Validation of algorithms for energy expenditure evaluation in children using raw acceleration data
MetadataShow full item record
Backround: The relation between energy expenditure (EE) in childhood physical activity and childhood or later adult health status is not clearly established. Thus, a better understanding of relation between energy expenditure and health in children are needed. An accessible approach for estimation of energy expenditure is important and may be established by using acceleration data. Aim: Determine the accuracy of Brandes´ linear regression equation for estimation of EE when applied to children (7-15 years old) using raw acceleration data.Furthermore, stepwise regression equations will be developed to improve accuracy of estimating EE from children acceleration data. Method: A cross-sectional study examining 42 children (20 girls and 22 boys; 7-15 years) wearing two accelerometers (to the lower back and thigh) and a portable indirect calorimetry as reference measure for total EE estimation was performed. The children performed several physical activities including walking With different speed, jogging and running. The accuracy of an existing data model (1) for evaluation of EE was tested. Further, stepwise regression analysis was used for developing new regression equations based on walking activities (NTNUwalking), vigorous activities (NTNUvigorous) and all activities combined (NTNUall int.). The accuracy of the different equations was assessed using correlation, coefficient of determination and Bland-Altman plots. Further, Bland-Altman plots is presented as mean bias and limits of agreement (LoA). EE is presented both as an absolute measure (kJ/min) and relative to body mass (J/kg/min). Results: The smallest difference between measured and estimated EE using Brandes et al.´s equation was found when walking moderately both for absolute (mean bias -0.78 and 95% LoA -9.80 to 8.24) and relative measures (mean bias -57.24 and 95% LoA -213.92 to 99.44). Further, NTNUwalking presented a greater accuracy for walking activities compared to Brandes et al.´s equation (absolute units: mean bias 0.24 and 95% LoA -3.23 to3.72. Relative units: mean bias 18.06 and 95% LoA -43.71 to 79.84). In addition to NTNUwalking, NTNUall int. presented the greatest accuracy for jogging and running when absolute values for EE were used (NTNUwalking: mean bias -1.23 and 95% LoA -12.88 to 10.40. NTNUall int.: mean bias 1.22 and 95% LoA -9.29 to 11.75). Conclusion: Brandes´ equations estimated EE in children most accurately for absolute values when walking moderately. However, newly developed equations were more accurate for both walking and vigorous intensities, supporting childspecific regression equations for estimating EE in children. The above equations worked reasonably well for estimating group estimates. However, further adaptations are needed to enhance accurate individual EE estimation.