Compositional data analysis for physical activity, sedentary time and sleep research

Abstract

The health effects of daily activity behaviours (physical activity, sedentary time and sleep) are widely studied. While previous research has largely examined activity behaviours in isolation, recent studies have adjusted for multiple behaviours. However, the inclusion of all activity behaviours in traditional multivariate analyses has not been possible due to the perfect multicollinearity of 24-h time budget data. The ensuing lack of adjustment for known effects on the outcome undermines the validity of study findings. We describe a statistical approach that enables the inclusion of all daily activity behaviours, based on the principles of compositional data analysis. Using data from the International Study of Childhood Obesity, Lifestyle and the Environment, we demonstrate the application of compositional multiple linear regression to estimate adiposity from children's daily activity behaviours expressed as isometric log-ratio coordinates. We present a novel method for predicting change in a continuous outcome based on relative changes within a composition, and for calculating associated confidence intervals to allow for statistical inference. The compositional data analysis presented overcomes the lack of adjustment that has plagued traditional statistical methods in the field, and provides robust and reliable insights into the health effects of daily activity behaviours.

Prolonged sitting at work is associated with a favorable time course of low-back pain among blue-collar workers: a prospective study in the DPhacto cohort

“…Thus, it was not possible to estimate risks among initially pain-free workers. Furthermore, this study showed small associations between sitting at work and time course of LBP, implicating that future studies should take the entire exposure from body postures and physical activities into account, like in a compositional data analysis (50).…”

Sedentary behavior patterns and adiposity in children: a study based on compositional data analysis

“…Given that the time in a day is constrained to 24 h, a change in the duration of one time-use component (e.g., time spent in PA) inevitably results in a change in the duration of one or more of the remaining time-use components (e.g., time spent in SB) [20]. In this context, in order to assess the association between SB and health, it is recommended to use a statistical approach that includes SB and PA variables in the same model [18,[20][21][22]. The use of compositional data analysis (CoDA) has been recommended over traditional multivariable models as it respects the compositional properties of the data by representing them as log ratios [22].…”

Compositional analysis of the associations between 24-h movement behaviours and cardio-metabolic risk factors in overweight and obese adults with pre-diabetes from the PREVIEW study: cross-sectional baseline analysis

“…In contrast, compositional analysis recognises the constrained nature of such data and uses log ratios to express the composition in terms of ratios of its parts. Conveying compositional data as log ratio coordinates transforms them from the constrained simplex to the unconstrained real space in which traditional multi variate statistics can be applied [12,14,15]. The shift towards compositional analysis in physical activity research was pioneered by Chastin et al [11], Carson et al [16], and Dumuid et al [4,15] and allows the examination of the combined effect of the activity composition on indicators of health.…”