ATTNhealth
Request deck Join the pilot
Join the clinician pilot Request the investor deck

Science · Last updated October 10, 2026

Every measure, explained. Every limit, stated.

What ATTNhealth measures, how each number is computed, how we plan to validate it, and what we don't know yet. Written for clinicians, with references.

In developmentNot FDA clearedNot for clinical use

Four cognitive functions, measured through performance.

Rating scales ask how attention feels. These tasks are designed to measure how it performs, trial by trial, in milliseconds.

Sustained attention

Task
Continuous performance test (CPT). Letters or shapes appear one at a time for several minutes. The person responds to targets and holds back on everything else.
What it shows
Whether attention holds over minutes, not seconds. Missed targets (omissions) are the main sign of lapses. The CPT has been studied since 1956.

SourceRosvold et al., Journal of Consulting Psychology, 1956

Response inhibition

Task
The same CPT, scored for responses to non-targets (commissions) and for how responses change when targets are frequent or rare.
What it shows
How well the person stops an action that has become automatic.

Working memory

Task
Short tasks that require holding and updating information for a few seconds, such as deciding whether the current item matches one shown two steps earlier.
What it shows
How much a person can keep in mind while using it. Working memory is commonly studied in both ADHD and early cognitive decline.

Processing speed

Task
Simple, speeded decisions where accuracy is easy and speed is the question.
What it shows
How quickly simple information gets processed. It gives context for every other score: slow and accurate is different from fast and careless.

The cognitive-decline battery will add memory tasks. Task selection is in progress.

Beyond right and wrong.

Two people can make the same number of errors for different reasons. These metrics help separate them.

Signal detection: d′ and criterion

  • d′ (d-prime) is sensitivity: how well someone tells targets from non-targets. Higher means sharper discrimination.
  • Criterion is response bias: how readily someone presses. A cautious responder misses targets. A trigger-happy responder hits non-targets.
  • Why it matters: Error counts mix ability and style. d′ and criterion pull them apart.

Response time, beyond the average

  • Intra-individual variability is how much response times swing from trial to trial. Across 319 studies, children, adolescents, and adults with ADHD showed more variability than typically developing peers.
  • Ex-Gaussian modeling splits a person's response-time distribution into a typical-speed part (μ, σ) and a slow tail (τ). The tail captures occasional very slow responses, a plausible signature of attention lapses. A 2023 meta-analysis found τ and σ were generally larger in ADHD samples.
  • Why it matters: An average can hide the pattern. Someone can be fast most of the time and drift off every so often. The average won't show that. The distribution will.

SourcesKofler et al., Clinical Psychology Review, 2013 · Neuropsychology Review, 2023 (ex-Gaussian meta-analysis)

First question: was the effort real?

A score only means something if the person was engaged. In one study of 392 adults referred for ADHD evaluation, 16% failed performance validity testing. At home, there are more ways for a session to go wrong: a phone, a sibling, a slow laptop.

SourceOvsiew et al., The Clinical Neuropsychologist, 2023

So every ATTNhealth session is designed to check, before any score is shown:

  • Effort. Embedded indicators that flag implausibly poor or inconsistent performance.
  • Engagement. Patterns such as runs of identical responses, long gaps, or leaving the browser tab.
  • Device timing. Whether the computer recorded response times precisely enough to trust.

If a session fails, the report says so on page one and holds back domain scores.

Norms are earned, not borrowed.

A score means little without a reference group. We don't have ATTNhealth norms yet. Until we do, no ATTNhealth score will be presented as normative.

Our plan

  • Collect reference data from a sample that spans age, sex, and education, recruited under an IRB-approved protocol
  • Publish the sample size and demographics behind every norm we use
  • Compare patients with their own baseline as well as with the reference group, for repeat testing

Rules, models, and what neither does.

Will do (planned)

  • Rules, not AI. Set working-memory difficulty and practice trials by fixed, published rules inside a standardized protocol. The core attention task runs a fixed sequence so scores can be compared with norms
  • Models, later. Flag disengagement patterns for validity review, and draft plain-language interpretive notes that a clinician reviews, edits, and signs. Any model that shapes the report would be part of the regulated device and reviewed by FDA before clinical use

Doesn't

  • Diagnose, or output a probability that someone has a condition
  • Make clinical decisions
  • Replace scoring that can be done with transparent, published formulas. d′, criterion, and ex-Gaussian parameters are computed with standard methods.

Not validated yet. Here's the plan.

ATTNhealth is pre-validation. Below is the sequence of studies, in order. Each has a status chip. We'll add results only once they're published.

  1. Step 1In progress

    Task build and prototype

    Core attention, working-memory, and processing-speed tasks running in the browser.

  2. Step 2Planned

    Technical verification

    Response-time precision across common browsers, operating systems, and hardware.

  3. Step 3Planned

    Reliability

    Test–retest consistency in healthy volunteers.

  4. Step 4Planned

    Clinical validation (ADHD)

    An IRB-approved study in adults referred for ADHD evaluation, comparing ATTNhealth with established measures and expert clinical diagnosis, with the analysis plan set in advance.

  5. Step 5Planned

    Normative reference data

    Age-stratified sample, published with demographics.

  6. Step 6Planned

    Clinical validation (cognitive decline)

    A parallel study in older adults with memory concerns.

Regulatory and trust

What's done, and what's still a goal.

Current status

ATTNhealth is in development. It is not FDA cleared or approved and is not available for clinical use.

"Today" lists what is true now. "Target" lists goals that have not been achieved.
AreaToday (achieved)Target (not yet achieved)
FDANot clearedNot cleared or approved. No submission made.Target510(k) clearance for the ADHD battery (planned pathway; FDA pre-submission first)
Clinical validationNot yet doneTargetIRB-approved studies, published
Patient data securityThe clinical product handles no patient data yetTargetBuilt toward HIPAA requirements
Independent security auditNone yetTargetSOC 2 Type 2
This websiteIn placeCollects only form entries. No patient health information.—

The intended FDA pathway

The FDA has a device type that appears to fit this kind of product: "recorder, attention task performance," product code LQD. The classification database still lists it as unclassified (pre-amendment). Computerized attention tests have been cleared under it through the 510(k) pathway since at least 2004. 510(k) is our planned pathway for the ADHD battery, with a cleared device of this type as the predicate. Before submitting, we plan an FDA pre-submission meeting to confirm the device type, the predicate, and the data FDA expects. Our device would not measure motor activity, so its indications may be narrower than those of camera-based predicates. Clearance isn't guaranteed, and timing depends on our validation data. The cognitive-decline battery may follow a different pathway. We'll decide that with regulatory counsel.

SourcesFDA Product Classification database, product code LQD · FDA 510(k) K040894 (2004)

Research directions, not product claims.

Cognition and genetics.

Over time, and only in consented research, we want to understand how cognitive measures relate to genetic risk.

Voice.

Researchers are studying whether speech carries information about attention and cognition. We're watching that work. Nothing is built.

More indications.

Traumatic brain injury, post-COVID cognitive symptoms, pre-surgical baselines, and learning disorders, each only after the first indications are validated.

What we don't know yet.

These are the questions our validation has to answer. We'd rather you hear them from us.

  • These aren't biomarkers. Under the FDA–NIH BEST glossary, a biomarker is a measured indicator of a biological process, and "a biomarker is not a measure of how an individual feels, functions, or survives." Task performance is how someone functions, so our measures are performance-based assessments, not biomarkers. None is a validated marker of ADHD, cognitive impairment, or dementia.
  • No single test diagnoses ADHD. Performance on attention tasks overlaps between people with and without ADHD. That's why the report shows several measures and never a verdict.
  • Variability isn't specific to ADHD. In the 319-study meta-analysis, adolescents and adults with ADHD were indistinguishable from other clinical groups on response-time variability.
  • The slow-tail evidence may be inflated. The 2023 ex-Gaussian meta-analysis found larger τ in ADHD, but its authors judged publication bias in τ probable, because τ is often the only parameter reported. Its formal funnel-plot test for τ was not significant.
  • Browsers aren't lab equipment. Timing precision varies across devices. We have to show ours is good enough, device by device.
  • Home isn't a quiet room. Validity checks reduce the noise of testing at home. They don't remove it.
  • Practice effects. Repeat testing can improve scores for reasons that have nothing to do with attention. Change measures must account for that.
  • Who the norms represent. Norms built on a narrow sample can mislead for everyone else. Ours will need breadth in age, education, and language. At launch, the battery will likely be English-only.
  • Regulation can change. In June 2021, an FDA advisory panel agreed this device type should be classified as Class II with special controls. Final requirements may change before we submit.

SourcesKofler et al., 2013 · Neuropsychology Review, 2023 · FDA Neurological Devices Panel materials, 2021 · FDA–NIH BEST glossary

References

  1. Rosvold HE, Mirsky AF, Sarason I, Bransome ED Jr, Beck LH. A continuous performance test of brain damage. J Consult Psychol 1956;20(5):343–350. https://doi.org/10.1037/h0043220
  2. Kofler MJ, Rapport MD, Sarver DE, et al. Reaction time variability in ADHD: a meta-analytic review of 319 studies. Clin Psychol Rev 2013;33(6):795–811. https://pubmed.ncbi.nlm.nih.gov/23872284/
  3. Bella-Fernández M, Martin-Moratinos M, Li C, Wang P, Blasco-Fontecilla H. Differences in ex-Gaussian parameters from response time distributions between individuals with and without ADHD: a meta-analysis. Neuropsychol Rev 2024;34:320–337 (online 2023). https://link.springer.com/article/10.1007/s11065-023-09587-2
  4. Ovsiew GP, Cerny BM, De Boer AB, et al. Performance and symptom validity assessment in ADHD: base rates of invalidity, concordance, and relative impact on cognitive performance. Clin Neuropsychol 2023;37:1498–1515. https://pubmed.ncbi.nlm.nih.gov/36594201/
  5. U.S. FDA. Product Classification: recorder, attention task performance (LQD). accessdata.fda.gov
  6. U.S. FDA. 510(k) premarket notification K040894 (2004). accessdata.fda.gov
  7. U.S. FDA. Neurological Devices Panel, June 4, 2021: classification of attention task performance recorders (LQD). Meeting page · 24-hour summary
  8. FDA-NIH Biomarker Working Group. BEST (Biomarkers, EndpointS, and other Tools) Resource: Glossary. https://www.ncbi.nlm.nih.gov/books/NBK338448/