Longitudinal compositional data analysis applied to accelerometry data over childhood and adolescence

Compositional data analysis techniques have emerged in recent years to analyse physical activity data including objective measurements from accelerometers. Compositional data analysis accounts for the dependence of physical activity behaviours compared to standard statistical techniques, which incorrectly assumes behaviours are independent. Although compositional data analysis applied to cross-sectional data, including physical activity data has been widely used, there has been limited progress in extending this over multiple time points (i.e. longitudinally).A review of the current literature on compositional data analysis applied to longitudinal data has been done as part of this project and found that no studies to date have analysed more than two time points of accelerometry data. In addition, the methods used for analysing two timepoints were limited and often could not explore trends within individuals.Statistical methods that can model compositional data over multiple time points, which also account for trends within individuals, will be demonstrated. Multivariate techniques, such as mixed effects models that are suitable for modelling the association between multiple compositional physical activity behaviours and body weight, will be presented and applied to more than two time points of compositional accelerometry data in the EarlyBird data set.EarlyBird is longitudinal cohort study of 347 healthy children, recruited from 54 primary schoolsin the City of Plymouth in the UK, and their parents (Voss et al. 2003). Annual observations of the children from age 5 years up to and including age 16 were collected, with accelerometry data and BMI from age 5 and body composition from age 9. The methods described above can give a more informative picture of the association between physical activity and body weightover childhood and adolescence.