Location: Children's Nutrition Research CenterTitle: Prediction of energy expenditure from heart rate and accelerometry in children and adolescents using multivariate adaptive regression splines modeling) Author
Submitted to: Obesity
Publication Type: Abstract Only
Publication Acceptance Date: 6/1/2009
Publication Date: 11/1/2009
Citation: Zakeri, I., Puyau, M., Adolph, A., Vohra, F., Butte, N.F. 2009. Prediction of energy expenditure from heart rate and accelerometry in children and adolescents using multivariate adaptive regression splines modeling [abstract]. Obesity. 17(2):S150. Interpretive Summary:
Technical Abstract: Free-living measurements of 24-h total energy expenditure (TEE) and activity energy expenditure (AEE) are required to better understand the metabolic, physiological, behavioral, and environmental factors affecting energy balance and contributing to the global epidemic of childhood obesity. The specific aim of this study is to develop a model based on direct ambulatory monitoring of heart rate (HR) and accelerometry to accurately predict energy expenditure (EE), and hence 24-h TEE and AEE in children and adolescents. Multivariate adaptive regression splines (MARS) will be used to model the complex, nonlinear relationships of HR, accelerometer counts (ACs) on EE by a series of spline functions on different intervals of the independent variables. MARS models for the prediction of minute-by-minute EE and hence 24-h TEE from HR and AC and other potential subject covariates were developed in 109 and validated in 61 normal-weight and overweight children (aged 5-18) using Actiheart and 24-h room respiration calorimetry. MARS model for the prediction of minute-by-minute EE consisted of 28 linear spline basis functions using AC, 1-min and 2-min lagged values of AC, 1-min and 2-min lead values and 2-min lagged values of HR, and subject characteristics of gender, weight, and height. Absolute prediction errors for 24-h TEE in the development and validation groups were 0.0 +/- 155 kcal and -68 +/- 155 kcal, with root mean square (RMS) errors of 154 kcal and 168 kcal, respectively. Prediction errors and RMS for sleep EE and awake EE were similar. Another MARS model for the prediction of AEE was developed based on 30 basis functions using AC, 1-min and 2-min lagged values of AC, 2-min lead and 2-min lagged values of HR, and subject characteristics of gender, weight, and height. Absolute prediction errors for AEE in the development and validation groups were 0.0 +/- 108 kcal and -66 +/- 104 kcal, with RMS errors of 108 Kcal and 122 kcal, respectively. Multivariate adaptive regression spline modeling provides a useful predictive model for 24-h EE and AEE in children and adolescents based on heart rate monitoring and accelerometry.