Answers are grounded in this paper and cite their sources.
Digital biomarkers derived from smartphones, wearables, and connected sensors are increasingly combined with artificial intelligence (AI) to detect and monitor neurodegenerative diseases, yet the signals on which these models depend are not static. This critical review, based on a structured search of biomedical, engineering, and regulatory literature, examines four interacting sources of instability: temporal drift arising from disease progression, aging, treatment, and symptom fluctuation; technological drift caused by changes in devices, sensors, software, and processing pipelines; behavioral and contextual drift associated with routines, adherence, engagement, and missing data; and population drift that arises when deployment cohorts differ from narrow development cohorts. We map these sources onto established machine-learning concepts, including covariate shift, label shift, and concept drift, and summarize disease- and modality-specific evidence in Parkinson disease, Alzheimer disease, Huntington disease, and related conditions. Across the literature, separating genuine disease progression from measurement-related, behavioral, and population-related change remains difficult, and validation based on random data splits, short follow-up, homogeneous cohorts, and single devices can overestimate durability. These limitations can produce false deterioration signals, unstable individual predictions, and inequitable performance after deployment. We propose a Detect → Attribute → Quantify → Adapt → Validate → Monitor (DAQAV-M) framework with illustrative operational criteria, attribution strategies, and version-controlled updating rules, and we position it relative to existing dataset-shift, model-monitoring, and regulatory lifecycle approaches. Trustworthy digital biomarkers require lifecycle validation that separates biomarker validity from model performance and tests stability within individuals, across technologies, and across diverse populations.
No takes yet. Share an insight, caveat, or question.
Oluwadare et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: