Comparing two wrist-worn accelerometers (Axivity AX3 and Matrix 003) for measuring movement behaviors in British and Chinese older adults

Journal article

Introduction: Two nationally representative cohorts, the English Longitudinal Study of Ageing (ELSA) and the China Health and Retirement Longitudinal Study (CHARLS), recently introduced accelerometry to measure movement behaviours. However, the use of different brands (Axivity AX3 and Matrix 003) may hinder data harmonisation. This study assessed whether the raw acceleration data and machine learning-derived physical activity and sleep outcomes were equivalent between these two accelerometers in both British and Chinese adults.

Methods: Eighty-five British and 117 Chinese adults aged ≥50 years wore both accelerometers in a random order on their dominant wrist for up to eight days. Data were processed using UKderived machine learning algorithms to generate outcomes such as average acceleration (mg), time in 24-hour movement behaviours (hours/day), daily step count, and sleep duration (hours/night). Equivalency was assessed using 95% equivalence tests (±10% equivalence zone).

Results: In both British and Chinese adults, average acceleration, sedentary time, time in bed, and sleep duration were equivalent between the two accelerometers, while time in moderatevigorous physical activity (MVPA) was not (±17.7% in British and ±28.2% in Chinese adults). Time in light physical activity (LPA) was equivalent in British (±6.2%) but borderline in Chinese adults (±10.3%), whereas the opposite was observed for daily step count (±10.5% in British and ±2.9% in Chinese adults).

Conclusion: Average acceleration was comparable between the Axivity AX3 and the Matrix 003 in both British and Chinese adults. Machine learning-derived outcomes were also largely comparable; however, the cross-nationality differences highlight the need for further populationspecific algorithm development and validation.

Authors: Brocklebank, L; Steptoe, A; Zhao, Y; Acquah, A; Yuan, H

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Human Kinetics
• Journal: Journal for the Measurement of Physical Behaviour
• Volume: 8
• Issue: 1
• Article number: jmpb.2025-0011
• Publication date: 2025-09-26
• Acceptance date: 2025-07-08
• DOI: 10.1123/jmpb.2025-0011
• EISSN: 2575-6613
• ISSN: 2575-6605
• Language: English
• Keywords: raw acceleration; machine learning; physical activity; sleep; data harmonisation; cross-nationality comparison