iRhythm Technologies

A deep learned algorithm reads every Zio recording before anyone else does, and the service around it also carries hardware and human work. A Zio patch records continuous single lead ECG for up to 14 days; the ZEUS software, which the company calls the first FDA cleared, deep learned algorithm, reads roughly a million heartbeats per recording; qualified cardiac technicians verify its output; and a physician receives the report.

At the company's volume, more than 12 million patients analyzed by its home page count, recordings of that length are beyond what people could read unaided, and successive clearances extended the algorithm from rhythm classification to beats, beat types and heart rate.

The patch, lighter in its current generation and worn through exercise, showering and sleep, and the technician verification layer carry value of their own, and the company describes the AI as verified by technicians rather than autonomous.

Who checks the model is described, and how they override it is not. The deep learned algorithm screens the full recording first, qualified cardiac technicians verify every report, and only then does a physician receive it; the company consistently describes the algorithm as verified by technicians rather than autonomous, and reports 99 percent physician agreement with end of wear reports from its commercial data. That is a human check on every case, kept even as the algorithm improved.

What is missing is how the parts interact: how often technicians change the algorithm's findings, what happens when they disagree, whether corrections feed back into training, and, for the Zio AT device, the criteria and timing for notifying a physician of an urgent arrhythmia during wear.

The approach is published in a journal, without the version and update discipline behind it. Hannun and colleagues (Nature Medicine, 2019) describe the deep neural network, its training data and its performance against cardiologists in the open literature.

The company adds that the model moved from machine learned to deep learned methods, detects thirteen arrhythmia classes, was extended from rhythms to beats, beat types and heart rate, and learns from an archive the home page puts at more than 2 billion hours of curated heartbeat data, and it discloses both the 21 percent sensitivity gain since 2010 and the markets where the deep learned algorithm is available.

What is missing is versioning: nothing tells a customer which algorithm version analyzed a given study or how changes to a continuously improving cleared algorithm are communicated, which the ZEUS clearance sequence (2009 to 2026) makes a live question.

The chain is short at the model layer. The ZEUS algorithm is iRhythm's own, cleared in its name through successive 510(k)s, trained on the company's own archive of recordings from its service (more than 2 billion hours by the home page's count) and described in Nature Medicine; the Zio Watch version was developed and cleared with Verily, which is named. The website privacy policy names marketing tools (Freshmarketer, HubSpot, Google Analytics).

The hosting provider and subprocessors behind the analysis pipeline are not named.

The evidence covers both the algorithm and the service. Hannun and colleagues (Nature Medicine, 2019) reported that a deep neural network classified a broad range of arrhythmias at diagnostic performance comparable to cardiologists, and the company describes more than 100 original research manuscripts on the service, including comparative effectiveness of ambulatory monitoring strategies in the American Heart Journal, a Medicare beneficiary utilization study, a managed care analysis of monitoring variation and work in an electrophysiology journal; the comparative effectiveness and utilization studies ask whether the monitoring strategy changes outcomes and cost, not only whether the algorithm agrees with a reader.

The home page's 99 percent physician agreement figure comes from the company's review of end of wear reports across device generations. The company notes that a key comparative study used previous generation device data, with current devices deemed substantially equivalent.

The home page reports more than 2 billion hours of curated heartbeat data and more than 12 million patients analyzed, and the company has described this archive as the basis for algorithm training and quality assurance. Each hour is a recording from an identifiable patient whose physician ordered a diagnostic test, and curation implies human labeling. The website privacy policy (effective January 2026) defers patient cardiac monitor data and the MyZio app to separate notices, including a HIPAA Notice of Privacy Practices, and says reasonable safeguards are used.

The basis for using patient recordings in training, and how they are de identified, is not published.

iRhythm bills payers for the diagnostic service it performs, and its website privacy policy (effective January 2026) says protected health information is governed by HIPAA and handled under a separate Notice of Privacy Practices, with patient cardiac monitor data and the MyZio app under their own policies; the trust center has a Privacy section. The website policy itself covers site visitors, and the company holds the raw ECG of every patient it monitors.

No business associate agreement or contract terms for health systems are published.

The company's site footer links a trust center with Security, Quality, Ethics and Compliance and Privacy sections; its overview carries leadership statements on integrity and quality, and the website privacy policy says reasonable safeguards are used. The surface is broad: an archive of labeled ECG recordings from more than 12 million patients, a provider portal and app, the MyZio patient app, record integrations and connected wearables. Premarket cybersecurity documentation exists for cleared connected devices, and as a public company iRhythm reports material incidents in securities filings.

The trust center's security page states that Zio by iRhythm holds a SOC 2 Type II report from an independent third party covering security, availability, confidentiality and privacy, adds a FIPS 140-2 certificate (number 3118) and regular third party HIPAA audits, and names the report rather than publishing it.

FDA's database lists 18 510(k) submissions from iRhythm, among them the ZEUS analysis software (K091075, 2009; K202527, May 2021; K222389, February 2023; and the ZEUS Platform, K252859, May 2026; product code DQK), the Zio Watch version of ZEUS for atrial fibrillation detection and burden (K213409, July 2022), the Zio patch and Zio monitor (K113862 and K121319, 2012; K202359, 2021; K243650, August 2025; DSH), Zio SR (K143513) and the Zio AT device (K240029, QYX, and K240177, DSI, October 2024).

The home page's claim of the first FDA cleared, deep learned algorithm verified by QCTs matches the cleared ZEUS software, and each capability step (rhythms, then beats, beat types and heart rate, then the watch) carried its own submission. Company material reports CE and UKCA marking and Japanese approval, and states that the deep learned algorithm is available only in the United States, EU, Switzerland, United Kingdom and Japan, so studies read in other markets do not use it.

Hannun and colleagues (Nature Medicine, 2019), with iRhythm authors, reported a deep neural network classifying a broad range of arrhythmias at diagnostic performance comparable to cardiologists, and the company states that average rhythm detection sensitivity has improved 21 percent since its first algorithm in 2010, acknowledging that performance changes with versions. For a single lead adhesive patch, skin contact and signal quality can vary with skin tone, body hair and body habitus, which changes how much of a recording the algorithm can classify, and atrial fibrillation prevalence varies by ancestry.

Results by sex, age, ancestry, body habitus or skin tone are not published.

The service puts people between the model and the patient: the algorithm reads the recording first, qualified cardiac technicians verify its output, and a physician receives the report and decides, so a missed arrhythmia is a failure of a reviewed process rather than of an unsupervised model. The home page reports 99 percent physician agreement with end of wear reports, based on the company's commercial data, which measures acceptance rather than error. The harm that matters is delayed: a missed atrial fibrillation during the wear period can surface as a stroke months later.

No accuracy commitment, indemnity or remediation terms are published.

The company describes a provider portal and mobile app for managing patients and reports, electronic health record integration that simplifies data collection and entry and widens access to reports, and the MyZio patient app. For a diagnostic service the payload is a report with coded findings, and the company ties its arrhythmia classes to procedure codes.

The record systems it integrates with, and whether integration delivers a document or discrete coded results, are not named.

The service is centralized by design: patients return or transmit recordings, analysis and technician verification run on the company's infrastructure, and reports return through a provider portal, record integrations and the MyZio app, so the company holds every patient's raw ECG. The website privacy policy (January 2026) states certification under the EU US and Swiss US Data Privacy Frameworks, a mechanism for transferring personal data to the United States, and the company operates in the United States, the EU, Switzerland, the United Kingdom and Japan.

Where recordings are processed and stored is not published.

Zio is a reimbursed diagnostic service: a physician orders a study, the patient wears the patch, and the service is billed to a payer under procedure codes rather than licensed to the practice. The company says its thirteen detected arrhythmia classes were included in the procedure code review process, and it publishes Medicare utilization and managed care cost research, so for a practice the relevant figures are the payer's rate and coverage.

What a practice or health system pays iRhythm directly, including when a study is not covered, is not published.

Zio covers thirteen arrhythmia classes over up to 14 days of continuous ambulatory recording: Zio monitor and Zio XT for long term recording, Zio AT with transmission during wear and Zio Watch for atrial fibrillation burden, each form cleared separately, with the algorithm tested in the peer reviewed literature and the service studied in Medicare beneficiaries and in comparative effectiveness research.

The setting is ordinary life: patients exercise, shower and sleep in the patch, which is why intermittent arrhythmias are caught. The company reports US clearance, CE and UKCA marking and Japanese approval, and states the deep learned algorithm is available only in the United States, EU, Switzerland, United Kingdom and Japan.