Episodes of Care Solutions, Ltd. (Frank Opelka)

October 16, 2026
To: U.S. Food and Drug Administration
Re: Docket No. FDA-2026-N-7874 — Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback
Dear FDA Colleagues:
Thank you for inviting public comment on the regulation of generative AI-enabled medical devices. I am pleased to submit the attached comments in my capacity as Principal of Episodes of Care Solutions, Ltd. (EOCS, Ltd.).
FDA’s discussion paper presents an important opportunity to rethink how clinical intelligence is evaluated as its capabilities extend beyond discrete software products into clinical decisions, care pathways, and increasingly coordinated actions. I commend the agency for emphasizing competency assessment, clinical confirmation, and postmarket monitoring.
My accompanying comment, “Toward a Clinical Intelligence Assurance Framework,” proposes an additional perspective: clinical intelligence should be assessed not only for technical competency and safety, but also for its performance within actual care delivery, the effectiveness of human supervision, and its contribution to meaningful patient outcomes.
Three accompanying figures illustrate this proposal: a One HHS Map depicting opportunities for interagency coordination; a Clinical Intelligence Assurance Framework spanning evaluation, deployment, surveillance, and improvement; and a colorectal cancer screening and follow-up pathway showing how multiple intelligence functions influence care over time.
I particularly encourage FDA to consider how its existing responsibilities can connect with complementary authorities and capabilities elsewhere within HHS. The objective is not to expand FDA jurisdiction by implication, but to encourage a more coordinated approach to assuring intelligence-enabled care while respecting existing statutory boundaries.
Thank you for considering these recommendations. I would welcome opportunities for further professional discussion as FDA and its partners continue this important work.
Respectfully submitted,
F. George Opelka, MD, FACS
Principal
Episodes of Care Solutions, Ltd. (EOCS, Ltd.)
www.eocs.ltd

EOCS, LTD. / INDEPENDENT PUBLIC COMMENT / FDA-2026-N-7874

October 16, 2026

To: U.S. Food and Drug Administration​
Re: Docket No. FDA-2026-N-7874 — Considerations for the Regulation of Generative AI-Enabled
Medical Devices: Discussion Paper and Request for Feedback

Dear FDA Colleagues:

Thank you for inviting public comment on the regulation of generative AI-enabled medical devices. I
am pleased to submit the attached comments in my capacity as Principal of Episodes of Care
Solutions, Ltd. (EOCS, Ltd.).

FDA's discussion paper presents an important opportunity to rethink how clinical intelligence is
evaluated as its capabilities extend beyond discrete software products into clinical decisions, care
pathways, and increasingly coordinated actions. I commend the agency for emphasizing competency
assessment, clinical confirmation, and postmarket monitoring.

My accompanying comment, “Toward a Clinical Intelligence Assurance Framework,” proposes an
additional perspective: clinical intelligence should be assessed not only for technical competency and
safety, but also for its performance within actual care delivery, the effectiveness of human
supervision, and its contribution to meaningful patient outcomes.

Three accompanying figures illustrate this proposal: a One HHS Map depicting opportunities for
interagency coordination; a Clinical Intelligence Assurance Framework spanning evaluation,
deployment, surveillance, and improvement; and a colorectal cancer screening and follow-up
pathway showing how multiple intelligence functions influence care over time.

I particularly encourage FDA to consider how its existing responsibilities can connect with
complementary authorities and capabilities elsewhere within HHS. The objective is not to expand
FDA jurisdiction by implication, but to encourage a more coordinated approach to assuring
intelligence-enabled care while respecting existing statutory boundaries.

Thank you for considering these recommendations. I would welcome opportunities for further
professional discussion as FDA and its partners continue this important work.

Respectfully submitted,

Frank Geo. Opelka, MD, FACS​
Principal​
Episodes of Care Solutions, Ltd. (EOCS, Ltd.)​
www.eocs.ltd

Submitted independently

Working public comment — October 16, 2026 • Not submitted
EOCS, LTD. / INDEPENDENT PUBLIC COMMENT / FDA-2026-N-7874

Toward a Clinical Intelligence Assurance Framework
From device competency to meaningful patient benefit | FDA docket FDA-2026-N-7874
Frank Geo. Opelka, MD, FACS | Principal, Episodes of Care Solutions, Ltd. (EOCS, Ltd.)

The opportunity
FDA’s discussion of generative AI-enabled medical devices is a valuable starting point. I submit these comments
independently of the American College of Surgeons. The challenge ahead is broader than assessing whether an
individual AI tool produces accurate outputs: clinical intelligence may identify risk, recommend action, orchestrate
referrals, shape treatment decisions, and monitor outcomes across settings and years. Assurance must address both
the performance of each function and the patient’s actual experience of the assembled care process.
I propose a Clinical Intelligence Assurance Framework (CIAF) as a policy construct—not an existing FDA regulatory
category. It asks three separate questions: (1) Does this clinical intelligence function require assurance? (2) What
evidence and controls are appropriate for its intended use? (3) Which agency, organization or professional body has
authority or accountability to supply that assurance? The answers will not always fall within FDA device jurisdiction.

1 | One patient, one HHS map
Figure 1 expresses the organizing problem visually: the person is at the center, while discovery, verification, data
exchange, deployment, population surveillance, and care delivery rely on distinct HHS functions. AI could help reveal
missed connections, reconcile evidence and reporting, and make handoffs auditable across organizational
boundaries. It should support coordination—not transfer legal authority from agencies or decisions from accountable
professionals.

Figure 1. Original EOCS “One HHS Map” concept, reproduced from the author’s October 2026 paper; agency pairings are proposed
coordination roles, not established legal assignments.

Working public comment — October 16, 2026 • Not submitted
EOCS, LTD. / INDEPENDENT PUBLIC COMMENT / FDA-2026-N-7874

2 | Assure functions, workflows and changes over time
FDA’s existing premarket and postmarket vocabulary provides a useful discipline for functions that fall within its
medical-device authority. For excluded/non-device software, analogous prerelease, deployment and longitudinal
assurance may need to arise from health systems, specialty societies, contracts, other laws or future policy. CIAF
would make these distinctions visible rather than imply that FDA regulates all consequential clinical software.
The assurance assessment should jointly consider function, clinical consequences, and demonstrated human control.
A theoretical override is not enough: the clinician must recognize a problem, have adequate time and information to
act, and be able to prevent or reverse harm. Patient and clinician corrections should change the assessment and
leave an audit trail.

Figure 2. Proposed CIAF lifecycle: functional evidence, real-world oversight, and controlled change across distinct authority lanes.

What should be measured?
Clinical correctness; consequential error and time to harm; subgroup performance; clinician–AI team effectiveness;
workflow delays and near misses; ability to correct or stop the function; version-specific drift; patient-reported
experience and goal attainment. Technical competency and guideline conformance are necessary observations, but
neither alone establishes meaningful patient benefit.

Working public comment — October 16, 2026 • Not submitted
EOCS, LTD. / INDEPENDENT PUBLIC COMMENT / FDA-2026-N-7874

3 | A colorectal cancer pathway makes the challenge tangible
Consider risk-based colorectal cancer screening and early detection. An intelligence function might synthesize age,
family history and prior testing, distinguish screening from diagnostic evaluation when symptoms warrant it, support
an informed choice, follow the referral, reconcile results, and ensure abnormal findings reach appropriate diagnostic
resolution. A patient reporting new family history must be able to trigger transparent reassessment. A perfect
screening recommendation is of little benefit if the test is not completed or an abnormal result is never closed.

Figure 3. Illustrative CRC pathway; intervention points are bidirectional and the treatment arc continues beyond diagnosis. No clinical
decision rule is validated by this schematic.

The same assurance problem continues in treatment. Stage, frailty, molecular findings and patient wishes inform
decisions that change as multimodal treatment is delayed, modified or discontinued. Probabilistic projections can
support discussions only when their published assumptions, uncertainty, transportability and observed calibration are
disclosed. A documented deviation by patient choice or clinical judgment is not automatically poor-quality care.

An inspectable assurance record
For each consequential function, I suggest a concise, versioned record of intended population, data and guideline
provenance, validation evidence, consequences of error, controls, human and institutional owners, patient-valued
outcomes, monitoring rules, and the conditions for suspension or revalidation. Such a record could support clinical
review, product oversight where applicable, and learning across settings.

Working public comment — October 16, 2026 • Not submitted
EOCS, LTD. / INDEPENDENT PUBLIC COMMENT / FDA-2026-N-7874

4 | Recommendations to FDA and HHS partners
1.​ FDA should clarify how proposed generative-AI evaluation principles address genuinely effective human
oversight, including intervention latency, user comprehension, realistic context and safe fallback.
2.​ Clinical confirmation should assess the clinician–intelligence team, patient-relevant outcomes and the
performance of the assembled workflow, not simply an AI model operating alone.
3.​ Where FDA has statutory authority, premarket evidence and postmarket monitoring should capture intended use,
model/knowledge version, downstream action, near misses, failures, drift and outcomes. Comparable clinical
assurance may be appropriate for non-device functions under other governance mechanisms.
4.​ FDA and the Secretary of HHS should explore a shared language and interoperable monitoring architecture
across FDA, CMS/CMMI, ASTP/ONC, NIH/AHRQ, CDC, OCR and clinical institutions while preserving each
entity’s lawful powers.
5.​ A bounded demonstration in colorectal screening and follow-up could test shared clinical assurance records,
patient corrections, referral closure, and accountable surveillance before recommending broader standards.

Closing
FDA’s initiative deserves support. The additional opportunity is to assure not merely competent AI instruments but
intelligible, adaptive, clinically beneficial care. Figures 1–3 depict a possible organizing architecture: one person,
interconnected institutions, a continuous assurance lifecycle, and an observable patient journey. I welcome further
discussion of practical, clinically grounded approaches consistent with statutory authority and specialty expertise.
Respectfully submitted,​
Frank Geo. Opelka, MD, FACS​
Principal, Episodes of Care Solutions, Ltd. (EOCS, Ltd.)

Selected source anchors
FDA, Considerations for the Regulation of Generative AI-Enabled Medical Devices (discussion paper and request for feedback),
docket FDA-2026-N-7874. https://www.regulations.gov/docket/FDA-2026-N-7874
FDA, Clinical Decision Support Software, final guidance (January 2026).
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software
FDA, Predetermined Change Control Plan guidance for AI-enabled device software functions (August 2025).
https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermine
d-change-control-plan-artificial-intelligence
Opelka, One HHS Map for Clinical Super Intelligence (October 2026), author-provided source graphic; Metadiagnostics colorectal
navigator simulations are illustrative and are not clinically validated.

Working public comment — October 16, 2026 • Not submitted