Comparison of equations for predicting energy expenditure from accelerometer counts in children A. Nilsson 1 , S. Brage 2 , C. Riddoch 3 , SA. Anderssen 4 , LB. Sardinha 5 , N. Wedderkopp 6 , LB. Andersen 4,6 , U. Ekelund 1,2 1 School of Health and Medical Science, Örebro University, Sweden, 2 MRC Epidemiology Unit, Cambridge, UK, 3 School for Health, University of Bath, UK, 4 Department of Sports Medicine, Norwegian School of Sport Sciences, Norway, 5 Faculty of Human Movement, Technical University of Lisbon, Portugal, 6 Institute of Sports Science & Clinical Biomechanics, University of Southern Denmark, Odense, Denmark Several prediction equations developed to convert body movement measured by accelerometry into energy expenditure have been published. The study aim was to examine the degree of agreement between three different prediction equations, when applied to data on physical activity in a large sample of children. We examined 1321 children (663 boys, 658 girls; mean age 9.6  0.4 y.) from four different countries. Physical activity was measured by the MTI accelerometer (model 7164). One equation, derived from doubly-labelled water (DLW) measurements was compared to one treadmill-based and one room calorimeter-based equation (mixture of activities). Predicted physical activity energy expenditure (PAEE) was the main outcome variable. In comparison with DLW-predicted PAEE, the laboratory-significantly (p < 0.001) overestimated PAEE by 17% and 83% respectively when based on a 24-hour prediction, while significantly (p < 0.001) underestimated PAEE by 46% and 3% respectively, when based on awake time only. Predicted PAEE differ substantially between equations, also depending on time frame assumptions. These equations can not be used interchangeably and interpretations of average levels of PAEE in children from available equations should be made with caution. Further development of equations applicable to free-living scenarios is needed. The prevalence of overweight in children has been reported to have increased in many western countries (Lobstein et al., 2003; Troiano & Flegal, 1998; Willms et al., 2003), which has raised scientific interest in the potential relationship between components of total energy expenditure (TEE) and the development of overweight and obesity in childhood (DeLany, 1998; Salbe et al., 2002) and other features of the metabolic syndrome (Andersen et al, 2006). Physical activity energy expenditure (PAEE) is the most variable component of TEE and therefore key for the regulation of TEE and of specific interest to measure accurately in epidemiological studies. However, the direct assessment of daily PAEE is problematic since the methods with the highest degree of validity are either expensive (doubly labelled water) or impractical in field settings (respiratory gas analysis) (Schutz et al., 2002). Therefore, assessing physical activity with other objective methods, such as accelerometry, is an approach that has been successfully used in large-scale epidemiological studies in children (Riddoch et al., 2004; Ekelund et al., 2004). The outcome from accelerometry (i.e. activity counts) can be transformed to predict PAEE by regression analysis. This relationship may also be used when interpreting children’s physical activity behavior. For example, when analyzing time spent at different intensity levels of physical activity (e.g. low, moderate, high) or when examining proportions of children who reach recommended levels of physical activity, the relationship between activity counts and EE is used to establish intensity thresholds. One of the most widely used accelerometers, the MTI accelerometer is valid for assessing the total amount of physical activity on a group level (Ekelund et al., 2001), and provides detailed information of activity at different levels of intensity (Nilsson et al., 2002; Trost et al., 2002; Puyau et al., 2002; Freedson et al., 1998; Hendelman et al., 2000; Swartz et al., 2000). However, activity counts from accelerometry do not provide direct information on TEE and its sub- components. A number of prediction equations have been developed to convert activity counts to components of TEE (Ekelund et al., 2001; Trost et al., 2002; Puyau et al., 2002; Freedson et al., 1998; Hendelman et al., 2000; Swartz et al., 2000). Almost all prediction equations have been developed during laboratory-restricted activities using respiratory gas analysis as the criterion measure (Trost et al., 2002; Puyau et al., 2002; Freedson et al., 1998; Swartz et al., 2000), and these equations can also be used to create specific cut-off limits using activity counts corresponding to certain intensity levels. To determine appropriate intensity levels from calorimetry measurements, the Metabolic Energy Turnover (MET) classification system is often used, 1