Development Of Prediction Models For Activity Energy Expenditure Using Wrist-worn ActiGraph GT3X+ In Adults

PurposeOn the basis of classifying activity type,regression model is developed to predicting energy expenditure for dominant and non-dominant wrist-worn ActiGraph GT3X+accelerometer in adults.Objective to provide the scientific evidence for improving estimates of energy expenditure.Methods62 adults between the ages of 18 and 40 years(31 males and 31 females,Age:29.3±6.2years,BMI:22.4±2.4kg/m2)were recruited to participate in the study.Each participant performed 13 daily physical activities including hand writing,mobile phone using,computer working,clothes washing,goods sorting,self-paced walking,mopping floor,5.6km/h brisk walking,6.4km/h slow running,8.4km/h fast running,table tennis playing,aerobics and basketball playing.Participants were asked to wear ActiGraph GT3X+accelerometer on the dominant wrist and non-dominant wrist.Energy expenditure data were collected using Cosmed K4b2 simultaneously.Vector magnitude(VM)for distinguishing between sedentary behavior(SB)and non-SB was developed using receiver operating characteristic(ROC)analysis.Coefficient of variation(CV)was developed using ROC analysis to distinguish between regular physical activity and irregular physical activity.These thresholds were evaluated in terms of area under the curve(AUC),sensitivity and specificity.Linear regression analyses were used to develop regression equations.Prediction equations were assessed through cross-validation procedures using the predicted residual sum of squares(PRESS)method.Results1 For the sedentary behavior threshold using vector magnitude:VM?299 counts/10s for non-dominant wrist;VM?320 counts/10s for dominant wrist.2 For the regular physical activity threshold using coefficient of variation:CV?15.95%for non-dominant wrist;CV?13.95%for dominant wrist.3 Regression equations for non-dominant wrist-worn accelerometer:Irregular physical activity METs=1.925+0.0013*VM(R2=0.577,SEE=1.483 METs)and regular physical activity METs=2.686+0.0012*VM(R2=0.725,SEE=1.065 METs);regression equations for dominant wrist-worn accelerometer:Irregular physical activity METs=1.389+0.0013*VM(R2=0.715,SEE=1.222 METs)and regular physical activity METs=2.564+0.0012*VM(R2=0.762,SEE=1.007 METs).4 Using cross-validation of the PRESS method,slightly lower of R2(R2?<0.0027)and slightly higher of SEE(SEE?<0.0029METs)for two-regression model of both wrists.ConclusionThe regression equations have good accuracy for the prediction energy expenditure of both wrist-worn ActiGraph GT3X+accelerometer.Regression equations show good stability through cross-validation procedures of PRESS.With regard to regression equations to predict energy expenditure,dominant wrist placement achieved a higher accuracy compared to the non-dominant wrist.The wrist-worn accelerometer was superior predicting energy expenditure of regular physical activity.According to the results,we suggest to select dominant wrist as attachment sites of ActiGraph GT3X+accelerometer for measuring physical activity.Using activity type specific prediction model,when we estimate energy expenditure of physical activity.