Persivia

Real inference sits on top of a rules and content foundation the company describes openly. Soliton, which the company calls its AI engine, is described as bringing predictive, prescriptive and generative AI to population health: machine learning models predicting high cost cohorts, natural language processing extracting diagnosis codes from physician notes, speech interfaces for analytics, and generated tasks, notes and patient communications. Those are learned components doing real work.

The foundation is something else. The company describes CareSpace as built on more than a decade of clinical rules engine and analytics work, and the Soliton page says program design rests on algorithm development and semantic sets. The library of more than 200 programs, the data aggregation and the quality reporting would still function with the models removed, which makes AI a capable layer rather than the product.

Autonomy is described plainly and the scope is sensibly bounded, and the escalation path is missing.

The candor counts in its favor. The company states that software agents conduct outreach without requiring additional staff capacity, which is an accurate description of what autonomous patient contact is for. It does not dress the same function up as augmenting a care team while quietly selling the headcount saving.

The scope is bounded in a way that reduces risk. Named uses are follow ups, appointment reminders, medication adherence prompts and chronic care education, which are low acuity, high volume tasks where the cost of an imperfect interaction is small and the cost of not doing it at all is a missed care gap. Confining autonomous contact to that range of tasks is a defensible design choice rather than a lapse in oversight.

The analytics layer sits on the other side of the line, with risk stratification, alerts and care gap lists surfacing to care teams who decide what to do.

What is absent is what happens when a bounded conversation stops being bounded. A patient answering a medication adherence prompt may report chest pain, describe a fall, or say something indicating crisis. Nothing published describes how an agent recognizes that it has left its competence, who is notified, within what time, or whether a human reviews conversation transcripts at all. An automated channel that patients treat as a route to their care team creates an expectation of a listener, and no published material establishes that there is one.

The engine is named and described by function; nothing inside it is identified. Soliton is presented as the AI engine behind CareSpace, combining predictive, prescriptive and generative AI: machine learning for predicting high cost cohorts, natural language processing for extracting diagnosis codes from physician notes, speech interfaces for analytics, and generated worklists, notes, assessments and patient communications. The company describes the platform as resting on more than a decade of clinical rules engine work, so the boundary between authored rules and learned models is the useful thing to know, and it is not drawn.

The Soliton page states up to 90 percent accuracy in predicting high cost cohorts and 98 percent accuracy in extracting diagnosis codes from notes, with no method, metric definition, dataset or period behind either. No model, version, training data, language or speech model provider, or update practice is named.

The architecture is described and no model provider is named. CareSpace combines a data cloud described as a lakehouse for aggregation and management, the Soliton AI engine, a point of care tool (CareTrak), a laboratory results interface (SmartLab) and a marketplace. No subprocessor list, hosting provider, model or speech provider, data licensor or content source is named on the home, certifications, Soliton or privacy pages, and the privacy policy says only that information is shared with vendors and service providers to support or improve the platforms.

The generative layer is where the missing names matter most. Soliton generates patient communications and supports speech interfaces, and which language or speech model reads or hears patient content, under what retention terms, is not stated. The library of more than 200 programs encodes clinical guidance whose sources and update process are also unpublished.

Named customers with vendor reported results, and no method behind them. The home page attaches figures to named organizations: McLaren Health Care, $34 million in savings, which its case study page says made it the sixth best performing MSSP ACO in the country; HCA Florida Oak Hill Hospital, a 65 percent reduction in readmissions and 1,282 readmissions avoided; Prime Healthcare, $17 million in incentive payments; Harmony Cares, 85 to 100 percent HCC pass rates. The Soliton page adds up to 90 percent accuracy in predicting high cost cohorts and 98 percent accuracy extracting diagnosis codes from notes. No period, denominator, comparison or method accompanies any figure, and the case studies sit behind a download form.

One figure can be checked independently. CMS publishes MSSP results by ACO, so the McLaren ranking can be confirmed for the year it refers to, which the page does not state.

The NCQA Data Aggregator Validation designation is outside validation of the data the platform produces, which is a different question from whether its programs improve outcomes. The company cites no peer reviewed publication.

General assurances that do not answer the questions artificial intelligence raises. The platform privacy policy says Persivia may aggregate or de identify data collected through its platforms and use it for any purpose, keeps information for as long as an account is active or as needed to provide the service, and shares information with vendors and service providers only to support or improve the platforms. It does not say whether patient data trains the Soliton models, what is kept from generated patient communications or speech interfaces, or whether an external model provider processes that content.

The generative layer sharpens the question. Soliton is described as generating notes, assessments and patient communications, so unstructured protected health information passes through models whose retention and training terms are not published. The home page also describes continuous feedback loops that improve over time without saying whether that learning stays inside one customer's data.

A credible compliance posture at the certification level, with the contractual layer entirely private.

The supporting evidence is real. HITRUST certification of the platform maps healthcare regulatory requirements onto an audited control set rather than leaving compliance as an assertion, and the aggregation validation credential involves an external body examining data handling. Business associate status is the correct posture for a processor holding identified clinical, claims and social data on behalf of providers and plans.

What is absent is every negotiable term. No template business associate agreement, breach notification window, liability cap position, audit rights statement, subcontractor flow down provision or data return and destruction clause is published.

One feature of this product raises a specific obligation. Automated outreach contacts patients directly by text message and voice, which means protected health information leaves the platform through conversational channels into consumer devices. The compliance questions attached to that are specific: what is transmitted in an unencrypted text message, how patient consent to that channel is captured and recorded, and how a patient revokes it. The site addresses none of the three.

Two recognized attestations are named as the company's own, without the artifacts. The certifications page states HITRUST certification and a SOC 2 Type 2 report with HITRUST mapping. The company's March 2024 announcement identified the HITRUST certification as the implemented, one year (i1) level for CareSpace; the certifications page does not state the current level or date, and an i1 certification lasts one year. The same page lists NCQA Data Aggregator Validation, NCQA HEDIS measure certification and CMS public health reporting certification, which validate data and measures rather than how security is operated.

No report period, scope statement or trust portal is published, so nothing can be opened without asking.

Outside device regulation by function, with the patient facing layer sitting closer to the line than the analytics.

Risk stratification for program targeting, quality measurement, care gap identification and contract performance reporting are administrative and financial activities rather than diagnosis or treatment, so no clearance is required and none is claimed.

The regulatory weight the company carries sits in the payment domain and is substantial. Data validated under the aggregation program feeds quality submissions that determine payment under federal risk contracts, and the platform is explicitly built around Medicare shared savings participation, accountable care organization risk models and Medicare Advantage star ratings. Accuracy failures there are federally consequential even though they are not device matters.

The boundary case is patient facing education delivered by software. An agent conducting chronic care education by voice or text is communicating health information to a patient without a clinician composing the message, and where such software stays clear of device classification depends on whether it delivers authored educational content or generates individualized guidance. A language model conversing about medication adherence can drift from the first into the second without anyone deciding that it should. The company does not say which description applies or what constrains the agent's output.

The result is a product correctly outside device regulation, consequential in payment, and unaddressed at the boundary its newest capability approaches.

Nothing published on how model behavior is governed or tested. The home, certifications, Soliton and privacy pages carry no bias testing, subgroup results, fairness review, statement of target variables or model documentation.

The first mechanism at risk is one the company presents as a selling point. Soliton advertises up to 90 percent accuracy in predicting high cost cohorts, and predicting cost is the target Obermeyer and colleagues showed in Science in 2019 to understate the illness of Black patients, because less had been spent on their care at the same level of need. A cohort model built on cost directs care management away from the patients that failure affects unless it is corrected, and nothing says whether it is.

The second is social determinants targeting. Deprivation and demographic flags steer outreach toward groups least likely to meet outcome thresholds, which can improve equity and also places proxies for protected characteristics in a decision about who receives attention; which objective the targeting optimizes is not stated.

The third arrives with speech interfaces and generated patient communication, where recognition and language performance differ across accent, dialect and age, often for the patients a program exists to reach.

Nothing published on what happens when the system is wrong. The home, certifications, Soliton and privacy pages carry no warranty on analytic or generated output, no accuracy commitment, no indemnity, no service credit tied to data quality or availability, and no route to dispute an output.

The financial failure modes are substantial. Quality measures computed on incompletely ingested data move a reported score and the shared savings or star rating attached to it. A risk stratification that misses patients sends care management to the wrong people and a contract underperforms. For an independent accountable care organization, a single performance year can decide continued participation.

The patient facing exposure is distinct. Soliton generates patient communications, and automated outreach reaches patients about medication and chronic disease without a clinician composing the message. If generated content misstates an instruction, fails to escalate a patient describing a serious symptom, or reaches the wrong patient, the harm lands on a person. The provider whose name the message carries is the party the patient will hold responsible, and nothing published allocates that responsibility between the provider and the vendor whose software spoke.

Intake is broad and externally validated, with no record system named. Seven source types are named: clinical records from electronic health record systems, claims, health information exchange feeds, admission and discharge messages, patient reported data, home device readings and social determinants information. The NCQA Data Aggregator Validation designation on the certifications page is an outside assessment of this exact capability, and the platform produces validated clinical documents in a standard format. Named deployments stand behind it: McLaren Health Care unified more than 30 data sources, and Prime Healthcare unified data across 51 hospitals on several record systems.

Admission and discharge feeds are what make care transition programs work at all, since a program that learns of a discharge from a claim arrives weeks late.

What holds it below the top is mechanics. No record system is named, identity resolution across sources is not described, and nothing says whether insights write back into the record; the point of care tool, CareTrak, is listed without integration detail.

Nothing published about where the system runs or where data rests. The home, certifications, Soliton and privacy pages name no hosting provider, region, tenancy model, residency commitment, recovery objective or availability commitment; the data product is described as a lakehouse without saying where it sits.

The gap matters for a platform that assembles identified longitudinal records from clinical, claims, social, device and exchange feeds across whole attributed populations and runs care management and quality reporting on them daily. The site also presents a Saudi Arabia offering alongside the United States business, which makes residency a live question rather than a formality.

Pricing stays inside the sales conversation. No price, unit of charge, tier, implementation fee or minimum commitment appears on the site, and nothing states whether the products listed (CareSpace, the data cloud, CareTrak, Soliton, SmartLab and the marketplace) are sold together or separately.

Speed is the lead commercial argument, with go live in as little as nine weeks and five or six point solutions replaced, and that is an argument about cost published without the cost. The library of more than 200 programs raises the same question: included, bundled or licensed individually.

Broad on both clinical and buyer coverage, and the breadth is built rather than claimed.

Clinical coverage runs through a library of more than 200 evidence based programs spanning chronic disease management, care transitions, preventive care, behavioral health and social determinants. That is coverage a customer can inspect item by item rather than an assertion. Including behavioral health matters because it is the category most often left out of population health programs and the one whose omission most distorts a risk picture.

Buyer and contract coverage is similarly wide. Payers, providers and hospital systems are all addressed, and supported programs include Medicare shared savings participation, accountable care organization risk models, Medicare Advantage star rating improvement, commercial payment transformation and newer federal models, with performance monitored across multiple contracts at once, which is the operational reality for most organizations in this market.

Data source coverage supports the clinical claim. Clinical records, claims, health information exchange feeds, admission and discharge messages, patient reported data, home device readings and social determinants information all feed the record.

What is missing is evidence of depth. Program count establishes breadth of content and not the quality of any individual pathway, no customer count or installed base figure is published, and nothing establishes how the platform performs in a setting outside ambulatory and accountable care.