Snapshots Diagnose. Trajectories Explain.

A Thought Experiment

Nobody would accept a single blood pressure reading, taken once, as a lifetime cardiovascular profile. Nobody would measure resting heart rate on one quiet afternoon and treat it as a permanent description of cardiac health. We understand, intuitively, that physiological systems fluctuate — and that meaningful signal lives in the pattern of change over time, not in any single measurement.

Cognitive testing is, in many clinical and research settings, still built around the opposite assumption. A person sits down, completes a standardized test — the MMSE, the MoCA, a neuropsychological battery — receives a score, and goes home. That score is compared against a population norm, not against the person's own history. It is a snapshot, taken under artificial conditions, on a single day that may or may not be representative of anything.

The Limits of the Single-Timepoint Test

The methodological problems with one-time cognitive assessment are well documented, even if they are not always foregrounded in clinical practice. Performance on standardized cognitive tests is sensitive to sleep quality, mood, time of day, test anxiety, and — critically — practice effects from prior exposure to similar tasks, which can mask genuine decline or inflate apparent improvement on repeat testing.^[1]

More fundamentally, a single test administered once or even annually cannot distinguish a stable low performer from someone who is actively declining but happened to have a good day. Research on intra-individual variability — how much a person's own performance fluctuates from occasion to occasion — has shown that this variability is not simply noise to be averaged away. It is itself a signal, and increased short-term variability has been linked to underlying neurological change, sometimes appearing before decline is detectable in mean test scores.^[2]

What Changes Are We Actually Missing?

The concept of subjective cognitive decline — self-perceived worsening of memory or thinking ability in the absence of abnormal performance on standard tests — has become an important research focus precisely because it captures something that discrete testing misses. People frequently notice change in themselves well before that change is measurable on a single administered test, and a meaningful proportion of individuals reporting such decline go on to show objective impairment years later.^[3]

What might that early change actually look like, before it is large enough to register on a standardized score? Plausibly, something like this:

• Slightly longer hesitation before making everyday decisions.
• A gradual narrowing of routine — doing the same things, the same way, more often.
• Reduced initiative to try something unfamiliar or effortful.
• Subtle withdrawal from social interaction that is easy to attribute to other causes.
• Small, cumulative drift in the timing and consistency of daily habits.

None of these, observed once, means anything. Everyone has an off week. But a consistent drift across many such small signals, sustained over months, is a different kind of information entirely — and it is exactly the kind of information that a single clinical visit, by design, cannot capture.

From Passive Behavior to Longitudinal Signal

This is the premise behind the growing field of digital biomarkers: physiological or behavioral measures, collected through smartphones, wearables, or other digital devices, that can be tracked continuously rather than at scheduled intervals.^[4] Instead of a single test administered once a year, digital tools can in principle observe reaction time, typing rhythm, app engagement, movement patterns, and routine stability as an ongoing stream — capturing not a score, but a trajectory.

Reviews of home-based and remote monitoring approaches for cognitive impairment have converged on a similar conclusion: passive, repeated, ecologically valid measurement may detect functionally meaningful change earlier than periodic clinical testing, precisely because it is sensitive to the accumulation of small deviations from a person's own baseline rather than to deviation from a population average.^[5]

Why Machine Learning Needs Trajectories, Not Points

This shift matters even more once machine learning enters the picture. Statistical and machine learning methods for disease progression modeling — approaches that reconstruct the likely sequence and timing of biomarker changes across a population — have shown that the *order and pace* at which measures change over time carries information that a single cross-sectional snapshot structurally cannot contain.^[6] A model trained to recognize patterns in how people change is answering a different, and arguably more clinically useful, question than a model trained to classify a single test score.

Longitudinal, repeated-measurement designs — sometimes called measurement-burst designs in the psychological literature — were developed specifically to separate stable individual differences from within-person change over time, and to do so with enough temporal resolution to detect the latter before it becomes clinically obvious.^[7] Digital tools make this kind of design dramatically more feasible at scale than it has ever been with in-clinic assessment alone.

What This Means for NeuroFlex

This is one of the questions we are exploring through NeuroFlex. Rather than treating brain games, daily routines, and habit tracking purely as isolated wellness activities, the underlying architecture is designed to preserve the process of interaction over time — timing, hesitation, consistency, drift — so that, with appropriate consent and research infrastructure, a person's own behavioral trajectory can eventually be compared against their own history, not just against a population norm.

None of this replaces clinical diagnosis, biomarker testing, or the judgment of a qualified physician. What it may offer, if the underlying hypothesis holds, is an additional layer of observation — one built for detecting change over time rather than performance at a single moment.

Perhaps the next meaningful breakthrough in early Alzheimer's detection will not come from a better one-time test. Perhaps it will come from asking a more patient question — not "how did this person perform today?" but "how has this person changed?"