Page 182 - Contributed Paper Session (CPS) - Volume 4

CPS2169 Carmen D. Tekwe et al.
school-based interventions as targeted interventions designed to reduce
sedentary behaviour among children. An example of such behavioural school-
based intervention is the activity permissive learning environment (Benden, et
al. 2014). Activity permissive learning environments introduce stand-biased
desks into classrooms as a means of increasing physical activity among school-
aged children. By reducing sedentary behaviour, physical activity behaviour is
encouraged during the school day, and devices such as physical activity
monitors are used to assess the behavioural patterns of physical activity. These
devices provide estimates of school day energy expenditure (SDEE), the total
amount of energy or calories expended by the body, to perform physical
activity and routine bodily functions during the school day. Overweight and
obesity in children are defined based on age- and sex- adjusted body mass
indexes (BMI) in the upper percentile ranges. However, most studies assessing
impacts of interventions on BMI rely on traditional linear regression models
designed to assess intervention effects on children within ``normal'' BMI
percentile ranges, limiting assessments of how interventions affect children at
higher risks for overweight and obesity. Thus, statistical approaches that
permit evaluations of covariates effects across the entire distribution of BMI
are preferable for assessing their effects on subjects at higher risks for
developing overweight or obesity (Koenker, 1978). Quantile regression is a
statistical technique used to estimate effects of predictors on quantile
functions of a response. Examples of quantile functions include the median,
the 85th and the 95th percentiles of the outcome. A drawback to the use of
classical mean regression models in modelling BMI as an outcome is that these
methods provide incomplete answers to questions related to BMI values that
lie within the tails of its distribution. Additionally, covariates such as SDEE and
age may influence the quantile functions differently. Therefore, statistical
approaches that allow one to determine covariate effects across the full
spectrum of quantile functions of BMI is preferable in obesity studies (Koenker,
1978).
Our current work was motivated by a problem in childhood obesity
research. In a recent study, standbiased desks were introduced to three
elementary schools in a Texas school district as a means of increasing physical
activity. A research question of interest was to determine the impact of SDEE
obtained at baseline on subsequent risks for obesity. The recruited children
were given BodyMedia SenseWear® armband devices (BodyMedia,
Pittsburgh, PA) to measure their energy expenditure during school hours,
while sex- and age- adjusted BMI was used as an indicator for obesity. Physical
activity monitoring devices are designed to measure the intensity of physical
activity. Data from these devices are collected either at the second or minute
level over multiple days resulting in high dimensional longitudinal data that
appear as curves. Thus, SDEE data are collected over time and can easily be

171 | I S I W S C 2 0 1 9