Energy Expenditure Estimation And Intensity Classification Physical Activity In Daily Tasks Using Raw Accelerometer Data

BACKGROUND: Accelerometers are often used to estimate sedentary time and energy expenditure associated with physical activity. Since there is no universally accepted method, several metrics and accelerometer positioning protocols can be used for this purpose. PURPOSE: Several metrics were obtained using accelerometer data from the hip, dominant and non-dominant wrist to develop regression equations for energy expenditure (EE) quantification and cut-points for classification of daily tasks intensity. METHODS: 49 healthy adults (18 to 65 yrs) performed a protocol including seven different intensity daily activities (from lying down to running) while wearing three Actigraph GT3X+ accelerometers (right hip, dominant wrist, and non-dominant wrist). Oxygen consumption (V̇O2) and carbon dioxide production (V̇CO2) were measured by indirect calorimetry (K5, Cosmed) while tasks were being performed. Accelerometer metrics analyzed were activity counts, Mean Amplitude Deviation (MAD) and Euclidean Norm Minus One (ENMO). EE was predicted using linear mixed models and receiver operating characteristic curves were used to classify sedentary activities (SA) and different physical activity intensities. Bland-Altman plots, prediction accuracy, Kappa statistic and percent agreement were calculated using leave-one-out cross-validation. RESULTS: All prediction models presented a quadratic equation that had as predictors one of the accelerometer metrics, age, sex, and body mass index. Predicted EE indicated a good agreement with a mean absolute % error below 16% on the hip using ENMO. The equation based on the accelerometer placed at the non-dominant wrist and using the MAD metric had the lowest value of MAE (0.89 kcal·min-1) and RMSE (1.25 kcal·min-1). Global classification agreement from the developed cut-points was categorized as almost perfect with a % agreement above 63% using the accelerometer on the non-dominant wrist and the accelerometer metric MAD. CONCLUSIONS: SA are better predicted with accelerometers placed at the non-dominant wrist while vigorous activities are better predicted with hip placement. Although hip placement seems to be better because it is more central, the use of the non-dominant wrist should be considered in long-term studies as it increases adherence to the protocol.