Cognivue

A genuine middle case between BrainCheck and Linus Health, both indexed. Adaptive psychophysics is real computational adaptation: the test dynamically adjusts as the patient progresses, calibrating to that individual's visual and motor abilities, and collects over 130,000 data points to produce the score. That is substantively more than digitising a paper instrument, which is why this is not a C alongside BrainCheck.

But it is an adaptive testing algorithm rather than a machine learning model deriving novel signal from behaviour, which is why it is not an A alongside Linus Health's analysis of stylus dynamics and speech acoustics. The company describes proprietary technology and a proprietary algorithm without claiming machine learning, which is accurate rather than inflationary and is worth noting in a category where AI language is applied loosely.

A on candour and scope discipline, the MD-Staff precedent. The company states plainly that Clarity and Thrive are indicated as ADJUNCTIVE tools for evaluating cognitive function, that they are not stand-alone diagnostic tools, and that they do not identify the presence or absence of clinical conditions. Placed in product copy rather than buried.

There is an interesting structural inversion worth noting: the test is fully self-administered by the patient with no clinician proctor, which is MORE autonomous in administration than competitors, while the interpretation claim is more tightly constrained. Automating the administration and explicitly not automating the diagnosis is a coherent and well drawn boundary.

The mechanism is named specifically rather than behind an accuracy claim: adaptive psychophysics targeting cortical information processing, dynamic adaptation during the test, over 130,000 data points, and a single averaged output score derived from four perception and four memory subscores, with that output structure documented in the De Novo review file. Regulatory documentation being publicly retrievable is itself a transparency advantage. Graded B rather than A because the proprietary algorithm converting those data points into the clinical score is not described, and there is no published model card.

Three things here are better than the category norm and one of them is rare enough to name as a benchmark. The company publishes a field level inventory of exactly what the device collects on each test, naming the elements individually rather than assuring a reader that data is safeguarded, and a precise list is far more useful to a privacy office than a general assurance because it can be assessed against a policy line by line.

The architecture favours the buyer: information is stored on the device in a local database reachable only by on device processes, access is separated by role with administrators alone able to open stored reports, and reports leave the device only if an administrator configures a file share or enables messaging, both customer decisions rather than defaults. The benchmark item is the candour about the limits of its own feature.

The company states plainly that hiding reports administered before a chosen date does not delete the underlying database results; it removes the report files and conceals the session from the interface. Publishing the boundary of a feature a buyer would otherwise read as deletion is exactly the behaviour this index credits, and it is why the rest of the record can be read at face value.

Held below the top grade because no retention period, deletion procedure or end of contract disposition is published, and the one mechanism resembling deletion is disclosed as not being deletion. Ask how records are permanently removed, and what happens when a device is serviced or retired.

Strong, independent, and notably focused on the right question for this instrument class. A peer reviewed validation published in Neurological Sciences and Neurosurgery reported Cognivue as equally effective as the Montreal Cognitive Assessment with SUPERIOR test-retest reliability, which is the metric that actually matters for an instrument intended to establish a baseline and monitor change over time.

Notably the company reports the correlation with MoCA honestly at r = 0.38, a modest figure it could easily have omitted or dressed up. The FOCUS study enrolled 452 participants across six US study sites, validating against RBANS, a gold standard battery used in both clinical practice and Alzheimer's trials, and separately established age-normed scores and measured practice effects. Original clearance rested on a large-scale 2014 comparison against standard neuropsychological tests. Evidence spans validity, reliability, normative range development and diverse population performance.

Cognivue publishes a field level inventory of exactly what the device collects on each test: patient first name, middle initial and last name, date of birth, gender, subtest and domain scores, and optionally an email address. Very few vendors assessed here enumerate the data elements at all, and a precise list is more useful to a privacy office than a general assurance about safeguarding.

The architecture favours the buyer. Information is stored on the device itself in a local database reachable only by on device processes, so in ordinary operation patient records stay inside the customer's facility rather than moving to vendor infrastructure. Access is separated by role, with administrators alone able to open stored reports and operators limited to the report generated immediately after a test. Reports leave the device only if an administrator configures a network file share or enables an HL7 message, and both are customer decisions rather than defaults.

The candour about the limits of its own deletion feature is creditable and uncommon. Cognivue states plainly that hiding reports administered before a chosen date does not delete the underlying database results; it removes the report files and conceals the session from the interface. Publishing the boundary of a feature a buyer might otherwise read as deletion is the behaviour this index credits, and it is the reason the record can be read at face value.

What is missing is the other half. No retention period, deletion procedure or end of contract data disposition is published, and the one mechanism resembling deletion is disclosed as not being deletion. Nothing addresses whether test data is used for any secondary purpose, including model development. Buyers should ask how patient records are permanently removed from the device database, what happens to that data when a device is returned, serviced or retired, and what the vendor can reach through the remote management channel.

No business associate agreement, template or summary of terms could be located. The device security document refers to HIPAA regulations as part of the company's compliance responsibility alongside FDA guidance on medical device cybersecurity, but that is a statement of intent rather than a description of the instrument. The company's published privacy policy addresses website visitors only and contains no HIPAA section at all.

The architecture makes this a live question rather than an academic one. Cognivue Clarity and Thrive hold patient information locally on the device inside the customer's facility, which in ordinary operation keeps the vendor out of the data path and might suggest no agreement is needed. But the same document states that the devices are managed and monitored remotely by the vendor through AWS Systems Manager, which is the kind of arrangement that tends to create business associate status. Nothing published resolves whether Cognivue considers itself a business associate, and if so on what terms.

Buyers should ask directly whether Cognivue will execute a business associate agreement, request the template before contracting, and ask specifically whether it covers the remote management channel as well as the device. A provider deploying these devices should not assume that on device storage removes the need for one.

Cognivue publishes a numbered device security document describing the controls on Cognivue Clarity and Thrive. The controls named are specific and checkable rather than atmospheric: an on device firewall blocking all incoming and outgoing connections except those required for the report file share, HTTPS for all network traffic, access to the underlying operating system restricted to network and printer configuration, automatic operating system security updates, ClamAV antivirus running on the device, a password protected folder share whose password the administrator can change, and configurable password expiration. Role separation is enforced in the product rather than by policy: device administrators alone can view stored reports, while operator users can see a report only immediately after the test concludes and cannot open older ones. The devices are managed and monitored remotely through AWS Systems Manager over HTTPS.

Held below the top band because no independent assessment exists. No SOC 2, HITRUST, ISO 27001 or other third party attestation was located, and there is no trust centre or security portal. The document is the entire disclosure, and it is the vendor describing its own controls.

Two points a buyer should raise. The document names Ubuntu 18.04 LTS as the device operating system. Standard support for that release has ended, and while extended maintenance is available, the document itself dates from 2023, so a buyer should ask what operating system baseline currently ships, how it is patched, and how long it will be supported. Second, remote management through AWS Systems Manager gives the vendor a channel into a device holding patient information, and the scope of that access is not bounded in the document. Ask what vendor staff can reach, under what authorisation, and whether that access is logged and reviewable by the customer.

The strongest regulatory position in this category and a category-creating one. Cognivue holds a De Novo classification, DEN130033, with the review documentation publicly available, validated against the St. Louis University Mental Status examination. A De Novo is not a 510(k): it establishes a new device classification where no predicate exists, meaning Cognivue created the regulatory pathway that subsequent computerized cognitive assessment devices were cleared into.

This is the same distinction that earned Viz.ai an A for the first ever De Novo in computer aided triage, and it resolves an ambiguity across this whole category, where several vendors use cleared, registered and listed interchangeably. Both Clarity and Thrive operate under the same cleared technology, and the indicated use as an adjunctive tool is stated precisely.

The FOCUS study is among the most deliberate bias investigations by any vendor in the index, and it was designed for that purpose rather than reporting demographics incidentally. The published rationale states the problem directly: most cognitive instruments do not perform well in diverse samples of older adults, and because those same instruments gate clinical trial eligibility, they make recruiting minoritized adults into Alzheimer's studies harder.

The paper quantifies the inequity it is responding to, noting that women are 50.4 percent of the US population but only 41.2 percent of clinical trial participants, and that 63 percent of US residents lack a bachelor's degree while 55.8 percent of trial participants hold a college or advanced degree. The study then enrolled 452 participants across six sites stratified by age, education, sex, race and ethnicity to calculate age-normed scores and assess performance across those strata. A vendor publishing the structural inequity in its own field and then testing its instrument against it is the benchmark this axis should be graded against.

The mechanism is named specifically rather than behind an accuracy claim, covering adaptive psychophysics targeting cortical information processing, dynamic adaptation during the test, a stated volume of data points collected, and a single averaged output score derived from four perception and four memory subscores, with that output structure documented in a publicly retrievable regulatory review file. Two things follow that are worth crediting.

The decomposition tells a clinician the headline score is an average over eight components rather than a single measurement, which means two patients can reach the same number by different routes and a clinician who knows that will ask to see the subscores. And regulatory documentation being publicly retrievable is a transparency advantage in itself, since it gives an outside party an authoritative description the vendor did not write.

Held at C because the conversion is undescribed: the proprietary algorithm turning the raw measurements into the clinical score is not published, no model card exists, and no warranty, indemnity or remediation commitment attaches. That conversion is where the clinical meaning is created, and a score whose derivation is closed cannot be interrogated when it conflicts with a clinician's own assessment of a patient they know. Ask whether subscores are reported alongside the composite, the test retest variability, and performance in patients with sensory or motor impairment.

No named EHR integrations located. The product generates an immediate one-page report at the device rather than flowing structured results into the clinical record. Against BrainCheck, which names Epic and athenahealth, Linus Health, which offers health system EHR integration, and Neurotrack, which claims integration with all EHRs, this is the weakest interoperability position in the category and reflects its origins as a standalone hardware device.

Dedicated hardware. Clarity and Thrive are physical devices rather than software on a tablet or browser, calibrated identically across units to ensure consistency. That calibration is a genuine reliability advantage and supports the superior test-retest finding, but it is also the most capital-intensive and least flexible deployment model in this category, requiring device purchase and physical placement, with no remote administration capability located. Compare Linus Health running on an iPad with a remote option, and Neurotrack claiming any device.

No device price or licence fee published, but the company states plainly that the assessment is patient-administered and reimbursable, and the 5 minute Thrive product is explicitly positioned for pharmacy as well as office settings, which signals the intended billing and distribution model.

Graded B rather than A because, unlike BrainCheck, no specific CPT codes or reimbursement guidance are published, and for a hardware product the acquisition cost is a material undisclosed variable that competitors selling software do not impose.

Two products deliberately sized for different settings, a 10 minute full assessment for clinical evaluation and a 5 minute screen for high-throughput environments including pharmacy, which extends reach beyond the physician office into retail health. Used across neurology, geriatrics and primary care, and the FOCUS validation against RBANS positions it for clinical trial screening as well. Graded B rather than A because coverage is a single clinical domain, cognition, and because the hardware requirement constrains where it can physically be deployed.