Wrist-specific accelerometry methods for estimating free-living physical activity

Michael I.C. Kingsley, Rashmika Nawaratne, Paul D. O'Halloran, Alexander H.K. Montoye, Damminda Alahakoon, Daswin De Silva, Kiera Staley, Matthew Nicholson

Research output: Contribution to journalArticleResearchpeer-review

14 Citations (Scopus)


Objectives: To compare accelerometry-derived estimates of physical activity from 9 wrist-specific predictive models and a reference hip-specific method. Design: Prospective cohort repeated measures study. Methods: 110 participants wore an accelerometer at wrist and hip locations for 1 week of free-living. Accelerometer data from three axes were used to calculate physical activity estimates using existing wrist-specific models (3 linear and 6 artificial neural network models)and a reference hip-specific method. Estimates of physical activity were compared to reference values at both epoch (≤60-s)and weekly levels. Results: 9044 h were analysed. Physical activity ranged from 7 to 96 min per day of moderate-to-vigorous physical activity (MVPA). Method of analysis influenced determination of sedentary behaviour (<1.5 METs), light physical activity (1.5 to <3 METs)and MVPA (>3 METs)(p < 0.001, respectively). All wrist-specific models produced total weekly MVPA values that were different to the reference method. At the epoch level, Hildebrand et al. (2014)produced the strongest correlation (r = 0.69, 95%CI: 0.67–0.71)with tightest ratio limits of agreement (95%CI: 0.53–1.30)for MVPA, and highest agreement to predict MVPA (94.1%, 95%CI: 94.0–94.1%)with sensitivity of 63.1% (95%CI: 62.6–63.7%)and specificity of 96.0% (95%CI: 95.9–96.0%). Conclusions: Caution is required when comparing results from studies that use inconsistent analysis methods. Although a wrist-specific linear model produced results that were most similar to the hip-specific reference method when estimating total weekly MVPA, modest absolute and relative agreement at the epoch level suggest that additional analysis methods are required to improve estimates of physical activity derived from wrist-worn accelerometers.

Original languageEnglish
Pages (from-to)677-683
Number of pages7
JournalJournal of Science and Medicine in Sport
Issue number6
Publication statusPublished - Jun 2019
Externally publishedYes


  • Accelerometer
  • Actigraph
  • Artificial neural network
  • Hip
  • Physical activity
  • Wrist

Cite this