Oncology Nursing Society (ONS) is a professional association that represents more than 100,000 nurses
and is the professional home to more than 35,000 members. ONS is committed to promoting excellence in
oncology nursing and the transformation of cancer care. Since 1975, ONS has provided a professional
community for oncology nurses, developed evidence-based education programs and treatment information,
and advocated for patient care, all in an effort to improve the quality of life and outcomes for patients with
cancer and their families.