RhythmX AI

The company is AI native by construction rather than by repositioning. Generative and predictive models are the product, producing the recommendations that constitute the entire value proposition, and there is no prior system of record or services business underneath. It draws on the technical depth of its parent group, which the company describes as 250 data scientists and a thousand people dedicated to healthcare.

Note for the future rather than for this grade: the merger with a patient engagement company broadens the combined entity considerably, and a later assessment should check whether the AI remains the centre or becomes one component of a wider platform.

This product recommends clinical actions rather than reporting clinical facts, and that raises the bar it has to clear. Its stated outputs include recommended laboratory tests, imaging, medications and follow ups, plus routing a patient to a particular clinician type or channel.

This index has consistently graded order adjacent recommendation at C where nothing is published about how the recommendations behave, on the reasoning that an error in a note is a documentation problem while an error in a recommended order is a clinical action. No confidence threshold, no routing rule, no abstention behaviour for a thin record and no error rate was located, and no statement describes what a clinician must verify before acting. The drill down capability is a genuine mitigation and is the reason this is not lower, since a physician can interrogate the basis of a recommendation rather than accept it blind.

The platform is described as combining predictive algorithms with generative capability over longitudinal data, and clinicians can drill into a recommendation to see what sits behind it, which is a real verification affordance rather than a black box. Beyond that nothing is disclosed: no model or model family is named, no accuracy figure exists for any recommendation type, no validation methodology is published, and no model card was located. For a product that recommends laboratory tests, imaging and medications, per recommendation type performance is the disclosure that matters and none of it exists publicly.

Nothing on patient data handling was located: no retention, deletion, secondary use, model training, sub processor or residency position. Two structural features make that absence more consequential than for a typical platform of this size. First, the data foundation is not the company's own.

The platform is described as leveraging assets of its parent group, including that group's artificial intelligence platform and longitudinal data covering roughly three hundred million patients, several billion annual claims and more than a million healthcare professionals. Nothing published states where that corpus came from, on what legal basis it was assembled, whether it is de identified and by which method, or whether data flowing through a customer's deployment joins it.

A platform whose central claim rests on the scale of an inherited data asset owes an account of that asset's provenance, and inheriting it does not transfer the answer. Second, the inputs are unusually broad. The platform explicitly incorporates social determinants, lifestyle, dietary, physical, cultural and mental health factors alongside clinical data, and much of that category is not protected health information in the statutory sense.

That produces an uncomfortable result worth stating plainly: the most sensitive and most re identifying inputs to the platform may be the ones carrying the least legal protection, so a business associate agreement may not reach them at all. Ask which protections apply to the non clinical inputs specifically, and for the corpus provenance.

One named deployment carrying real weight, and one number that needs reading carefully. Presbyterian Healthcare Services expanded to 200 primary care clinicians in February 2026, described as a first full system deployment rather than a pilot, with the system's chief medical information officer named and quoted.

The reported early result carries a genuine denominator, which is more than most vendors offer: more than 200 combined HCC and non HCC conditions identified across more than 10,000 patient encounters within weeks of launch. That works out at roughly two percent of encounters, a modest and therefore plausible figure rather than an inflated one, though the vendor does not present it that way.

Held at B because it is a single site, vendor published, with no independent evaluation, and because conditions identified is a volume measure rather than a measure of whether the identifications were correct or acted upon. A clinical advisory board with named executives from Prime Healthcare, Sentara Health and Mass General Brigham adds credibility to the design process without being evidence about outcomes.

Nothing on patient data handling was located. There is no published statement on retention, deletion, secondary use, model training, subprocessors or residency.

Two structural features make the absence more consequential here than for a typical platform of this size.

First, the data foundation is not the company's own. RhythmX AI is described as leveraging assets of its parent group, including that group's artificial intelligence platform and longitudinal data covering roughly three hundred million patients, over four billion annual claims, and more than a million healthcare professionals across several hundred thousand facilities. Nothing published states where that corpus came from, on what legal basis it was assembled, whether it is de identified and by which method, or whether data flowing through a customer's deployment joins it. A platform whose central claim rests on the scale of an inherited data asset owes an account of that asset's provenance.

Second, the inputs are unusually broad. The platform explicitly incorporates social determinants, lifestyle, dietary, physical, cultural and mental health factors alongside clinical data. Much of that category is not protected health information in the statutory sense, which produces an uncomfortable result: the most sensitive and most re identifying inputs to the platform may be the ones carrying the least legal protection. Buyers should ask which protections apply to the non clinical data specifically, since a business associate agreement may not reach it.

No HIPAA statement, business associate agreement, template or terms were located in any retrieved material, and the company does not identify itself as a business associate anywhere found.

This is a straightforward gap rather than a considered position. The platform reads electronic health record data and delivers recommendations inside the clinician's workflow, which places it in a business associate relationship with every provider organisation deploying it. An agreement therefore exists in the contracting process; it is simply not visible to a buyer beforehand, and no public material addresses the relationship at all.

The corporate structure adds a question a buyer should ask early. The company operates inside a larger group, draws on that group's artificial intelligence platform and data assets, and has since been combined with a sister company into a joint offering. Buyers should establish which legal entity will be their counterparty, which entity signs the business associate agreement, and whether the agreement reaches the affiliated companies whose platform and data the product depends on. A group structure can leave the contracting entity thinner than the marketed capability implies.

No trust centre, security page, certification or third party attestation was located across two separate searches, and no SOC 2, HITRUST or ISO 27001 claim appears in any material retrieved from the company. No security controls are described.

A scope caveat belongs on this record. The company's name collides closely with several unrelated cardiac monitoring businesses that do publish detailed security credentials, which makes this a difficult question to search cleanly and produced a set of near misses that a casual reader could easily mistake for this vendor. That is worth stating because it cuts both ways: it may have obscured a genuine disclosure, and it also means third party summaries about this company should be treated with suspicion.

The absence is notable given the deployment scale claimed for the combined offering and the profile of the institutions on its clinical advisory board, whose own vendor security reviews would ordinarily require an attestation before production use.

Buyers should ask what the company holds in its own name, and should ask separately whether any attestation being offered belongs to the parent group or a sister company rather than to the contracting entity. A certification held elsewhere in a corporate group does not cover the product being purchased.

No clearance, authorisation or published regulatory position was located, and the grade records the absence of a stated position rather than a judgement that clearance is required.

The language the company uses to describe its own output is the reason this warrants attention rather than a shrug. The platform is described as delivering patient specific prescriptive actions and recommendations, generated by predictive and generative models, surfaced inside the electronic health record workflow. Prescriptive is a materially stronger word than informative, and a system that tells a clinician what to do for a named patient sits closer to the regulated boundary than one that retrieves relevant evidence and leaves the reasoning to the reader.

The clinical decision support exclusion under the 21st Century Cures Act turns substantially on whether the clinician can independently review the basis for a recommendation. The company states that recommendations can be drilled into through a natural language interface, which points in the right direction, but nothing published describes what a clinician sees when they do, or whether the underlying evidence is exposed at the level the exclusion contemplates.

A second function deserves its own question. The platform performs patient orchestration and routing to a chosen clinician type and channel. Routing decisions affect who a patient sees and when, which is an operational function with clinical consequences and is not obviously covered by the same analysis. Buyers should ask the company to state its regulatory position and to address the recommendation and routing functions separately.

The grade describes disclosure and incentive structure, and the counterweights belong alongside. Recommendations are traceable through a drill down interface, the clinical advisory board carries named senior clinicians from three major systems, and the deploying organisation's own chief medical information officer framed the adoption in terms of responsible AI use rather than novelty, which is a stance worth crediting.

Against that, the headline measured outcome at the named deployment is the identification of HCC and non HCC conditions, and the stated purpose includes strengthening performance in value based arrangements, which places the product on the risk adjustment gradient this index tracks across the category.

Separately, no fairness, subgroup or demographic performance disclosure was located, and that gap is consequential here in a way it is not everywhere: the platform explicitly incorporates social determinants and lifestyle data into recommendations, and a model that uses social data to shape clinical recommendations is precisely the kind that requires subgroup performance reporting rather than the kind that can defer it.

One verification affordance is real: clinicians can drill into a recommendation to see what sits behind it, so a suggestion arrives with its supporting material rather than as a bare instruction, and a recommendation resting on something the clinician knows to be wrong can be rejected on that basis. That is a genuine control and it is more than a black box score. Held at C because nothing measures any of it and the output types are not equivalent.

The platform recommends laboratory tests, imaging and medications, and per recommendation type performance is the disclosure that matters, because those three carry different costs when wrong: an unnecessary test wastes money and time, an unnecessary scan adds radiation and incidental findings that generate their own cascade, and a medication recommendation reaches the patient's body. A single figure across all three, which is not published either, would describe none of them.

No model or model family is named, no accuracy figure exists for any recommendation type, no validation methodology is published, no model card was located, and no warranty, indemnity or remediation commitment attaches. Ask for performance by recommendation type, the acceptance rate by clinicians, and what the system does when the evidence behind a recommendation is weak.

The company describes the platform as EHR agnostic and, more usefully, its named deployment demonstrates it: clinicians at Presbyterian Healthcare Services work with the data presented in a single consolidated workflow inside their Epic EHR. Ingestion spans more than ten source types including EHR data, clinical policies and guidelines, payer and financial data and social data, which is genuine breadth rather than a single feed.

Held at B because Epic is the only EHR evidenced, no other system is named, no integration standard or mechanism is described, and no marketplace listing or third party technical review was located to corroborate the agnostic claim.

The platform is described as electronic health record agnostic, which is a real architectural claim and the most useful deployment fact published. Beyond it, nothing: no hosting arrangement, cloud provider, infrastructure description, implementation timeline, uptime commitment or data residency statement was located.

One widely repeated figure needs care and is the reason this record should be read closely. Deployment scale of roughly a hundred and fifty health systems and eighty five million patients, including several million veterans, is reported for the combined offering formed when this company was merged with a sister patient engagement company inside the same corporate group. That footprint is the two businesses together and predominantly reflects the engagement platform's installed base. It should not be read as the deployed scale of this product, and a buyer evaluating the precision care platform specifically should ask how many organisations run that component, for how long, and at what stage of maturity.

Buyers should also establish where data is processed and stored, whether any deployment option keeps data inside their own environment given the platform draws on group data assets, and what is contractually committed on availability.

No pricing information of any kind was located: no list price, no pricing unit, no tier structure, and no indication of whether the platform is licensed per physician, per attributed patient, per health system or against measured performance.

What is published is capital rather than commercial terms. The company launched with fifty million dollars of initial funding from its parent group, and its corporate parentage and sister companies are named openly. That tells a buyer the business is well capitalised and who stands behind it, which is not nothing, but it says nothing about what the product costs or how the charge is structured.

The pricing question interacts with the corporate structure in a way worth raising directly. The platform depends on an artificial intelligence platform and a large longitudinal data asset owned elsewhere in the group. Buyers should ask whether access to those assets is priced separately or bundled, whether continued access is contractually guaranteed for the term, and what happens to a deployment if the group restructures those assets or the entity is combined again. The product has already been merged once into a joint offering with a sister company, so that is a live rather than a hypothetical question.

Deliberately focused rather than broad, and the focus is a strength. The platform targets primary care and whole person management of chronic conditions, which the company identifies as the setting where fragmented data and time pressure bite hardest, and the named deployment is a primary care clinician population.

Coverage extends across clinical, laboratory, social determinant and lifestyle dimensions within that setting, and routing logic reaches beyond the primary care physician to advanced practice providers and specialists. Graded B rather than A because the depth is within one care setting, the range of chronic and acute conditions covered is described as expanding rather than complete, and no specialty specific instrument level behaviour was located.