Bob Wachtl

See attached file(s)

TO: Division of Dockets Management (HFA-305), Food and Drug Administration
LOCATION: 5630 Fishers Lane, Rm. 1061, Rockville, MD 20852
DATE: October 6, 2026

Public Comment Response: Generative AI-Enabled Medical
Devices
Re: Docket No. FDA-2026-N-7874 — Considerations for the Regulation of Generative AI-Enabled Medical
Devices

SUBMITTED BY: Dr. Bob Wachtl, DNP, MHA
Professor of Nursing & Clinical Health Informatics Specialist
Ursuline College / Gannon University — Contact: bwachtl@gannon.edu

EXECUTIVE SUMMARY

As a nurse scientist, doctoral educator, and health informatics specialist, I strongly support the FDA Center for
Devices and Radiological Health (CDRH) in establishing a competency-based, total product life cycle (TPLC)
regulatory framework for generative artificial intelligence (GenAI) medical devices. Modeling medical software
oversight on clinician licensure, board credentialing, and ongoing practice review represents a practical and
clinically grounded step forward.
However, translating this regulatory vision into safe clinical practice requires examining how these tools
function at the bedside. In fast-paced hospital units and clinics, artificial intelligence does not operate in
isolation. It enters a dynamic clinical environment where human judgment, professional intuition, and direct
physical assessment must remain the final authority. To protect patients while advancing technology, the FDA's
regulatory framework must explicitly ensure that the bedside nurse retains a controlling voice in care decisions,
working in direct coordination with the broader interdisciplinary healthcare team.
1. PRESERVING HUMAN AUTHORITY & THE NURSE'S CONTROLLING VOICE

Evaluating an AI model's accuracy on static benchmark testing alone creates a false sense of security. In clinical
practice, software that performs exceptionally well in laboratory trials can still contribute to care failures if it
bypasses or undermines human reasoning at the point of delivery.
Core Clinical Principle: Premarket clinical confirmation must evaluate the integrated Human-AI Team
within real-world workflows. Software architecture must be designed to support human authority rather than
displace it. The bedside nurse—who maintains continuous contact with the patient—must serve as an essential
controlling voice, weighing AI suggestions against hands-on clinical findings before any care plan is altered.
Artificial intelligence recommendations must always function as supportive inputs to be evaluated, validated, or
rejected by the nurse and interdisciplinary team—never as primary or automated drivers of patient care.

Docket No. FDA-2026-N-7874 | Public Comment Page 1 of 3
2. RE-EVALUATING "ACTION-DIRECTING" INFORMATION RISKS

The FDA's proposed two-axis risk framework categorizes "Informational: Action-Directing" software outputs at
a lower risk level than direct action-taking functions. In real bedside practice, this theoretical boundary
frequently disappears.
When a GenAI tool presents a specific care recommendation—such as proposing a medication dose change,
suggesting a triage priority, or advising care escalation—clinicians working under high patient acuity and tight
time constraints experience strong functional pressure to accept the suggestion. Universal disclaimers like
"consult a physician" do not reduce this operational pressure.
REAL-WORLD CLINICAL WORKFLOW PRESSURE SHIFT

Bedside Operational Reality
Informational Action-Directing High Acuity + Severe Time Pressures
Non-Directive Prompt / Suggestion Converts "Information" into de facto Action

Recommendation: The FDA should classify any software function generating specific clinical
recommendations as a higher-risk activity unless the system provides transparent, immediate links to primarysource clinical guidelines that the care team can quickly verify at the bedside.
3. MAPPING BENCHMARKS TO CLINICAL JUDGMENT FRAMEWORKS

The FDA's proposed non-clinical benchmarking elements (S.1–S.3, E.1–E.4) offer a solid starting point, but they
require structural alignment with recognized clinical reasoning models. The agency should map these
benchmarking categories directly to the established NCSBN Clinical Judgment Measurement Model
(NCJMM).
FDA Benchmarking Element NCJMM Clinical Reasoning Step & Interdisciplinary Role

S.1 Safety Escalation & E.2 Gathering Recognizing & Analyzing Cues: Bedside nurse detects subtle
physiological changes and validates early warnings.

E.1 Clinical Knowledge Fidelity Formulating Hypotheses: Interdisciplinary team evaluates clinical options
using current evidence.

E.3 Quantitative Analysis Generating Solutions: Precise verification of dosage calculations, lab
values, and metabolic rates.

E.4 Communication Quality Evaluating Outcomes: Clear interprofessional handoffs, empathetic
patient care, and team debriefing.

Docket No. FDA-2026-N-7874 | Public Comment Page 2 of 3
4. MANAGING OVER-ESCALATION & PROTECTING CLINICAL INTUITION

CDRH correctly identifies both under-escalation and over-escalation as critical evaluation factors. While failing
to flag a deteriorating patient causes immediate harm, systematic over-escalation introduces severe long-term
risks. Frequent unnecessary alerts produce severe alert fatigue among nursing staff, overcrowd emergency
departments, trigger redundant diagnostic testing, and erode clinician trust in digital tools.
Premarket review must require developers to quantify and limit false-positive recommendations. Most
importantly, software algorithms must not be permitted to override or crowd out the seasoned intuition and
hands-on judgment of the bedside nurse and attending care team.
5. MANDATING FOUNDATION MODEL TRANSPARENCY (MAFS)

Relying on third-party foundation models creates significant visibility challenges for healthcare organizations
and software developers. While voluntary Foundation Model Master Files (MAFs) offer a practical path forward,
the FDA should require specific core disclosures before an MAF can be referenced in a device submission.
Participating model developers must disclose system guardrail architectures, safety refusal boundaries, finetuning data sources, and formal commitments to notify device sponsors prior to deploying underlying model
updates.
6. SUMMARY POLICY RECOMMENDATIONS

Maintain Human Final Authority: Ensure the bedside nurse holds an active controlling voice in care
decisions, working alongside the full interdisciplinary team.
Evaluate Human-AI Collaboration: Require premarket clinical trials to measure the combined
performance of the human-software team rather than testing algorithms in isolation.
Refine Activity Risk Classifications: Recognize that action-directing prompts carry high functional risk
during high-acuity, time-pressured clinical workflows.
Adopt Standardized Clinical Frameworks: Align premarket benchmarking rubrics directly with
established NCJMM clinical judgment steps.
Enforce Foundation Model Disclosures: Require clear guardrail documentation and advance update
notices in Foundation Model Master Files.
Respectfully submitted,

Dr. Bob Wachtl, DNP, MHA
Professor of Nursing & Clinical Health Informatics Specialist
Ursuline College / Gannon University
Email: bwachtl@gannon.edu

Docket No. FDA-2026-N-7874 | Public Comment Page 3 of 3