Objective: To investigate differences in sleep and activity digital phenotypes from passive, free-living wrist accelerometry across persons with Parkinson’s disease (PwP), individuals with risk markers (at-risk) and normal or reduced DaT SPECT Specific Binding Ratio (n/rSBR), and healthy controls (HC).
Background: Early detection of Parkinson’s disease (PD) would benefit from objective digital biomarkers capturing disease manifestations in daily living. Wrist accelerometry offers continuous measures of sleep and activity, yet its ability to detect prodromal PD from healthy aging remains unclear.
Method: We analyzed 434±205 days of wrist acceleration per participant from the PPMI (https://www.ppmi-info.org) Verily Study Watch cohort (n=322; PwP=152, nSBR=121, rSBR=17, HC=32). We extracted 228 sleep, activity, and wrist acceleration features and reduced them to 81 non‑collinear variables. Features were adjusted for age, sex, weight, height using OLS regression, and residual group differences were tested via Kruskal–Wallis and Dunn–Bonferroni post‑hoc tests. Effect sizes were quantified with Cliff’s Delta with bootstrapped 95% CIs.
Results: HC vs nSBR cohorts were nearly identical (1/81 significant features, δ=0.34, medium effect). HC vs rSBR had strong separation (23/81 significant features, all large effects, δ=0.67-0.74). rSBR subjects showed decreased movement variability and amplitude. HC vs PwP had robust discrimination (51/81 features significant, 31 large effects) with top discriminators: acceleration range (δ=0.75), acceleration entropy (δ=0.73), and REM sleep duration (δ=0.70). At-risk comparisons showed: nSBR vs rSBR had 15 discriminating features (10 large effects); nSBR vs PwP had 48 significant features (19 large effects), dominated by movement amplitude (δ=0.67) and REM sleep duration metrics (δ=0.63-0.64); and rSBR vs PwP had 8 significant features (3 large effects), with REM sleep duration metrics remaining discriminative (δ=0.62-0.66).
Conclusion: Wrist accelerometry distinguishes unaffected individuals from those with dopaminergic deficiency or PD, but not from at-risk individuals with normal SBR. These sleep and activity digital phenotypes show strong potential for machine learning models screening for dopaminergic dysfunction as a proxy for prodromal PD.
Funded by the European Union under GA ID 101080581
To cite this abstract in AMA style:
I. Gerasimou, A. Moustaklis, L. Hadjileontiadis, S. Hadjidimitriou. Differences in Sleep and Activity Digital Phenotypes from Wrist Accelerometry in Parkinson’s Disease, At‑Risk Individuals, and Healthy Controls [abstract]. Mov Disord. 2026; 41 (suppl 1). https://www.mdsabstracts.org/abstract/differences-in-sleep-and-activity-digital-phenotypes-from-wrist-accelerometry-in-parkinsons-disease-at-risk-individuals-and-healthy-controls/. Accessed October 1, 2026.« Back to 2026 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/differences-in-sleep-and-activity-digital-phenotypes-from-wrist-accelerometry-in-parkinsons-disease-at-risk-individuals-and-healthy-controls/