AliveCor

Deep neural networks are the engine of the interpretation, and the devices record ECGs a clinician can read without them. The home page leads with Kardia 12L, "the easiest way to acquire a 12-lead ECG", a pocket sized system using a single cable and five electrodes; the device is cleared as an electrocardiograph in its own right, and the interpretation software, marketed as KAI 12L, is cleared separately and uses several deep neural networks to measure and interpret resting ECGs across rhythm, morphology and interval determinations, including acute myocardial infarction.

The company also sells KardiaMobile personal ECG devices, provider software (KardiaStation, KardiaComplete) and the KardiaCare Plus subscription, and reports more than 3 million lifetime users. Remove the models and the devices still produce tracings, so the models are the engine of a core module rather than the whole product.

The oversight rule is published in the cleared indication, and the highest stakes determination lacks its operating detail. The current interpretation software (K252589) assists a healthcare professional in measuring and interpreting resting ECGs, is used by professionals or trained personnel in offices, hospitals and acute settings, and its analyses are an adjunct that must be reviewed by a qualified physician.

What is missing is how the system presents uncertainty: nothing describes confidence reporting for a myocardial infarction determination, how borderline or poor quality recordings are flagged, or whether a conventional 12 lead confirmation is advised before acting, in acute and unconventional settings where the staff acting first may not be cardiologists. KardiaMobile personal devices return determinations to patients directly, and nothing published says what a patient is told to do with an abnormal result.

The approach, training data and cleared scope are stated, without performance figures. The current 510(k) summary (K252589) describes several deep neural networks trained on about 1 million 12 lead ECGs from about 400,000 patients at Emory University Hospital between 1985 and 2010, with the training population's sex, age and race composition, performing rhythm, morphology and interval analysis.

The company enumerates its 39 determinations, including acute myocardial infarction, and the clearance sequence (Corvair in June 2024, Corvair Monza in January 2026) marks versions. The Kardia 12L acquisition design, a single cable and five electrodes, is described.

What is missing is performance, since no sensitivity, specificity or PPV is published for any determination, and a description of how the models handle the reduced electrode set relative to a conventional ten electrode placement.

The model chain is structurally short: the interpretation models are AliveCor's own, cleared in its name (K231010, K252589) so they cannot be quietly swapped, and their training source is named in the filing as about 1 million ECGs from Emory University Hospital recorded between 1985 and 2010.

No hosting provider, subprocessor list or third party component is named for the consumer or clinical lines, and the agreement under which the Emory data was used is not described; recipients of data may be listed in the privacy policy. The interpretation software is cleared separately from the device and can work with other compatible hardware, which draws the model boundary explicitly.

Published research and deployment figures, short of published diagnostic accuracy for the lead product. The home page cites more than 200 clinical publications and more than 3 million lifetime users in more than 40 countries, largely built on the KardiaMobile personal devices.

For Kardia 12L and its interpretation software, the company has reported a peer reviewed study in Heart Rhythm O2 finding a 29 percent reduction in ECG acquisition time, a workflow result, use in more than 250 US practices with expansion to India, Australia and New Zealand, more than 4,000 identified cases of myocardial infarction and ischemia, and Medicare payment for hospital outpatient use approved in 2025.

The current 510(k) summary (K252589) reports that sensitivity, specificity and PPV were tested with no figures. The headline claims, the easiest 12 lead acquisition and heart attack detection from a reduced leadset, are answered on workflow and not on accuracy: detections counted are not misses measured.

The training data's source is named and the terms that govern patient data are not published. The company's 510(k) summary (K252589) names Emory University Hospital as the source of about 1 million training ECGs recorded between 1985 and 2010; the data use agreement, deidentification standard and research authorization behind that transfer are not described.

The company's privacy policy may state how ECGs recorded on its personal devices and clinical products are retained and whether they inform model development. The consumer line matters here: people recording their own ECGs on KardiaMobile generate data governed by consumer privacy law rather than HIPAA, where secondary use varies widely, and the company reports more than 3 million lifetime users. No retention, training use or deidentification position is stated elsewhere, and no safety engineering is published.

No HIPAA position is stated on the company's home page, though the privacy policy may give one. The data governance shape is unusual: KardiaMobile recordings made by consumers outside a clinical relationship fall under consumer privacy law rather than HIPAA, while Kardia 12L studies and provider software run inside provider relationships, so identical ECG data is governed differently by product, and nothing published on the pages available explains how the company separates the two or what happens when a consumer shares recordings with a physician. KardiaACCESS, launched in September 2026 under the CMS ACCESS Model, adds clinician guided care for Medicare beneficiaries, which carries its own HIPAA obligations.

No security certification, testing statement or disclosure route is named on the company's home page, and its security page (alivecor.com/security) is where one would be stated. The attack surface includes a consumer element: personal ECG devices pair wirelessly with patients' own phones running the Kardia app, a device class with a documented history of pairing, firmware and app weaknesses and specific FDA guidance, alongside clinical systems used in acute settings whose determinations inform urgent treatment. Premarket cybersecurity documentation exists for cleared connected devices and is reviewed by regulators rather than published for customers.

The regulatory position is identifiable and matches what is sold. FDA's database lists AliveCor clearances on 7 June 2024 for Corvair (K231010, product code MHX, ECG analysis) and Impala (K232035, DPS, an electrocardiograph), and on 9 January 2026 for Corvair Monza (K252589, MHX, with K231010 as predicate); the dates match the company's announcements of dual clearance for the KAI 12L interpretation software and the Kardia 12L device in June 2024 with 35 determinations, and of a further clearance in January 2026 adding short PR interval, atrial and ventricular bigeminy and left and right axial deviation to reach 39.

The database lists them under those development names, so a buyer matching KAI 12L to its clearance needs the company's statement or the date. The current indication assists a healthcare professional in measuring and interpreting resting diagnostic ECGs in offices, hospitals and acute settings, as an adjunct that a qualified physician must review, which matches the clinical positioning. KardiaMobile personal devices carry their own earlier clearances.

The training population is characterized in the company's filing and no performance by group is published. The 510(k) summary for the current version of the 12 lead interpretation software (K252589, cleared 9 January 2026) says its deep neural networks were trained on about 1 million 12 lead ECGs from about 400,000 patients at Emory University Hospital recorded between 1985 and 2010, 52 percent male and 48 percent female, mean age 61, 56 percent white, 33 percent African American, 2.2 percent Asian and 9 percent other, and that performance was tested with sensitivity, specificity and PPV, with no figures and no results by sex, age, race or body habitus.

A single US academic center's ECGs from before 2011 is the population the models learned from, while the product is sold into acute and unconventional settings in the United States, India, Australia and New Zealand. Two exposures bear on it: electrode contact with five electrodes varies with body habitus, chest hair, breast tissue and skin, and infarction presentation in women is more often missed in cardiac care.

A physician review requirement is part of the cleared use, and nothing stands behind the output. The current interpretation software's indication (K252589) says its analyses are an adjunct to clinical history, symptoms and other tests and must be reviewed by a qualified physician, so a clinician can challenge a determination before acting.

The summary states that sensitivity, specificity and PPV were assessed and performance is substantially equivalent to the predicate, but publishes no figures for any determination, including acute myocardial infarction or atrial fibrillation, so no error rate is on the record. No accuracy commitment, indemnity or remediation position is published.

A false negative infarction determination in an acute setting can delay escalation and a false positive can activate a catheterization lab; nothing says how a clinician should weigh a negative result in a symptomatic patient or a determination made on a poor recording. KardiaMobile personal devices return determinations to patients with no clinician in the loop.

One interoperability property is documented and record integration is not described. The interpretation software is cleared separately from the Kardia 12L electrocardiograph, and the company has said it can work with other compatible acquisition devices, so the model is not tied to one piece of hardware. Studies reach clinicians through the company's provider software, and Medicare billing for outpatient studies implies documentation reaching a record and a claim.

No electronic health record, standard (such as HL7 or DICOM ECG), API or marketplace listing is named, and nothing describes whether a completed study arrives in the chart as a document, discrete coded findings or a standard ECG file, including from ambulances, rural clinics and care homes where the company markets the device, or how a patient's own KardiaMobile recordings reach their physician.

Nothing published says where the system runs or where recordings rest. Personal devices pair with patients' own phones and recordings typically sync to an account; the clinical system's studies reach clinicians through provider software (KardiaStation, KardiaComplete). Where the interpretation executes (on the device, the phone or a cloud service), which hosting provider holds recordings, and in which region, is not stated, though the company's privacy policy may set out those terms.

The clinical system is sold in the United States, India, Australia and New Zealand, with European review under way, and India's data protection law and European transfer rules bear on where health data sits. For a device used to detect myocardial infarction in acute settings, whether a determination depends on connectivity is operationally material and unstated.

No price is published for the product the record leads with, and how it is paid for is discoverable. For Kardia 12L, the company has reported Medicare payment approval for hospital outpatient use in 2025, which is what the payer pays the provider for a study, not what a practice pays AliveCor, and it publishes a purchasing route for providers; no device, software, per study or subscription price is published.

In September 2026 it launched KardiaACCESS under the CMS ACCESS Model, giving eligible Original Medicare beneficiaries clinician guided care and connected devices for blood pressure, diabetes and heart health at no out of pocket cost, which settles the patient's cost, not the buyer's. KardiaMobile personal devices are retail products, a side line to the Kardia 12L clinical system.

Coverage is named from home to acute care, with validation behind part of it. The current interpretation software (K252589) is indicated for the general population, from healthy subjects to patients with cardiac and non cardiac abnormalities, in offices, hospitals and acute settings; KardiaMobile personal devices are cleared for people recording their own rhythm at home, and KardiaACCESS (September 2026) brings connected devices and clinician guided care to Medicare beneficiaries for blood pressure, diabetes and heart health.

The company markets Kardia 12L for ambulances, rural clinics, care homes and field settings, and sells it in the United States, India, Australia and New Zealand. Validation stands behind part of this: the clearances and the company's research on its personal devices, but no published accuracy figures for the 12 lead determinations and nothing validating the unconventional settings or populations outside the US academic training data.