Improving Energy Expenditure Estimates From Wearable Devices: A Machine Learning Approach

Background: Limited evidence exists examining associations between objectively measured physical activity and body composition in adults following clinically significant weight loss (¡Ý5%). Methods: 1,207 participants recruited to a weight loss maintenance study were included in the analysis. Minute-level physical activity and heart rate data were collected over 14 days using a Fitbit Charge 2 (FC2). Time spent in sedentary behaviour (SB), light, moderate and vigorous activity was estimated using the heart rate reserve method and data on steps was obtained from FC2. Body mass index (BMI) and body composition were collected at the start of 14 days using bioelectrical impedance analysis. Hierarchical multivariable regressions were used to investigate the effect of physical activity on BMI and body composition. Results: The FC2 was worn 97% of t he day. 933 ¡À 164 mins/day was spent in SB (65%) and 26 ¡À 25 mins/day (2%) in vigorous physical activity. An average of 10,710 steps were recorded per day. Vigorous physical activity and steps were consistently negatively associated with BMI and positively associated with fat-free mass following adjustment for BMI and sedentary behaviour. An interaction effect was found between vigorous physical activity and sedentary behaviours. Conclusions: The results of this cross-sectional analysis support associations between physical activity, BMI and body composition. We found evidence of an effect of physical activity on body composition independent of BMI. Further, the interaction between VPA and SB is relevant to BMI and body composition.
Refereed journal
Output Tags
Theme 3: Health and Wellbeing (RESAS 2016-21)