Correction of estimation bias of predictive equations of energy expenditure based on wrist/waist-mounted accelerometers

Abstract

Background. Using wearable inertial sensors to accurately estimate energy expenditure (EE) during an athletic training process is important. Due to the characteristics of inertial sensors, however, the positions in which they are worn can produce signals of different natures. To understand and solve this issue, this study used the heart rate reserve (HRR) as a compensation factor to modify the traditional empirical equation of the accelerometer EE sensor and examine the possibility of improving the estimation of energy expenditure for sensors worn in different positions. Methods. Indirect calorimetry was used as the criterion measure (CM) to measure the EE of 90 healthy adults on a treadmill (five speeds: 4.8, 6.4, 8.0, 9.7, and 11.3 km/h). The measurement was simultaneously performed with the ActiGraph GT9X-Link (placed on the wrist and waist) with the Polar H10 Heart Rate Monitor. Results. At the same exercise intensity, the EE measurements of the GT9X on the wrist and waist had significant differences from those of theCM(p<0.05). By using multiple regression analysis-utilizing values from vector magnitudes (VM), body weight (BW) and HRR parameters-accuracy of EE estimation was greatly improved compared to traditional equation. Modified models explained a greater proportion of variance (R2) (wrist: 0.802; waist: 0.805) and demonstrated a good ICC (wrist: 0.863, waist: 0.889) compared to Freedson's VM3 Combination equation (R2: wrist: 0.384, waist: 0.783; ICC: wrist: 0.073, waist: 0.868). Conclusions. The EE estimation equation combining the VM of accelerometer measurements, BW and HRR greatly enhanced the accuracy of EE estimation based on data from accelerometers worn in different positions, particularly from those on the wrist.