Validity of wearable actimeter computation of total energy expenditure during walking in post-stroke individuals

Recent studies reported that wearable sensor devices show low validity for assessing the amount of energy expenditure in individuals after stroke.

We aimed to evaluate the validity of energy expenditure calculation based on the product of energy cost and walked distance estimated by wearable devices in individuals after hemispheric stroke.

We recruited individuals with hemispheric stroke sequelae who were able to walk without human assistance. The participants wore a tri-axial accelerometer (Actigraph GT3x) and a pedometer (ONStep 400) on the unaffected hip in addition to a respiratory gas exchange analyzer (METAMAX 3B) during 6min of walking at their self-selected walking speed and mode. The energy expenditure was calculated from the product of energy cost measured by the METAMAX 3B and the distance estimated by wearable devices. It was compared to the energy expenditure measured by the METAMAX 3B and the energy expenditure values recorded by the devices according to the manufacturer's algorithms. The validity was investigated by Bland-Altman analysis (mean bias [MB], root mean square error [RMSE], limits of agreement [95%LoA]), and Pearson correlation analysis (r).

We included 26 participants (mean [SD] age 64.6 [14.8] years). With the pedometer, the energy expenditure calculated from the product of energy cost and walked distance showed high accuracy and agreement with METAMAX 3B values (MB=-1.6kcal; RMSE=4.1kcal; 95%LoA=-9.9; 6.6kcal; r=0.87, P<0.01) but low accuracy and agreement with Actigraph GT3x values (MB=15.7kcal; RMSE=8.7kcal; 95%LoA=-1.3; 32.6kcal; r=0.44, P=0.02) because of poorer estimation of walked distance. With the pedometer, this new method of calculation strongly increased the validity parameter values for estimating energy expenditure as compared with the manufacturer's algorithm.

This new method based on the energy cost and distance estimated by wearable devices provided better energy expenditure estimates for the pedometer than did the manufacturer's algorithm. The validity of this method depended on the accuracy of the sensor to measure the distance walked by an individual after stroke.