Retrieve Medical

The extraction and surfacing engine is the whole product. There is no EHR, no services business and no workflow suite underneath it, and both commercial products, the inpatient documentation integrity application and the outpatient visit preparation tool, are the same natural language processing and machine learning engine pointed at different settings. Remove the model and nothing remains.

The human gate is explicit and structural rather than implied. The company describes surfaced comorbidities being presented to the physician's screen for validation and potential inclusion if deemed necessary by the attending physician, and states plainly that the clinician decides what is relevant.

Verification is supported rather than merely permitted, because any highlighted item opens to the underlying note or result in its original context inside the EMR, which is a stronger affordance than linking to the vendor's own copy of a document. Held at B because nothing quantifies how the gate performs: no acceptance or rejection rate is published, no confidence signal or threshold routes anything, and no abstention behaviour is described for a thin record. The write back path also deserves a question, since a single action inserts data into the chart; ask what is logged about who accepted what, and whether the entry is attributable and reversible.

The technology is described only as advanced natural language processing and artificial intelligence. No model or model family is named, no accuracy figure is published, no omission or false negative measure exists, and no evaluation methodology accompanies the study the company promotes.

The one real transparency property is at the output level and it is a good one: any surfaced item can be opened in place to show the actual note or result in its original context in the EMR, so a physician checks the source system rather than the vendor's rendering of it. That establishes where a statement came from without establishing how often the engine is wrong or what it misses.

One layer is named where almost nobody names it, and one product feature raises a question this index has not had to ask before. The credit: identity verification is supplied by a named partner under an announced arrangement, and naming that layer rather than leaving it implicit is worth crediting, because proofing access to a medical record places a biometric identity company in the chain and a buyer would otherwise never think to ask what it retains. The feature is the physical card.

It is a wallet sized card carrying a code that resolves to a concise medical summary, and the stated design is that a clinician can read that summary in an emergency, which by construction means without the patient participating at the time. Deeper record access is said to require explicit consent; the summary tier is not consent gated at the moment of use, which makes the card a bearer token to a medical summary.

Establish what the summary contains, whether the code resolves to a fixed payload or a server lookup, whether access can be revoked once a card is lost, and what prevents someone who photographs the card from retaining access afterwards. Two further flows sit alongside: the clinical products write back into the permanent chart, and a consumer held account consolidates records from a nationwide exchange into a standing longitudinal copy outside any hospital's control. No retention, training or de identification position was located for any of them.

A real study at real scale measuring the wrong thing, which is a pattern in this category. The company reports analysing 10,542 patient charts and identifying 12,506 additional comorbidities and diagnoses, alongside a substantial reduction in physician queries. The chart count is a genuine denominator and larger than most vendors here can show.

But the study is vendor run and announced by press release rather than peer reviewed, no comparator or baseline is described, and the outcome measured is the number of items the software surfaced, not whether they were correct, clinically material, or accepted.

Note the precise gap, because it mirrors another record in this category from the opposite direction. Regard publishes how many of its recommendations clinicians accepted without saying how many were made, and Retrieve publishes how many were identified without saying how many were accepted. Neither permits an acceptance rate to be calculated, and that ratio is the number that matters. Ask for it.

No retention period, no statement on whether customer data is used to train or improve models, and no de identification posture was located for any product in the portfolio.

The surface this axis has to cover has grown substantially, and the questions are now harder than when the company sold one product into one medical record. Three distinct flows exist. The clinical products read the record inside Epic, Cerner or Meditech and write back into it, so the vendor authors content into the permanent chart, and nothing indicates whether an entry it generated remains identifiable as such to a clinician reading the note months later. PreviewMD optionally reaches health information exchanges, so records held by organisations other than the customer pass through the platform. And Retrieve Passport consolidates a person's records from a nationwide exchange into a consumer held account, which is a standing longitudinal copy of a complete medical history sitting outside any hospital's control.

The Passport card deserves its own question. It is a wallet sized physical card carrying a QR code that resolves to a concise medical summary, and the stated design is that a clinician can read that summary in an emergency, which by construction means without the patient participating at the time. Deeper record access is said to require explicit consent, but the summary tier is not consent gated at the moment of use. Establish what the summary contains, whether the code resolves to a fixed payload or a server lookup, whether access can be revoked once a card is lost, and what prevents someone who photographs the card from retaining access afterwards.

Identity verification is supplied by CLEAR under a partnership announced in July 2026. Naming that layer rather than leaving it implicit is credited. Ask what the identity provider retains, since proofing access to a medical record places a biometric identity company in the chain.

No HIPAA compliance statement and no business associate agreement terms were located on any retrieved surface. That absence now matters more than it did, because this company has moved onto both sides of the regime's boundary and publishes nothing that distinguishes them.

Retrieve Dx and PreviewMD operate inside a hospital's electronic medical record on the hospital's behalf, which is ordinary business associate territory, and an agreement plainly exists wherever the software is deployed even though its terms are not published. Retrieve Passport is a different arrangement. An individual directs their own records out of a nationwide health information exchange into a consumer controlled account, and a personal health record obtained that way generally sits outside the business associate framework, with the Federal Trade Commission Act and the health breach notification rule governing instead of the HIPAA Security Rule. The customer in that flow is the patient rather than a covered entity, so there is no covered entity to sign an agreement with.

The risk this creates is specific and worth naming. A health system that has executed a business associate agreement covering the clinical products may reasonably assume the relationship extends to everything the vendor does with records originating in its systems. That does not follow. Ask which regime the company treats the Passport platform as operating under, whether it considers itself subject to the health breach notification rule for that product, and whether data flowing through the consumer platform is segregated from the clinical estate or shares infrastructure with it.

No SOC 2, HITRUST, ISO 27001 or other independent attestation was located, and no trust centre or security page exists on any retrieved surface. Searches were run against the company name, against its named products and against its filing identity as a quoted company.

What moves this from an unretrieved absence to a graded finding is that the company makes an affirmative security claim without substantiating it. Launch material for Retrieve Passport states the product entered commercial production after years of development, regulatory alignment, interoperability integration and rigorous security validation. None of those phrases names a framework, an auditor, a standard or a scope. An unsubstantiated assertion that validation occurred is weaker than silence, because it invites a buyer to stop asking.

Two retrieval caveats belong on this row and a buyer should weigh the absence accordingly. The company name collides with the entire medical records retrieval sector, so general compliance queries return that industry rather than this vendor. And the company's own website disallows automated retrieval, so anything published only there was not reachable here. Ask for the attestation, its type, its scope and its period directly, and ask specifically whether it covers the consumer Passport platform as well as the clinical products, since those are different systems serving different customers.

No FDA clearance, device authorisation or clinical decision support exemption analysis was located. Graded C on that absence rather than on an assessed weakness.

The scoping question is worth raising even though the product is positioned as documentation rather than diagnosis, because the marketing describes alerting a clinician to issues that could affect medical decision making and surfacing potential risk factors and next steps. Documentation integrity and decision support are different regulatory conversations, and the product's own language reaches into the second. Ask which framework the company believes governs each surface, and to see the analysis written down.

The grade describes incentive structure and disclosure, not wrongdoing, and the counterweights are real and belong in the same paragraph. The physician validates every item and decides relevance, the underlying evidence is viewable in the original record before anything is accepted, the stated goal is documentation that is accurate and appropriate and reflects the complexity of care actually delivered, the reported reduction in physician queries is a genuine burden benefit rather than a revenue one, and the company is chaired by a past president of a major clinical society.

Against that, the revenue argument is made openly and is the company's own framing: surfacing additional comorbidities raises the Case Mix Index, that index directly affects reimbursement levels, and hospitals can anticipate increased revenue over time. Severity capture and risk adjustment are among the most litigated areas of United States healthcare compliance, and this makes a fourth product in this index whose documentation moves the payment unit at the admission level. Separately, no fairness, subgroup or demographic performance disclosure of any kind was located.

The verification affordance here is stronger than the citations credited elsewhere in this category, and the distinction is worth drawing. Any surfaced item can be opened in place to show the actual note or result in its original context in the record system, so a physician checks the source system rather than the vendor's rendering of it.

A citation tells a reader where a statement supposedly came from and still leaves them reading the vendor's presentation of it; opening the original in its own system removes the vendor from the verification path entirely, and a fabricated or misattributed item has nowhere to open to. For a summarisation product that is the strongest form of point of use checking available. Held at C because nothing measures the engine.

The technology is described only in general terms with no model or model family named, no accuracy figure, no omission or false negative measure, and no evaluation methodology accompanying the study the company promotes. Omission is the failure that matters for a summariser, because the clinician who can open every surfaced item still has no way to see what was never surfaced, and the verification affordance does nothing about it. No warranty, indemnity or remediation commitment was located. Ask for the omission rate on clinically significant content against a reviewed sample, and the methodology behind the promoted study.

Stronger than most of this category on the property that matters least often and matters here: it writes back. Most products in this lane read the record and hand the clinician a summary to act on manually. This one pulls the relevant data and enters it into the chart in the correct format and the correct place once the physician accepts it, which removes the recurring transcription cost that read only tools leave behind.

Three EHRs are named explicitly, Epic, Cerner and Meditech, and the outpatient product offers optional health information exchange access for records the institution does not hold.

Held at B because the accompanying claim to integrate seamlessly with all EHRs is unfalsifiable as written, no integration mechanism or standard is described, and no marketplace listing or third party technical review was located to corroborate the breadth.

Described as a real time cloud based application, which establishes a hosted delivery model with no customer hosted or on premise option offered. Nothing further is published: no cloud provider, no region, no residency commitment and no statement about where record content is processed relative to the hospital that holds it. Graded C because the delivery model is clear and the residency picture is entirely absent.

No price, tier or pricing mechanism is published on any retrieved surface.

One supplier continuity note belongs alongside rather than inside the commercial assessment: the parent is publicly traded, so financial disclosures are available to a buyer performing vendor due diligence, an option that does not exist with most private vendors in this category. That is recorded as a procurement fact and is not used for or against product quality.

Both major settings are covered by purpose built products rather than by one tool stretched across them. Retrieve Dx addresses inpatient documentation integrity, working in comorbid and major comorbid condition terms and Case Mix Index, while PreviewMD addresses outpatient visit preparation and produces consult requests and admission notes with ICD-10 codes paired to diagnoses.

The company reports rapid implementation with minimal IT involvement and staff training inside a single day, which lowers the adoption barrier for smaller organisations. Graded B rather than A because coverage is by setting rather than by specialty, and no specialty specific behaviour or instrument level depth was located in either product.