Measuring Digital Behavioral Biomarkers Before Symptoms of Alzheimer's Disease Appear

Section 1

Executive Summary

NeuroFlex is being developed as a two-layer platform: a research-oriented platform architecture, with a free cognitive wellness Android application available today as its first user-facing tool, and a longitudinal behavioral research infrastructure designed for tomorrow.

The core scientific question it is built to investigate is whether long-term patterns in everyday digital behavior contain measurable signals associated with pre-symptomatic cognitive change — specifically in the context of Alzheimer's disease (AD) and related dementias.

The platform collects behavioral data across six wellness domains — cognitive training, hydration, movement, nutrition, social connection, and relaxation — and is architecturally designed to preserve not just outcomes but the process of interaction: timing, hesitation, correction, avoidance, and change over time. These process-level signals are hypothesized to carry information that final scores or completion rates do not.

The current version (Phase 1) operates entirely offline, with no data collection. It establishes the user-facing product, the behavioral event architecture, and the participant experience that future research phases will build upon.

Section 2

The Problem: Late Diagnosis, Early Signals

Alzheimer's disease is the leading cause of dementia, affecting an estimated 55 million people worldwide, with numbers projected to reach 139 million by 2050 as populations age. Despite decades of research investment, clinical diagnosis typically occurs at a stage when significant neurodegeneration has already taken place.

The pathological process — amyloid plaque accumulation, tau neurofibrillary tangle formation, synaptic loss, and neuroinflammation — is now understood to begin 15–20 years before the onset of clinically detectable cognitive impairment. This pre-symptomatic window is increasingly recognized as the most important period for intervention, prevention research, and participant recruitment for clinical trials.

Traditional cognitive assessments (MMSE, MoCA, neuropsychological batteries) are also limited in this context: they are administered infrequently, are subject to practice effects and examiner variability, and capture only a snapshot of performance at a single point in time. They do not capture how cognition changes — its trajectory — across the daily texture of a person's life.

NeuroFlex is designed around a different premise: that everyday digital interactions, measured consistently over time, may offer a complementary observational layer — one that is scalable, non-invasive, and capable of capturing longitudinal behavioral trajectories at a granularity that clinical assessments cannot reach.

Section 3

The Approach: Digital Behavioral Biomarkers

A digital biomarker is a physiological or behavioral measure collected through a digital device that can be used to explain, influence, or predict health-related outcomes. The field has expanded rapidly over the past decade, with research demonstrating measurable signals in keystroke dynamics, speech patterns, gait analysis, sleep patterns, and smartphone interaction behavior.

NeuroFlex focuses on behavioral biomarkers derived from structured application interactions: not passive background sensing, but purposeful engagement with cognitive tasks, wellness activities, and daily routine management. This approach offers several advantages:

• Ecological validity: behavior is captured in naturalistic conditions, not in a clinical testing environment.
• Longitudinal density: daily interactions create a continuous behavioral time series across weeks, months, and years.
• Process capture: structured interactions preserve not just outcomes (did the user complete the task?) but process metadata (how long did they hesitate? how many corrections did they make? how did their performance change over repeated sessions?).
• Low participant burden: wellness activities provide intrinsic value to users, independent of research goals, supporting long-term engagement.

The hypothesis is not that any individual behavioral measurement predicts cognitive change. It is that trajectories — patterns of change across extended time periods, across multiple behavioral domains — may contain signal that would not be detectable in any single observation.

Section 4

Scientific Hypotheses

NeuroFlex is developing a series of formal, testable scientific hypotheses that the research programme is designed to investigate. Two are currently published as working hypothesis papers:

Both hypotheses are formulated as falsifiable, testable propositions. Neither asserts that NeuroFlex will detect Alzheimer's disease. Both assert that specific behavioral patterns, measured longitudinally, may carry signal that warrants systematic investigation with appropriate research infrastructure and clinical ground truth.

Section 5

Sex Differences and Model Stratification

Women account for approximately 60–65% of all Alzheimer's patients globally. For decades, this disparity was attributed primarily to differential longevity. Current evidence indicates a more complex biological picture involving at least five interacting mechanisms:

• Estrogen and the menopausal transition: estrogen has documented neuroprotective and neuromodulatory effects. Perimenopausal estrogen decline is associated with measurable reductions in cerebral glucose metabolism — a marker closely linked to AD risk — years before cognitive change.
• Tau pathology: women show stronger associations between amyloid burden and tau accumulation; for equivalent amyloid load, women tend to develop more tau pathology, the biomarker most closely correlated with cognitive decline.
• Cognitive reserve dynamics: women may have higher verbal memory reserve in certain domains, masking pathology longer — leading to later diagnosis at more advanced disease stages and an apparently more rapid post-diagnosis decline.
• APOE ε4 interaction: the risk conferred by a single APOE ε4 allele appears to be substantially higher in women than in men.
• Neuroinflammation: sex differences in microglial activation and neuroinflammatory response may contribute to differential disease progression dynamics.

These mechanisms have a direct consequence for AI-driven behavioral analysis: a model trained on sex-undifferentiated data may produce systematically biased predictions for both male and female participants. NeuroFlex is designed from the outset to treat sex as a primary modeling variable — not a covariate to be controlled for, but a structural feature of the prediction problem itself.

Section 6

Platform Architecture

The NeuroFlex platform is structured around six wellness domains, each designed to generate structured behavioral observations across different cognitive, physical, and social dimensions:

The platform also includes a structured daily check-in system (Pulse, Compass, Wander, and Haven modules) capturing self-reported mood, energy, motivation, focus, stress, loneliness, confidence, memory perception, and perceived cognitive difficulty. These explicit self-report signals complement the implicit behavioral metadata captured during task interaction.

Section 7

Data Model and Behavioral Event Architecture

The NeuroFlex behavioral event model is designed to ensure that data collected in Phase 1 (local, offline) can map cleanly into a future research-grade backend without architectural changes. The core principle is: never store only final scores. Store the process.

The event model is structured across ten levels:

• System events: app lifecycle signals.
• Navigation events: screen transitions, button interactions, invalid taps, repeated taps.
• Activity lifecycle: starts, completions, abandonments, retries, skips.
• Cognitive events: task-level behavioral metadata — timing, hesitation, error type, correction sequences.
• Emotional events: self-reported mood, energy, stress, loneliness, memory self-rating.
• Wellness events: domain-specific completion and engagement signals.
• Routine events: task creation, completion, postponement, and skip patterns.
• Notification events: scheduling, receipt, opening, dismissal, and action selection.
• Behavioral pattern events: challenge avoidance, difficulty adjustment, long pauses, return-after-gap signals.
• Consent events: research consent management, data export requests, data deletion.

The derived metrics that can be computed from this event stream include:

Section 8

Research Methodology

NeuroFlex adopts a four-cohort research design:

• Cohort A: participants with an existing Alzheimer's or dementia diagnosis.
• Cohort B: family members of diagnosed individuals — elevated genetic and environmental risk without current diagnosis.
• Cohort C: biomarker-positive individuals without clinical diagnosis — the pre-symptomatic population of primary research interest.
• Cohort D: general adult population without known risk factors — provides normative behavioral baseline data.

This design avoids the methodological problem of training models only on diagnosed versus healthy populations. The research process is structured in four stages:

The ground truth strategy relies on integration with clinical partner data where available, supplemented by longitudinal self-report and, where possible, linkage to existing biomarker studies. The research is designed to be additive to — not a replacement for — established clinical and imaging biomarker programmes.

Section 9

Privacy and Ethics Framework

Privacy is an architectural principle in NeuroFlex, not a compliance afterthought. The framework is designed to meet GDPR requirements and to be defensible under the standards of independent ethics review.

Current version (Phase 1): entirely offline. No personal data is collected, transmitted, or stored outside the user's device. No analytics, no account creation, no identifiers.

Future research phases: the following principles will govern any transition to data collection:

• Anonymous identifiers by default — no name, email, or government-issued ID required.
• Layered, explicit, informed consent with the ability to withdraw at any time and request data deletion.
• Separate consent for each data tier (Tier 1: core wellness events; Tier 2: detailed interaction metadata; Tier 3: advanced motor signals).
• Independent ethics committee review before any research data collection activates.
• Published, transparent data governance documentation.
• No data sharing with commercial third parties under any circumstances.

The research design explicitly excludes sensitive data categories (precise location, contact lists, camera, microphone, medical records) from default collection. Advanced motor data (Tier 3) requires separate, specific consent and is intended only for participants explicitly enrolled in clinical research partnerships.

Section 10

Development Roadmap

Section 11

Positioning and Related Work

NeuroFlex occupies a specific position in the digital health landscape that is distinct from several existing platform categories:

The key differentiators that NeuroFlex offers relative to this landscape are: consumer-scale deployment potential; longitudinal depth (years, not weeks); dual-purpose design serving both participant wellness and research value; sex-stratified modeling by design; and a Personal Baseline architecture that treats each participant's own behavioral trajectory as the primary reference — not a population average.

Section 12

Limitations and Open Questions

Intellectual honesty about the limitations of this approach is a scientific requirement, not an optional disclaimer. The following represent the most significant methodological challenges:

• Engagement decay: sustained longitudinal participation is the platform's most significant operational challenge. The research value of NeuroFlex depends entirely on multi-year data. Strategies to support engagement without compromising scientific validity are a central design consideration.
• Healthy participant bias: individuals who voluntarily engage with a wellness tool within the NeuroFlex platform may not be representative of the general population — or of the at-risk population — in ways that are difficult to fully characterize or correct for.
• Label uncertainty: the pre-symptomatic population (Cohort C) is by definition undiagnosed. Developing reliable ground truth for behavioral biomarker validation requires integration with clinical partner data.
• Confounding variables: behavioral trajectories are influenced by many factors (depression, medication, life events, digital literacy, screen fatigue) that are not uniquely attributable to cognitive change.
• Diagnostic delay bias: women's higher cognitive reserve may cause them to present for diagnosis later, at more advanced disease stages — which may systematically affect the behavioral data available for model training if not explicitly accounted for.
• Validation requirement: no behavioral biomarker finding from this platform will constitute scientific evidence without independent replication and peer review. NeuroFlex is a hypothesis-generation infrastructure, not a diagnostic system.

The research design described in this document is explicitly constructed to minimize these biases where possible, acknowledge them where they cannot be eliminated, and structure the research outputs so that they are verifiable and replicable.

Section 13

Vision and Next Steps

NeuroFlex is designed around a conviction: that the pre-symptomatic window of Alzheimer's disease is not unreachable. The signals are there. The question is whether we are measuring the right things, at the right granularity, over a long enough time period, in a way that is ethical, scalable, and scientifically defensible.

The immediate next steps are:

• Establishing the first academic research partnership with a neurologist or research group working in the AD biomarker space — ideally within the Slovak or Central European context, given proximity and existing relationships.
• Completing Phase 1 development and beginning structured user feedback collection.
• Initiating ethics committee consultation for the Phase 2 research infrastructure design.
• Developing a formal research protocol for the first observational study, including power calculations and cohort design.

For collaboration enquiries, partnership discussions, or access to research documentation, contact us at info@neuroflex.health.