Validation Of Exercise And Post-exercise Energy Expenditure Estimates Using Hip And Wrist Accelerometer Algorithms

Accelerometer-based devices are objective monitors of physical activity (PA) that can be used to estimate energy expenditure (EE). Most EE estimation algorithms are based on steady-state data and do not consider excess post-exercise oxygen consumption (EPOC). PURPOSE: To quantify the error between linear and machine-learning accelerometer-based EE estimates due to EPOC after varying durations of high-intensity treadmill running. METHODS: Nine adult males participated in three study visits. Visit 1 included a treadmill VO2 peak test to determine the treadmill speed at 80% VO2 peak for visits 2 and 3. Visit 2 included a seated 20 min baseline and three vigorous treadmill running bouts (30s, 60s, 120 s) each followed by 20 min seated rest. Visit 3 included a supine 60 min baseline and a 30 min treadmill running bout followed by 3 hrs of supine rest. Fifteen EE estimation algorithms using either a non-dominant wrist or right hip ActiGraph GT3X+ accelerometer (AG) were compared to measured EE using indirect calorimetry (IC). The 95% CI of EE bias (AG - IC) was used to determine significance. RESULTS: Linear regression EE algorithms tended to overestimate EE after each exercise bout (mean bias kCals [95% CIs]; 30s: 12.5 [3.30, 21.6], 60s: 9.6 [0.37, 18.9], 120 s: 6.5 [-2.74, 15.7], 30 min: 177.5 [84.6, 262.1]). Adding criterion-measured EPOC to the linear EE estimates resulted in additional EE overestimation. Linear EE estimates after exercise were dependent on the y-intercept, which was higher than the measured resting EE resulting in EE overestimation. The 4 non-linear algorithms underestimated EE after the short bouts and demonstrated no significant bias after the 30 min bout (30s: -7.91 [-10.2, -5.57], 60s: -10.9 [-14.0, -7.88], 120 s: -15.4 [-15.9, -14.8], 30 min: 14.83[-39.5, 69.2]). Adding EPOC to the non-linear EE estimates reduced bias after the short bouts but resulted in overestimation after the 30 min bout. CONCLUSIONS: Generally, the addition of measured EPOC to the estimated values increased the amount of EE overestimation during the post-exercise period. When calculating EE estimates from wearable-device data, researchers must be deliberate in selecting an appropriate algorithm based on the activities the algorithm is being applied to and the activities included in the calibration of that algorithm.