Estimating prospective change in clinical scores using sleep and physical activity data from a wearable device in Huntington’s disease [Conference Abstract]

Doheny, Emer, Drew, Cheney ORCID: https://orcid.org/0000-0002-4397-6252, Busse, Monica, Witkowski, Greg, Jung, Hans, Cubo, Esther, Arnesen, Astri, Rosser, Anne ORCID: https://orcid.org/0000-0002-4716-4753 and Lowery, Madeleine 2026. Estimating prospective change in clinical scores using sleep and physical activity data from a wearable device in Huntington’s disease [Conference Abstract]. Journal of Neurology, Neurosurgery and Psychiatry 97 (Supp 2) 10.1136/jnnp-2026-ehdn.157

Full text not available from this repository.

Abstract

Background: Sleep and physical activity are associated with disease progression in Huntington’s Disease (HD). Following recent validation of sleep and activity features from Fitbit Charge 4 in HD3,4, we investigate associations between sleep, activity and disease progression. Aim: To investigate if Fitbit-estimated sleep and activity metrics can predict prospective change in composite Unified HD Rating Scale (cUHDRS). Methods: Sixty-one participants with HD wore a Fitbit during an observational study with cUHDRS obtained at baseline and 12-months (32 male, ΔcUHDRS=-0.54±1.54; Imputed HD-ISS ≤1/2/3/unknown: N=4/10/43/4). Following a 7-day adjustment period, Fitbit-estimated total sleep time (TST), rapid eye movement sleep (REM), deep sleep, REM latency and step count were averaged over days 8-21. Quadratic logistic regression, with sequential feature selection, was used to develop three models to detect participants exhibiting declined cUHDRS at 12-months: ModelDemo considered demographic/genetic data (age, CAG, CAP), ModelFB considered Fitbit-estimated features, ModelAll considered Fitbit and demographic/genetic features. Models were evaluated via cross-validation using the area under the receiver operating characteristic curve (ROCAUC) and F1 score. Results: Greater performance was observed for models including sleep/activity features, compared with models using demographic/genetic data alone. ROCAUC was highest for ModelAll (0.73±0.24), while F1 was highest for ModelFB (0.66±0.18). The lowest ROCAUC and F1 were observed for ModelDemo. (ROCAUC 0.37±0.28, F1 0.56±0.18). Conclusions: Results suggest that sleep and activity data contain information that may be useful in predicting clinical deterioration in HD.

Item Type:	Short Communication
Date Type:	Publication
Status:	Published
Schools:	Schools > Medicine Research Institutes & Centres > Centre for Trials Research (CNTRR)
Publisher:	BMJ Publishing Group
ISSN:	0022-3050
Last Modified:	26 Oct 2026 12:45
URI:	https://orca.cardiff.ac.uk/id/eprint/190431

Actions (repository staff only)

Edit Item