Combining Acceleration and Skin Temperature Can Improve Accuracy of Physical Activity Trackers

Medicalresearch: In medical and sport science research, body-worn accelerometers are widely used to provide objective measurements of physical activity. However, accelerometers collect data continuously even during periods of nonwear (i.e. periods when participants may not be wearing their monitor, such as during sleeping). It is important to distinguish time of sedentary behaviours (eg. watching television) from time of nonwear. The clinical consequence of misclassification of accelerometer wear and nonwear would overestimate or underestimate physical activity level, and mislead the interpretation of the relationship between physical activity and health outcomes.

Automated estimation of accelerometer wear and nonwear time events is particularly desired by large cohort studies, but algorithms for this purpose are not yet standardized and their accuracy needs to be established.
This study presents a robust method of classifying wear and nonwear time events under free living conditions for triaxial accelerometers which combines acceleration and surface skin temperature data. The new findings are: Either acceleration data or skin temperature data alone is inadequate to accurately predict wear and nonwear events in some scenarios under a free living condition; This study provides a simple and efficient algorithm on use of short time periods of consecutive data blocks for accurately predicting triaxial accelerometer wear and nonwear events; Combining both types of acceleration and skin temperature data can significantly improve the accuracy of accelerometer wear and nonwear events classification in monitoring physical activity.
Clinicians and researchers would benefit from using the reported method to generate more accurate estimations of time spent in sedentary and active behaviours in free living conditions, and gain correct interpretation of relationships between physical activity, energy expenditure and health outcomes.
The reported methodology has its wide generality for combining multiple sources of accelerometer data to increase accuracy of free-living physical activity. It is recommended to apply such wear and non-wear detection method to large cohort studies that investigate the link between baseline physical activity assessment and health risks and disease outcomes over long time periods.