Novel daily energy expenditure estimation by using objective activity type classification: where do we go from here?

Abstract

ACCURATE ESTIMATION of daily energy expenditure (DEE) in population studies is challenged by factors such as feasibility, cost, and validity of the method under consideration. The use of wearable sensors, in particular accelerometers, is possibly the most promising currently available method to estimate DEE. Numerous accelerometer-based prediction models for estimating DEE and physical activity energy expenditure (PAEE) have been developed (5). Most of the prediction models published so far are based on linear regression analysis in experimental populations. To increase precision of these regression models, group-specific or even individual-specific regression models may be necessary. An alternative and potentially more effective and feasible approach may be to assess other individually specific characteristics that explain those yet unexplained differences between populations and between individuals. Assessing the major types of physical activity usually performed in daily life may improve the explained variation in DEE. In their article in the Journal of Applied Physiology, Bonomi and colleagues (3) present results from a study aimed to develop a model for estimating energy expenditure from accelerometer-based classification of types of physical activity in combination with published data on the energy cost [metabolic equivalent (MET) compendium values] of these activities (1, 3). The method proposed is novel because it does not rely on the use of regression analysis for predicting energy expenditure, nor does it rely on manufacturer-dependent information such as activity counts, the usual output from activity monitors based on accelerometry. Bonomi and colleagues compare their newly developed method with the more traditional method of using activity counts for DEE prediction. The high level of transparency of the proposed method is likely to inspire future research in this area. However, there are some additional issues that remain to be elucidated. The importance of classifying activity types when estimating energy expenditure is not yet fully understood. Previous studies have shown that the slope of the linear relation between energy expenditure and body acceleration varies across types of physical activity (6, 9). Theoretically, knowledge about the slope for each type of activity explains the variation in energy expenditure between types of activity within individuals that cannot be explained by acceleration (i.e., counts) alone. Between-individual differences in time spent in each type of physical activity may therefore be related to the average individuals’ slope across activities over a day and thus explain between-individual differences in DEE. In the model proposed by Bonomi and colleagues the types of physical activity identified by accelerometry and an accelerometer-based estimation of the speed of movement are used to estimate the MET intensity level defined by the compendium (1). In contrast to the assessment of counts by accelerometry, the compendium, at least in theory, takes into account differences in slope between types of activity because the activities included are based on measured or estimated energy expenditure rather than body acceleration. Therefore, the activity classification from accelerometry may have accounted for differences in slope between types of activity via the compendium. Future research will have to evaluate whether the main contribution of activity classification truly lies in the ability to account for differences in slope, which would indicate that activity classification needs to focus on explaining differences in slope rather than the activity type itself. Additionally, the y-intercept of the linear relation may also be an important parameter.