Abstract
Accelerometry is a state-of-the-art procedure to capture physical activity. However, the large variety of accelerometry metrics and wearing positions hamper the comparability of outcomes. Since this is a well-known challenge, we investigated how outcomes can be translated across four metrics and six sensor positions. Twenty healthy adults conducted 32 structured and semi-structured activities while wearing accelerometers at the hip, chest, thigh, wrist, ankle, and upper arm. The raw data was converted into four common metrics: Movement Acceleration Intensity (MAI), Euclidean Norm Minus One (ENMO), Mean Amplitude Deviation (MAD) and counts per minute (CPM), as computed by the Actigraph GT3X+ algorithm. Relationships between acceleration metrics and sensor positions were quantified via Pearson correlations and scatterplots. Our results show that nearby sensor positions were highly correlated (e.g., MAD hip and thigh: r = .96), while correlations between more distant sensor positions were weaker and less linear (e.g., MAD wrist and thigh: r = .80). Correlations between MAI, MAD and ENMO were high (r = .9), while correlations between CPM and other metrics were substantially lower (r = .78), less linear, and influenced by activity type. Thus, linear conversion between MAI, ENMO and MAD are highly feasible, but converting CPM may be less accurate. Linear conversions between nearby sensor positions are accurate, yet linear conversions between more distant sensor positions appear challenging. Importantly, based on 32 activities as well as metric- and sensor-location-specific configurations, we provide a comprehensive overview of outcome measures that enables researchers to individually explore conversion opportunities towards their own data.