Octagos Health

The model is named, branded, positioned as the engine of the platform, and inseparable from what the company sells, which places this above both comparable records in this segment.

Atlas AI is described as the engine behind the platform: it reads incoming device data, triages transmissions, filters non actionable alerts, prioritises what requires clinical attention and surfaces findings across a population. Every operational claim the company makes traces back to it, from more than 25 percent noise reduction to reports delivered at 98 percent accuracy to five hours a week saved. Remove it and the platform is a viewer for transmissions the manufacturers already send.

Ask Atlas extends the model into a second mode. A natural language interface letting a clinic query its own monitoring data across patient care, operations, research and revenue is a distinct capability requiring language understanding over structured clinical data, and no other record in this segment offers anything like it.

The human layer does not reduce centrality, and the company's framing makes that clear. Certified specialists review the model's output rather than performing the triage themselves, so the model does the first pass across the whole transmission volume and humans verify. That is the same division iRhythm uses, and it is a quality control architecture rather than a hedge against a weak model.

Graded A. The one qualification is that the headline performance figures describe the combined system rather than the algorithm alone, so the model's standalone capability is not separately established.

The clearest and most deliberately constructed oversight design in this segment, and the company states the boundary in language it would be held to.

The architecture is a named product rather than a caveat. The Two Brain Approach pairs Atlas AI with specialists certified by the International Board of Heart Rhythm Examiners, and the company frames the reasoning directly: automation alone is not enough for near perfect performance in real world environments. That is a vendor arguing against full automation while selling automation, which is the opposite of the usual incentive.

The division of labour is stated precisely. Atlas AI reads incoming data, prioritises what needs attention and surfaces findings, and clinical teams review its output rather than handing judgment over to it. That sentence is doing real work: it locates the model upstream of judgement and the humans at the decision point, and it is written to be quoted back.

Certification matters here. Board certified cardiac device specialists are a recognised professional credential with an examination and maintenance requirement, so the review layer is staffed by people with independent professional standing rather than by trained operators.

The performance framing is consistent rather than opportunistic: the 99 percent figures are attributed to the combined system, so the company is claiming credit for the architecture rather than for the algorithm alone.

What holds this short of complete is that nothing describes what happens when the specialist disagrees with the model, whether overrides are recorded, or whether filtered transmissions remain auditable. Graded A on the strength of the design and the clarity of the stated boundary.

Function and role are described clearly, performance is published with an important qualification, and mechanism is absent.

The functional description is better than the segment norm. Atlas AI is named as a product rather than referred to as artificial intelligence generally, its role is stated as reading incoming device data, prioritising what needs clinical attention and surfacing findings across a population, and its position relative to the humans is explicit. Ask Atlas is described concretely with worked examples of the questions it answers, which tells a buyer what the interface actually does rather than what it is called.

Performance is published and, to the company's credit, attributed accurately to the combined system rather than to the algorithm. More than 99 percent accuracy, sensitivity and specificity for the Two Brain Approach, more than 25 percent noise reduction, 98 percent report accuracy. Naming what the figures measure is the disclosure that lets a reader compare correctly, and most vendors would have dropped the qualifier.

What is missing is everything mechanical. No architecture, no training data description, no evaluation method beyond the cited publication, no versioning and no update cadence for Atlas AI. For Ask Atlas nothing states whether the language capability is built in house or supplied by an external model provider, which is the disclosure a buyer most needs for a natural language interface over clinical data.

The algorithm's standalone performance is also unavailable, so a clinic cannot tell how much of the 99 percent is the model and how much is the specialists.

Graded C.

Nothing upstream is disclosed for either model capability, and the newer one makes the gap more consequential.

Atlas AI is named as a product and described by function, and nothing states whether it is built in house, licensed, or assembled on third party components. No training corpus, no framework, no infrastructure provider beyond the cloud marketplace relationship, and no bill of materials were located.

Ask Atlas is where this matters most. A natural language interface answering plain language questions about a clinic's own monitoring data almost certainly runs on a commercial large language model, because building one is not a plausible undertaking for a company of this size, and the company launched the capability without naming what powers it. Every unnamed provider in that path is a party potentially receiving clinical queries and the data used to answer them, and enumerating subprocessors is a basic requirement of a health system diligence process. A vendor introducing a conversational interface over patient data in 2026 without naming the model behind it leaves its customers unable to complete that assessment.

The device data supply chain is claimed broadly as rapid integration with all implanted devices, without enumerating manufacturers, where the competing Implicity record names all five.

One dependency is visible: the cloud platform through which the product is listed and with which it integrates.

Graded D as an absence of disclosure rather than evidence of a problem.

A publication in a leading cardiology journal, headline figures that need careful reading, and unusually specific customer accounts.

The publication is the anchor. The company attributes its performance claims to a study it describes as appearing in the Journal of the American College of Cardiology, which is among the most demanding venues in the field, and characterises the finding as demonstrating that artificial intelligence can reduce alert burden while preserving clinical vigilance. That framing is the right one, because preserving vigilance while reducing volume is precisely the tradeoff this product has to win.

The figures require a careful reading and the company does not hide the reason. More than 99 percent accuracy, sensitivity and specificity are attributed to the Two Brain Approach, meaning the model together with certified specialist review. That is a legitimate way to report a product that is sold as a combined service, and it is not comparable to an algorithm only figure. Implicity reports 98.3 percent sensitivity with specificity of 61.6 percent for its algorithm alone, and the difference between those two presentations is the human layer rather than model superiority. A separate and lower figure, 98 percent report accuracy, appears elsewhere in company material without the same qualification.

Operational evidence is credible and named. Chicago Cardiology Institute describes switching after exhaustive research with onboarding delivered to timeline, another customer describes device technicians working to the top of their licence, and a third describes population analytics identifying at risk patients early. A survey of allied professionals managing remote programmes was conducted with a professional society.

Graded B.

Nothing published addresses training data, retention or secondary use, and the platform's research positioning makes the omission pointed.

The company markets research readiness as a selection criterion in this category, publishing comparison material on queryable data, multi site support and published validation, and Ask Atlas explicitly names research as one of the four workflows it serves alongside patient care, operations and revenue. So the platform is positioned as a research instrument over clinic data, and nothing describes the governance that would make that appropriate: whether data leaves a clinic's tenancy for research, whether it is aggregated across customers, what de identification applies, or who may access it.

The training question is equally open. Atlas AI triages transmissions across a multi customer base and no statement addresses whether customer transmissions inform its continued development, whether learning is confined per clinic or pooled, or whether a customer can decline. A triage model improves with volume, so the incentive to pool is structural.

Ask Atlas adds a specific and newer concern. A natural language interface over clinical data sends queries and returns answers through whatever language model sits behind it, and nothing states whether that component is internal or an external service, what patient data reaches it, or whether queries are retained.

The specialist review layer is a third pathway, since vendor employed clinicians access patient transmissions routinely, and no access controls or retention terms are described.

Graded D on the absence.

No published position was located, and the service model widens the exposure beyond software alone.

The data flows are extensive. The platform ingests transmissions from implanted devices and ambulatory monitors, integrates bidirectionally with major record systems, runs population analytics across a clinic's whole cohort, automates billing documentation, and provides a clinic mobile application giving real time access to device data. Each is a protected data pathway.

The service layers add two more that software only vendors do not have. Certified specialists employed by the vendor review patient transmissions as a routine part of the product, so vendor staff have direct clinical data access at scale. And a dedicated patient connectivity team contacts patients to keep them transmitting, which means vendor personnel hold patient contact information and communicate with them directly. Both are legitimate service designs and both mean more people outside the covered entity are touching identifiable data than a platform sale would involve.

Nothing published addresses any of it. No business associate agreement is offered or described, no protected data handling summary exists, no retention position is stated, and nothing describes the controls around specialist access or patient contact.

Ask Atlas raises a further question specific to natural language querying. A clinic asking questions of its own data in plain language is generating queries that traverse whatever language processing sits behind the interface, and nothing describes where that happens or what is retained.

Marketplace availability implies the cloud provider's own vendor requirements were met, which is assurance about a listing rather than a published posture.

Graded D.

No published security posture was located, and a competitor's public claim about this company is the closest thing to evidence in either direction.

No trust centre, no service organisation control report, no information security certification, no penetration testing statement, no vulnerability disclosure policy, no subprocessor list and no incident notification commitment were found in anything examined.

The one external data point comes from PaceMate, which in announcing its own ISO/IEC 27001:2022 certification named Octagos among competitors relying on health privacy statute compliance and service organisation control attestations rather than the international standard. That is an interested source and it is also a specific, checkable public claim that this company has not contradicted, and it implies a service organisation control attestation exists even though none was located here. A vendor named in a competitor's comparison has an obvious opportunity to publish its own credentials in response.

The exposure is wider than software alone because of the service layers. Vendor employed specialists access patient transmissions routinely and a connectivity team contacts patients directly, so personnel security, access control and monitoring matter as much as platform security, and none of it is described.

Marketplace listing on a major public cloud implies the provider's vendor requirements were satisfied, which is assurance about a listing rather than a security posture.

Graded D on published evidence rather than on any judgement about the underlying engineering, and the contrast with the neighbouring PaceMate record on this axis is the sharpest in the segment.

No device clearance was located, and the company's regulatory attention is directed at reimbursement rules with real depth.

Nothing in the material examined indicates FDA clearance for Atlas AI or for the platform. As on the PaceMate record, that may be a defensible position for software that triages and routes data from already cleared devices without issuing its own diagnostic determination, and the human review layer strengthens the argument, since certified specialists rather than the model produce what reaches the clinician. Implicity took the alternative path and cleared three algorithms, so the choice was available.

The complication is the same one and it is sharper here because the claims are stronger. A model reporting more than 99 percent sensitivity that filters which transmissions a clinician sees is making a clinically consequential determination, and nothing published sets out the company's regulatory analysis or intended use statement.

Where the company demonstrates genuine regulatory command is payment. It publishes detailed guidance on the implanted device procedure codes, the separate codes for loop recorders and insertable monitors, the distinct remote physiologic monitoring code family, a code newly introduced for 2026 covering real time interactive communication, the 2026 sixteen day rule affecting one of those codes, and a forthcoming outcomes linked programme starting in 2027. That is a company tracking the rules its customers are paid under closely enough to teach them.

Graded C: an unstated position on device regulation, with strong command of the reimbursement regime.

A peer reviewed validation exists and no subgroup analysis was located, which places this above the segment floor without reaching disclosure.

The validation is real. Performance is attributed to a study the company describes as published in the Journal of the American College of Cardiology, examining whether artificial intelligence can reduce alert burden while preserving clinical vigilance. Submitting a triage system to peer review in a leading cardiology journal exposes the method to reviewers with no commercial stake, and the framing of the research question is the correct one for this application.

The company also demonstrates it thinks about population level variation. Population analytics across a clinic's whole implanted cohort is marketed as identifying patient trends and at risk groups, and a patient connectivity team exists specifically to reach patients who stop transmitting. That last point matters more than it appears, because patients who drop off remote monitoring are disproportionately older, poorer and more rural, and a company staffing a team to recover them is addressing an access gap that pure software vendors leave open.

What is absent is any performance breakdown. No figures by sex, age, device type, manufacturer or comorbidity were located, no model card exists, and no training population composition is described. The headline figures are aggregates for the combined human and model system, which makes model behaviour across subgroups doubly invisible, since a specialist reviewer may be compensating unevenly.

Graded C.

No published liability terms, and the service architecture allocates responsibility better than any other record in this segment.

The allocation is the substance. Certified specialists review the model's output before findings reach the clinic, and those specialists hold an independent professional credential from a recognised board. So a missed transmission is not a model failure delivered unchecked, it is a failure of a reviewed process in which a credentialed professional participated, and that professional is accountable in their own right. Compared with a pure software filter, where suppression happens with nobody in the chain, this is a materially stronger position for the customer.

The company's own framing supports it, stating that clinical teams review the model's output rather than handing judgment over to it, which places final responsibility with the clinic while the vendor carries the intermediate review.

The risk that remains is the one common to this segment. Triage decides what a clinician sees, a filtered transmission is reviewed by nobody downstream, and the published figures describe the combined system, so a clinic cannot separate how much of the 99 percent sensitivity comes from the model and how much from the specialists. If specialist staffing were reduced or review sampling introduced, performance would change in ways the published figure would not reveal.

Nothing published states accuracy commitments, indemnity, limitation, service levels for the specialist layer, or what happens if a missed event is later identified. The billing automation carries a secondary exposure of the kind recorded on the neighbouring record.

Graded C.

Bidirectional integration across major record systems, a specific commercial commitment attached to it, and independent customer confirmation.

The integration claim is bidirectional with major record systems, and the company publishes technical context rather than only assertions, describing how integration with the dominant enterprise systems typically works through interface feeds carrying data between the monitoring platform and the record. A vendor explaining the mechanism to prospective buyers is a vendor that has built it.

A named ambulatory record system integration is documented specifically, which matters because independent cardiology practices frequently run systems outside the two enterprise vendors and are routinely underserved by integrations built for hospitals.

The commercial commitment is what lifts this above the segment. Integration is offered at no cost, and a named customer confirms it independently, describing the transition as seamless with zero downtime and no added costs, while a second describes the company holding to its integration timeline. Record system integration is where implementations overrun, so a vendor absorbing that cost and customers confirming it delivered is a stronger interoperability signal than a list of logos.

Device side coverage is claimed as rapid integration with all implanted devices, and the platform also ingests ambulatory monitoring data, so two normally separate data classes arrive in one workflow.

Marketplace availability adds a procurement path integrating with the provider's cloud technologies.

Graded A.

The infrastructure relationship is named and the specifics are not, which places this at the segment midpoint.

What is disclosed is the cloud platform. The company lists in a major public cloud marketplace, describes integration with that provider's cloud services and technologies, and offers procurement with consolidated purchasing and billing through it. Naming the underlying cloud tells a security reviewer which control environment sits beneath the platform, and marketplace listing means some buyers can deploy under an existing enterprise cloud agreement rather than negotiating a new vendor contract.

A clinic mobile application is a further deployment surface, giving real time access to device data on personal or clinic devices.

What is absent is every specific: no region, no residency option, no subprocessor list, no retention position, no export or contract end terms, and no availability commitment for a service sitting in a monitoring pathway.

Two elements are distinctive to this vendor and undescribed. The specialist review layer means vendor employed clinicians access patient data, and where those staff are located matters for data handling; the company states United States based specialists, which is a partial answer and the only one given. And the patient connectivity team contacts patients directly, which implies systems holding patient contact details alongside clinical data with no described boundary between them.

Ask Atlas raises a third question, since natural language querying over clinical data implies processing somewhere that is never located.

Graded C on the strength of the named cloud platform and the stated location of the clinical staff.

No rate card, and two commercial commitments stated plainly enough to be held to.

The first is integration cost. A named customer describes the integration as no cost, with the transition seamless, zero downtime and no added costs, and the company markets the point directly. In a category where record system integration is routinely the largest line item in an implementation quote, committing to zero is a real and checkable commercial position rather than a discount.

The second is the absence of fees, stated as no fees and no surprises. That is ordinarily marketing language, and it carries more weight here because the company has published a detailed guide to the procedure codes that govern this market, including the 2026 updates and a newly introduced code for remote physiologic monitoring services. A vendor that educates buyers on the reimbursement rules is describing the economics its customers operate under rather than obscuring them.

Procurement is also easier than usual, with the platform available through a major public cloud marketplace since February 2026, which allows purchase through an existing cloud agreement with consolidated billing.

What is missing is the price itself. No per patient or per clinic rate, no service tier pricing for the specialist review component, no contract term and no minimum. The specialist layer is the significant unknown, since certified clinical review is a labour cost and nothing indicates whether it is bundled, tiered by volume or charged separately.

Graded B.

The broadest genuine coverage in this segment, spanning two monitoring workflows that normally sit in separate systems.

The distinguishing claim is combining implanted device monitoring with ambulatory electrocardiographic monitoring in one review workflow. The company's own chief medical officer frames the problem precisely: ambulatory volumes are growing faster than device clinic staffing, and in most clinics that work lives in a system separate from implanted device monitoring. Merging them changes where a clinician's time goes across both, and it removes a duplication most device clinics simply tolerate.

Device coverage is claimed as rapid integration with all implanted devices, and the company publishes a comparison of monitor types across holter, patch, mobile telemetry, event monitors and implantable loop recorders, which indicates working knowledge across the full ambulatory range rather than implanted devices alone.

Institutional coverage runs from individual device clinics to large cardiology groups, with dedicated material for the latter, and named customers include a cardiology institute and multi site programmes. Population analytics across a clinic's whole implanted cohort is a further coverage dimension, since it addresses the patients who are not generating alerts as well as those who are.

The service model extends coverage in a direction software alone does not reach. A dedicated patient connectivity team exists to keep patients transmitting, which addresses the patients who drop off remote monitoring entirely and are otherwise invisible.

Graded A.