iRhythm Technologies (Mintu Turakhia)

iRhythm Technologies Comments

October 19, 2026

Submitted Electronically

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

Re: FDA’s Discussion Paper on the Regulation of Generative AI-enabled Medical
Devices, Docket FDA-2026-N-7874

Dear Dr. Tarver:

On behalf of iRhythm Technologies, Inc. (iRhythm), we appreciate the opportunity to
comment on the Discussion Paper on the Regulation of Generative AI-enabled
Medical Devices.

iRhythm is a digital healthcare company dedicated to improving cardiovascular
outcomes through innovative diagnostic technologies, advanced artificial intelligence
(AI) and comprehensive clinical services. The company’s mission is to create trusted
solutions that detect, predict and prevent disease. iRhythm participates in Medicare
and Medicaid as an Independent Diagnostic Testing Facility (IDTF) and manufactures
an FDA-cleared, patch-based external wearable monitor, branded under the Zio®
name.

iRhythm appreciates FDA’s efforts to develop a risk-based framework for generative
AI-enabled medical devices. As a manufacturer of regulated medical devices
incorporating deep neural network (DNN) algorithms, we encourage FDA to clearly
distinguish generative AI from established classification and prediction DNN
technologies.

DNNs used for bounded diagnostic tasks—such as detection, classification,
prediction, or estimation from physiologic signals—should not be considered
generative AI solely because of their architecture, use of foundation-model
techniques, or integration with a generative component. The relevant distinction
should be the intended function and behavior of the output, including whether it is
bounded and objectively verifiable versus open-ended or generative.

For bounded DNN-based diagnostic functions, existing analytical and clinical
validation approaches remain appropriate. Premarket and postmarket performance
can be objectively characterized using prespecified measures such as sensitivity,
specificity, agreement, calibration, positive and negative predictive value. A
competency-based framework designed for open-ended generative systems should
not displace these established validation and quality approaches.
We also encourage FDA to consider the intended clinical workflow when assessing
risk. Human review, availability of underlying physiologic data, downstream
safeguards, reversibility, and whether an algorithmic output is one component of
clinician decision-making can substantially mitigate the consequences of an incorrect
output.

Finally, lifecycle requirements should distinguish locked, version-controlled DNNs
from continuously adaptive models. Post-market monitoring, real-world evidence, and
predetermined change control plans can provide important safeguards, but should
complement a risk-based framework rather than impose generative-AI-specific
requirements on established DNN technologies whose performance can be
conventionally and objectively validated.

A clear boundary between established DNN-based medical device algorithms and
genuinely generative or agentic AI will help preserve regulatory clarity while allowing
FDA to appropriately address the novel risks presented by generative AI. Thank you
for considering our comments. If you have any questions, please feel free to reach
out to me at mintu.turakhia@irhythmtech.com.

Sincerely,

Mintu Turakhia, MD, MS
Chief Medical and Scientific Officer