Comparison Of Omnidirectional Accelerometer Prediction Equations And Count Cutpoints In Children And Adolescents

Abstract

Interpreting accelerometer output for children and youth is challenging because characteristics related to growth and development can affect the resultant activity counts and the equations used to convert counts to energy expenditure (EE). PURPOSE: The purposes of this study were to: 1) compare different equations (Heil 1 regression (H1), Heil 2 regression (H2), and Puyau (P)) available for converting counts into EE data and 2) explore the classification accuracy of count cutpoints for different physical activity (PA) intensity levels (sedentary, light, moderate, vigorous, and moderate-to-vigorous (MVPA)) for each equation and the Evenson (E) cutpoints. METHODS: Children and youth (N = 101; 62 boys and 39 girls; 6-16 yrs old; 11.2 ± 2.8 yrs, 95% White) wore a portable metabolic analyzer (breath-by-breath measures) and an omnidirectional accelerometer (set to 15-sec epochs) during 12 different activities of various intensity levels, including supine rest. Most activities were performed for five minutes (rest was 10 minutes), and two minutes of steady state from each activity were analyzed. EE data were reduced to kcal/kg/min, and resting values were subtracted to provide activity energy expenditure (AEE) values. The relationship between observed and predicted AEE values was characterized using Pearson correlation and the weighted kappa statistic. Sensitivity and specificity calculations were used to evaluate the classification accuracy. RESULTS: The correlations between observed and predicted EE were r = 0.88, 0.87, and 0.89 (P < 0.0001) for the H1, H2, and P equations, respectively. Kappas were 0.54, 0.55, 0.49, and 0.63 for the H1, H2, P, and E cutpoints across activities, respectively. Sensitivities ranged from 1.4-100.0%, and specificities ranged from 62.4-98.8% across PA intensity classifications. Sedentary activities showed the highest sensitivities across equations (99.6-100.0%), while light activities showed the lowest (1.4-41.9%). Specificities were the highest for MVPA (97.8-98.8%) across activities, while they were the lowest for sedentary activities (62.4-81.8%). CONCLUSIONS: The three equations were similar in their abilities to estimate EE. Of the four sets, the Evenson cutpoints appeared to exhibit the best sensitivity and specificity values. Funded by NICHD (RO1 HD055400-02)