Cognivue

A learned add-on sits on a product whose value stands without it. The Clarity and Thrive tests run on adaptive psychophysics: the device adjusts each exercise to the patient's visual and motor responses across ten subtests (motor control, visual salience, letter, word, shape and motion discrimination and memory) and reports a composite score against age-normed cutoffs.

That is adaptive testing with fixed rules, not a model whose parameters are estimated from data, and the company does not describe it as machine learning. The learned component is the Cognivue Amyloid Risk Measure (CARM), launched in May 2025 as an optional feature on Clarity: the company's release says it was developed using machine learning and combines age with three subtests into a four point estimate of the likelihood of brain amyloid, published in Neurology and Therapy (2025).

Remove CARM and the cognitive assessment, the report and the reimbursement path are unchanged. The company's news library also carries a 2023 partnership with DigiCARE Realized, whose record screening AI is that company's product, not Cognivue's.

The scope is drawn plainly, and the routing from a result to a person is not. Every product page states that Clarity and Thrive are adjunctive tools for evaluating cognitive function, not stand-alone diagnostic tools, that they do not identify the presence or absence of clinical diagnoses, and that results are to be interpreted by a licensed clinician.

The test is self-administered, and the company's papers publish the cutoffs (66 on the global score in Bio-Hermes; age specific cut points in FOCUS). What is missing is the escalation rule for the settings the company sells into without a physician: Thrive is marketed to pharmacies and hearing care practices as a screen for non-physicians, and no page says what score sends a patient to a physician or how a missed impairment is caught.

The CARM release says the measure helps clinicians distinguish normal aging, preclinical Alzheimer's disease, Alzheimer's related impairment and decline from other causes, which sits uneasily beside the statement that the device does not identify clinical diagnoses.

The approach is named and the scoring method is described in the company's papers; update practice is not published. The technology page and device pages describe adaptive psychophysics that adjusts each exercise to the patient's visual and motor responses, name the ten subtests across memory, executive function and attention, discrimination and visuospatial domains, and describe the CogniWheel input. The papers publish the global score's cutoffs and age norms.

CARM is described as developed with machine learning from age and three named subtests, scored in four categories, in the company's release and in Neurology and Therapy (2025). Device generations are identified (Clarity and Thrive Model 2 in the FDA listing, Gen5 in the security document), but nothing says what changed between them, whether scoring changed, or how customers are told when it does.

The chain is structurally short. Scoring, including CARM, runs on the device from the company's own algorithms, and patient data is stored in a local database on the device rather than sent to a model provider. The one outside party named is Amazon Web Services, whose IoT service the company uses to manage and monitor devices remotely (Gen5 security features document, RES-002-A, 2025). What that channel carries, whether the customer portal holds test data, and who hosts it are not stated, and no subprocessor list is published.

Peer reviewed, prospective multi site studies with the method readable, and the accuracy they report is moderate. In Bio-Hermes (887 participants across US sites, organized by the Global Alzheimer's Platform Foundation), Galvin et al. (Scientific Reports, 2024) report sensitivity 73.9 percent and specificity 58.6 percent for cognitive impairment at a cutoff of 66, with a positive likelihood ratio of 2.17 in biomarker confirmed groups; a second analysis (Journal of Alzheimer's Disease, 2025; 505 participants) separates amyloid positive from negative and preclinical Alzheimer's disease from true controls, with correlations of r = -0.43 against amyloid PET and -0.40 against plasma p-tau217; and the CARM paper (Neurology and Therapy, 2025) reports an odds ratio of 3.67 for amyloid PET positivity in the high CARM categories.

FOCUS (452 participants, six sites) compares Clarity with the RBANS by age band (AUC 0.695 to 0.750). The authors are the company's Chief Scientific Officer, university colleagues and Cognivue staff. A competitor's analysis of the same Bio-Hermes cohort (Jannati et al., Alzheimer's Research and Therapy, 2025) reported Cognivue at AUC 0.70 for amyloid status, and the company's chief executive published a Matters Arising on that comparison's method (2026). The home page also cites a paper finding Cognivue as effective as the MoCA with better test-retest reliability.

What is collected and where it stays are stated specifically; how long it stays and what else it is used for are not. The Gen5 security features document (RES-002-A, 2025) lists the fields Clarity and Thrive store per test: name, date of birth, sex, subtest and domain scores and, optionally, email and phone number.

It says they are kept in a local database on the device, reachable only by on-device processes, under full disk AES-256 encryption, with administrators alone able to open older reports. The earlier document (RES-001-A) states plainly that hiding reports from before a chosen date does not delete the underlying results. The device is managed remotely through an AWS IoT service, and the reach of that channel into stored data is not described.

No retention period, deletion procedure or end of contract disposition is published, and nothing says whether test data from customer devices is used to develop CARM or other scores; the published CARM work used research cohort data. Nothing is published on safety engineering, such as how a scoring fault is detected and handled.

A compliance claim with no business associate position. The Gen5 security features document (RES-002-A, 2025) refers to the company's responsibility to comply with HIPAA regulations and lists the protected health information each device stores. No business associate agreement, template or statement of business associate status is published. The privacy policy covers website visitors only and carries an unfilled effective date.

Because the devices hold patient records on site while the company manages them remotely through an AWS IoT service, whether the company acts as a business associate is a live question the published material does not answer.

Controls are described specifically, with nothing independent behind them. The Gen5 security features document (RES-002-A, 2025) lists full disk AES-256 encryption, HTTPS for all traffic, automatic operating system security updates, ClamAV antivirus, restricted access to the operating system, role separated report access, automatic logout after 30 minutes and password rules, with devices managed remotely through an AWS IoT service.

The earlier document (RES-001-A, 2023) named Ubuntu 18.04 LTS as the operating system. No SOC 2 report, ISO/IEC 27001 or HITRUST certification, penetration test or trust center is published on the site, in the news library or in either document. Because Clarity and Thrive are listed as exempt from premarket notification, no premarket submission, and so no FDA cybersecurity review, stands behind the current devices either.

A De Novo authorization is named, and the devices sold today are later versions listed as exempt under the classification it created. FDA granted DEN130033 to the original Cognivue device, creating the computerized cognitive assessment aid classification (21 CFR 882.1470, product code PKQ).

The FDA registration and listing database shows Cognivue, Inc. (registration 3014389576) listing Cognivue Clarity, Cognivue Thrive and their Model 2 versions under product code PTY, computerized cognitive assessment aid, exempt, Class II; no 510(k) is held for them.

The site calls Clarity and Thrive tests based on FDA-cleared technology and states on every product page that they are adjunctive tools for evaluating cognitive function, not stand-alone diagnostic tools, and do not identify the presence or absence of clinical diagnoses.

The Cognivue Amyloid Risk Measure, a machine learning estimate of amyloid likelihood added to Clarity in 2025, sits outside that indication, and no page states its regulatory position; its release says it helps clinicians distinguish Alzheimer's related impairment from other causes.

Per group results are published with numbers, and they show where the test is weaker. In the Bio-Hermes analysis (Galvin et al., Scientific Reports, 2024; 887 participants), Clarity scores are reported for impaired and unimpaired participants within each group: non-Hispanic White (n=699) 71.7 against 57.6, Black (n=87) 63.6 against 56.7 (p=0.023), Hispanic (n=85) 71.5 against 57.8, and by education, age band, sex and APOE e4 status.

The authors state that unimpaired Black participants scored lower than White and Hispanic participants, so the gap between impaired and unimpaired is narrowest in that group, and that scores differ by age and education but not by sex.

The FOCUS study (Galvin et al., Journal of Alzheimer's Disease, 2024; 452 participants at six US sites, 31 percent African American, 46 percent with 12 years of schooling or less) reports performance against the RBANS by age band: AUC 0.695, 0.716 and 0.750, with sensitivity 65.1, 65.8 and 88.4 percent and specificity 65.1, 72.4 and 57.1 percent at cut points of 86, 81 and 68, and race and education effects similar to those of the RBANS. James Galvin, first author of both, is the company's Chief Scientific Officer, and Cognivue staff are coauthors. No subgroup results are published for the CARM amyloid estimate.

A published error rate with its method and denominator, cited on the company's own clinical data page, and nothing standing behind the output. Galvin et al. (Scientific Reports, 2024), led by the company's Chief Scientific Officer with Cognivue staff as coauthors, report Clarity in 887 Bio-Hermes participants (388 cognitively normal, 282 with mild cognitive impairment, 217 with probable Alzheimer's disease): at a cutoff of 66, sensitivity 73.9 percent, specificity 58.6 percent, positive predictive value 71.7 percent and negative predictive value 62.7 percent against clinical diagnosis, and a positive likelihood ratio of 2.17 and negative likelihood ratio of 0.29 in the biomarker confirmed groups.

These are study results, not performance in use, and the specificity means roughly four in ten unimpaired people screen positive. No warranty, indemnity or remediation commitment attached to a result is published, and no route for a patient or clinician to challenge one.

A standards route is described with no system named. The earlier security features document (RES-001-A, 2023) says an administrator can configure the device to create an HL7 message holding the report, or set up a password protected network file share to reach the report folder. The current Gen5 document (RES-002-A, 2025) mentions only printing, by USB or network.

No record system, interface engine or production deployment is named on the site, in the news library or in either document, so the integration cannot be verified. The device produces a one page report at the point of care.

A single on-premises option with the data location committed and the remote path left open. Clarity and Thrive are dedicated devices placed in the customer's office, and the Gen5 security features document (RES-002-A, 2025) says patient information is stored in a local database on the device, reachable only by on-device processes, under full disk encryption. Reports print to a USB or network printer.

What is left open is the processing path: the device is managed and monitored through an AWS IoT service over HTTPS, and the company runs a customer portal, and neither document says what data those channels carry or where it rests. No remote or at-home administration route is offered on the product pages.

No device, license or per-test price is published; how the product is paid for is discoverable. The reimbursement page, with a downloadable Clarity reimbursement guide and flashcard, explains that insurers pay the provider for test administration under the neuropsychological testing code the AMA created in 2019 and for the clinician's interpretation, sets out the medical necessity documentation payers ask for, and lists the circumstances in which claims are denied, such as screening when impairment is not suspected.

Those are what the provider recovers, not what the company charges. The products are dedicated devices, and pharmacy buyers are offered an integration program with each device, so the cost to size is the device plus any service terms.

Settings are named, with validation behind part of them. Clarity, the ten minute assessment, is sold to physicians in neurology, geriatrics and primary care; Thrive, the five minute screen, to pharmacies and hearing care practices for use by non-physicians; and CARM to support trial enrichment and treatment screening. Clarity is also sold for early detection of cognitive impairment and, with CARM, for estimating amyloid likelihood.

Validation covers research cohorts aged about 59 to 85 (Bio-Hermes) and 18 to 85 (FOCUS, six sites). The pharmacy and hearing care settings, where the result is read by a non-physician, carry implementation posters rather than published performance.