Malcolm Granat

PhD

University of Salford

Professor in Health and Rehabilitation Science

IDATA (Learn more about this study)

IDATA-49

Oct 21, 2021

Investigating the agreement of the composition and time spent in different physical activities derived from accelerometer data captured at different sensor locations

Accelerometer data captured from a range of sensor locations have been used to quantify the physical behaviour of an individual. The characteristics of the sensor location determine how the acceleration data is processed to identify both the class and the intensity of the activities. Two of the most common sensor locations are the hip and the thigh. From the hip location, accelerations are captured close to the body's centre of mass and have been used to quantify activity type and intensity using a counts based strategy. From the thigh location, thigh inclination and dynamic acceleration have been used to identify different postures. In order to compare research that uses different sensor locations, we need to be confident that the algorithms used to process accelerometer data for these sensor locations provide a consistent measure of an individual's physical behaviour.

While several studies have investigated the agreement between physical activity classifications obtained using these sensor locations, they have typically been small validation studies that have only considered daily volumes of physical activity. We believe that by considering the temporal pattern of physical activity accumulation, we may be able to identify periods where activity classification may be inconsistent across these sensor locations. This should allow us to identify potential classes of activity that may be incorrectly classified. This could enable us to develop classification algorithms that can better identify these activities and improve the accuracy of physical behaviour measurement.

We propose using the paired accelerometer data captured by the thigh-worn activPAL and hip-worn ActiGraph to quantify the temporal pattern of time spent in different physical activity classes (stepping, standing, sitting and lying). We will then test the agreement of the volume and temporal pattern of the different classes to investigate differences in how certain classes of activity are classified for the different sensor locations.