Digital Biomarkers for Cognitive Health: Typing, Gait, and Voice as Early Signals

A digital biomarker is a physiological or behavioral measure collected through a digital device — a smartphone, smartwatch, computer keyboard, or wearable sensor — that can serve as an indicator of health status. In cognitive health research, digital biomarkers are measures that can capture the subtle ways that cognitive function expresses itself in everyday behavior.

The intuition behind this approach is straightforward: cognitive changes that are too subtle to appear on a clinical memory test still show up in behavior. If someone's processing speed is declining, they may type more slowly or make more errors. If executive function is affected, their movement patterns may become less organized. If language processing is subtly changing, their speech patterns — hesitations, word-finding pauses, syntactic complexity — may shift before they or anyone around them notices.

Several categories of digital biomarker are being actively researched: keystroke dynamics (the timing patterns of keyboard typing), smartphone interaction patterns (scroll speed, tap accuracy, screen-on time distribution), voice and speech analysis, gait and balance measured by accelerometers, and physiological signals from wearables such as heart rate variability.

The appeal of digital biomarkers is their ecological validity — they capture behavior in the real world, not in a clinic, and they can do so continuously and passively, generating far more data than is possible through periodic clinical assessment.

Keystroke dynamics have been the most studied keyboard-based digital biomarker for cognition. A study from MIT's AgeLab demonstrated that typing speed and inter-key timing variability were sensitive to mild cognitive impairment and could track longitudinal cognitive change. A larger study using data from the Typing Biomechanics study at the University of Michigan followed adults over multiple years and found that changes in typing rhythm correlated with later neuropsychological test results.

The Apple Heart and Movement Study and related Apple-funded research have investigated smartphone and Apple Watch data as cognitive biomarkers. Features including gait speed, step symmetry, and dual-task walking (walking while performing a cognitive task) measured by the Apple Watch have shown correlations with cognitive status. Gait speed in particular is one of the most validated non-digital predictors of cognitive decline and is well-suited to wearable measurement.

Voice and speech analysis for cognitive detection has produced some of the most striking results. Research published in Frontiers in Aging Neuroscience demonstrated that automated analysis of speech — measuring pauses, phonation time, articulation rate, and semantic coherence — could distinguish early Alzheimer's patients from controls with accuracy exceeding 80%. The Boston Cookie Theft picture description task, a clinical assessment tool, has been used extensively as a stimulus for AI speech analysis studies.

The field is still establishing what constitutes a validated digital biomarker — one that has been replicated across multiple independent cohorts, shown to have sensitivity and specificity appropriate for clinical use, and demonstrated to add value beyond existing assessments. Most digital biomarkers for cognition are in earlier validation stages.

Digital cognitive monitoring tools are becoming available commercially, though it is important to distinguish validated research tools from wellness products making unsubstantiated claims. Apps and services that track reaction time, typing patterns, or simple cognitive tests are not equivalent to validated clinical assessments — but they may provide useful longitudinal signal within their own measurement system.

The most practically meaningful current application is daily cognitive testing — short, standardized tasks that provide a personal baseline and track change over time. This is the approach that Keel uses. By measuring processing speed, working memory, and other cognitive domains daily, it becomes possible to distinguish a bad day from a real trend, and to identify sustained changes worth discussing with a healthcare provider.

If you use an Apple Watch or similar device, the health data being collected — including gait metrics, heart rate variability, and activity patterns — is increasingly part of what researchers are studying as cognitive biomarkers. Sharing this data with your healthcare provider if you have cognitive concerns is a reasonable step as this field develops.

The key limitation to understand is that digital biomarkers are correlational — they can flag that something may have changed, but they are not diagnostic. A change in typing speed could reflect cognitive change, but it could also reflect arthritic finger pain, a new medication, fatigue, or a new keyboard. Interpreting digital biomarker changes requires clinical context.

The promise of digital biomarkers is a future where cognitive health is monitored continuously and unobtrusively — not through occasional clinic visits, but through the natural use of devices people already carry. In this vision, a sustained change in cognitive function would be detected by the person's phone or watch months before it reaches clinical attention, enabling earlier evaluation and earlier intervention.

The technical challenges are real but being actively addressed. Signal-to-noise ratio is high — many things influence typing speed and gait other than cognitive function. Personalizing algorithms to individual baselines (rather than population averages) is a key methodological advance. Privacy and data security are critical design constraints.

Several large longitudinal studies — including the UK Biobank cognitive extension and the All of Us Research Program in the United States — are collecting digital behavioral data alongside clinical cognitive assessments, genetic data, and biomarkers. These datasets will be foundational for validating digital biomarkers at scale and across diverse populations.

Keel represents one approach to validated digital cognitive monitoring: using well-designed short tests with established research pedigrees to measure processing speed, working memory, spatial reasoning, semantic fluency, and reaction time — not passively, but in an active daily check-in. The value of either approach — passive or active — is the longitudinal trend data that a single measurement cannot provide.