JMIR Publications

ABSTRACT

Background:

Wearable devices enable continuous measurement of physical activity, sedentary behaviour, sleep, and heart rate under free-living conditions. However, most validation studies rely on small, homogeneous samples, are conducted under laboratory conditions, or lack gold-standard ground-truth measurements, limiting the generalisability and accuracy of derived metrics. There is a pressing need for open-access, large-scale, free-living validation datasets that include multi-sensor data from diverse body locations and participant demographics to aid in model development.

Objective:

The OxWEARS study aims to: (1) validate accelerometer-based measurement of physical behaviours across five body sites against annotated camera data; (2) validate measurements of sleep and sleep staging from five different body sites against polysomnography (PSG); (3) validate wrist-worn photoplethysmography (PPG) heart rate measurements against chest-worn electrocardiogram (ECG); and (4) generate a comprehensive, annotated, and anonymised dataset for open-access research use.

Methods:

A total of ~160 adults (aged ≥40 years) stratified by age, sex, and BMI will be recruited from the Oxford Biobank. Over three days and four nights, participants will wear sensors on the wrists, chest, hip, thigh, and ankle. Ground-truth measures will be obtained from a chest ECG patch for heart rate, a first-person camera for activity annotation, ankle-worn accelerometer for step count, and at-home PSG for sleep. An under-mattress sensor will collect measures of sleep, respiration rate and bed time, and a subjective sleep diary will also be obtained. Signals from different wear locations will be compared against the ground-truth using precision, recall, F1, kappa, and agreement metrics.

Results:

Recruitment commenced in November 2024, with 15 participants enrolled by May 2025. Data collection is ongoing and expected to conclude in 2026, with the final annotated dataset made publicly available as soon as possible thereafter.

Conclusions:

The OxWEARS study will produce an openly accessible dataset exceeding 10,000 annotated hours in a stratified sample of adults. This will directly support scalable, generalisable human activity recognition efforts, whilst also enabling robust development and benchmarking of wearable-derived health metrics.