Carenostics

Machine learning drives the core function, and a monitoring and workflow platform carries value around it. Carenostics now sells an AI administrative assistant for transplant teams: it gathers records from the outside health systems a patient touches, extracts the clinically relevant findings from unstructured documents, and files updates and committee reports back into the record as structured data. The extraction and summarization depend on learned models. The rest of the product is monitoring and workflow: alerts when an outside hospitalization or overdue result occurs, checklist tracking for waitlisted patients, annual check ins and report assembly, which would still run as rules without the models, with far less to show.

The company's earlier product, models that find undiagnosed chronic disease in a health system's own data, is no longer sold on the current site.

Review steps are described, and the company's own copy promises the output is complete. Extracted findings go to physicians and surgeons for review with source links, but the assistant also files updates and assembles reports automatically, writing everything back to the record as structured data, and fires alerts when an outside hospitalization, missed lab or expired result changes a waitlisted patient's status. No threshold for what is written back without review is published.

The product page says of waitlist management that "no patient falls through the cracks", that for surgeons the company "makes sure the record is complete, verified, and current", and that the surgeon "makes the call with verified, complete data"; the home page promises complete, current data at every decision point. Those are claims about what is absent, which no review of the extracted findings can show, so they grade here whatever the review design: a surgeon who believes the record is complete will not look for the outside result the assistant never found.

The approach is described in general terms and nothing is identified. The product page describes three steps: gathering outside records by monitoring the health systems a patient touches, extracting the clinically relevant findings from unstructured records, and automating updates and reports written back as structured data. The company has a chief AI officer, and the home page cites more than 60 healthcare AI patents and 200 peer reviewed papers across the team's careers. No model, architecture, training data, evaluation method or version practice for the extraction is named on the home, product, about or news pages.

Nothing says which models read the records, whether any is a licensed foundation model, what they were trained or tuned on, or how a program learns that a model changed. The team's publications describe earlier chronic disease models, not this one.

The chain is sketched, and every party in it goes unnamed. The product page says patient data is hosted on a HITRUST certified healthcare cloud, that the assistant gathers records from outside health systems, extracts findings with its models and writes them back to the program's record. The cloud provider, any network used to retrieve outside records, and whether the extraction runs on the company's own models or a licensed model are not named anywhere on the site.

When the company sold chronic disease models, trained and run inside each health system, the chain had no transfer at all. The transplant assistant moves outside records into a hosted cloud, so a transfer now occurs, and who receives it is not stated.

No named deployment of the transplant assistant and no figure that claims to have been measured. The home page headlines 30 percent faster time to the waitlist, twice the referral throughput and a 30 percent higher active waitlist rate, and the product page labels its figures potential impact: up to 44 percent coordinator efficiency, up to twice as many patients to committee, up to 30 percent more patients active and 30 minutes saved per offer decision. None is attributed to a customer, period or baseline, and the only customer voice is an unnamed transplant coordinator.

The team's evidence belongs to earlier products. The news page lists peer reviewed and conference work on finding undiagnosed chronic kidney disease and on severe uncontrolled asthma, including results from a collaboration with Hackensack Meridian Health presented at a thoracic society meeting in 2025, and those do not test an administrative assistant for transplant teams.

General assurances, and the earlier on site commitment no longer describes the product. The product page says all data is hosted on a HIPAA compliant, HITRUST certified healthcare cloud, that the company handles compliance and security updates, and that the platform keeps zero local PHI, with single sign on, role based access and full audit logging. The company's earlier promise that no patient information ever leaves the hospital system belonged to the chronic disease product and is contradicted by the current hosted design, in which outside records and extracted findings are held in the company's cloud.

No retention period is stated for gathered outside records or extracted findings, nothing says whether patient records train the extraction models, no deidentification step is described, and the site has no privacy policy. The assistant also monitors outside health systems a patient has touched, so how those records are obtained and under what authority matters as much as how they are held.

Compliance is claimed, and no privacy document exists on the site. The product page describes the service as running on a HIPAA compliant, HITRUST certified healthcare cloud and says the company handles all compliance and security updates, with zero local PHI. The site links no privacy policy, terms or business associate material from any page, and the compliance phrase is attached to the hosting environment rather than stated as the company's own business associate status.

The claim stands unsupported, and no agreement, subcontractor flow down or breach timetable is published. The assistant gathers records from outside health systems a patient has touched and holds them in the company's cloud, which is business associate work for every transplant program it serves, and the patients whose outside records are retrieved are not party to that agreement.

Controls are asserted, and the one certification named belongs to the hosting cloud. The site footer carries a "HITRUST Certified" badge with no holder stated, but the product page's security section says what is certified: all data is hosted on a HIPAA compliant, HITRUST certified healthcare cloud. On the more specific page the certification attaches to the host, which is the host's attestation rather than the company's, and Carenostics has announced no HITRUST certification of its own. Single sign on, role based access control and full audit logging are described with no independent check behind them.

The company publishes no certification or report in its own name, such as SOC 2, HITRUST or ISO 27001, with its scope and date. Software that gathers a transplant candidate's records from every outside health system and holds them in a cloud is a high trust position.

No device claim is made, and the product is scoped as administrative, with one marketed function at the edge. The site's line is "Separate the clinical from the clerical", and most of what the assistant does is clerical: gathering outside records, extracting findings for review with source links, filing updates, assembling committee reports and tracking checklists. The product page also says post transplant monitoring "surfaces complications before the next appointment" and that alerts fire when action is needed. Surfacing a complication already documented at an outside system, with the source linked, is a reviewable basis; what the assistant counts as a complication and whether it ranks or prioritizes patients by it is not stated, and no regulatory position appears on the home, product, about or news pages.

The rules that govern this work are transplant program requirements (organ allocation policy and waitlist management obligations) and HIPAA for the records gathered.

Nothing is published on how the transplant assistant's models are governed or tested, and the team's research sits on earlier products. The news page lists peer reviewed and conference work by the team on finding undiagnosed chronic kidney disease, on severe uncontrolled asthma and on predicting retention in opioid use disorder treatment, and the home page cites more than 200 peer reviewed papers across the team's careers. None of it evaluates the current product, and no governance framework, validation process or subgroup result is published on the home, product, about or news pages.

The exposures follow from what the assistant does. It decides which findings in a 300 page outside file are clinically relevant and which are left out, for patients whose access to an organ turns on the completeness of that file; performance can differ by the outside system a record came from, by document quality and by the patient populations those systems serve, and none of that is measured on anything published.

Mechanisms exist to check an output against its source, and nothing stands behind it. The product page says extracted findings go to physicians and surgeons for review, that selection committee reports carry source links every reviewer can verify, and that the platform keeps comprehensive audit trails. That is how an output can be challenged: a reviewer can trace a finding back to the document it came from.

No warranty, indemnity, accuracy commitment or remediation obligation appears on the home, product, about or news pages, and the site links no terms of service or privacy policy at all. No error or omission rate is published. A finding missed from an outside record, and so absent from a committee packet or an organ offer decision, leaves no source link to follow, which is where recourse would matter most.

Integration is claimed through standards, with no record system named. The product page says the assistant is built for record system workflows on SMART on FHIR and HL7, integrates natively with major record systems with no custom interface builds, and writes everything back as structured data; the footer carries a "Major EHR partner" badge without naming the partner. It also monitors the outside health systems a patient has touched and gathers their records, without saying through which network or exchange.

Nothing on the site names a record system, health information exchange or record retrieval network, and no deployment is tied to an integration. The one deployment the company once described, at a single health network, belonged to the chronic disease product, and the current site no longer mentions it.

A single hosted service with its location implied. The product page says all data is hosted on a HITRUST certified healthcare cloud, that the company handles compliance and security updates, and that the platform keeps zero local PHI, deploying alongside the program's existing tools and writing back to the record. The cloud provider and region are not named on the home, product, about or news pages.

The chronic disease product was described as trained and run inside each health system with no patient information leaving it; the transplant assistant holds gathered outside records and extracted findings in the company's cloud, so residency, tenancy and what happens to those records at contract end are live questions again.

Pricing stays private, but the business case shows how the product is sold and roughly what it replaces. Carenostics sells to transplant programs, and the product page frames the cost against staffing: capacity equivalent to three coordinators for the cost of one, 20 percent more transplants a year from a larger active waitlist, and pre transplant administrative infrastructure described as reimbursable through the Medicare Cost Report, which the company says offsets a meaningful portion of contract cost for most programs. The route in is a 45 minute demo.

Neither the price nor the unit of charge is stated, whether per program, per patient on the list or per coordinator seat. The last announced financing was a $5 million seed round in September 2023, and the company has since changed its product focus.

The product is healthcare administration automation for transplant coordinators: an administrative assistant for referral and records work across referral, evaluation, listing and waitlist management.

Stated clearly for one specialty, with nothing validating it yet. The product serves solid organ transplant programs across the patient journey: referral, evaluation, listing and selection committee, waitlist management, organ acceptance and post transplant monitoring, for coordinators, transplant physicians and surgeons. The setting is the transplant center within a health system, and the market is the United States. The company describes itself as built for transplant rather than a general AI company.

The site names no transplant program as a customer, the only customer voice is an unnamed coordinator, and the impact figures are labeled potential. The earlier chronic disease case finding (kidney disease, asthma, lung disease and heart failure) survives only as the team's published research, not as a product the site sells.