Abstract
Background: Compositional Data Analysis (CoDA) describes a statistical methodology for handling relative data such as proportions of time spent in different behaviours during the 24-hour day. This was previously only applied to cross sectional studies, and techniques for communicating results were at an early stage in physical activity (PA) research.Objective: This thesis extends CoDA to prospective studies and explores ideas to improve communication of results.
Methods: Cross sectional studies based on physical activity research investigating the association between (i) composition of the 24-hour day using data from the Canadian Health Measure Survey (CHMS) and cardiometabolic health biomarkers, and (ii) composition of the waking day using data from Health Survey for England 2008 and cardiometabolic health biomarkers were performed. Multiple methods of communicating results both in terms of reported results and data visualization were attempted. The success of these techniques was reviewed retrospectively. Cox regression with compositional explanatory variables was developed and applied in a prospective study investigating the association between composition of the 24-hour day and all-cause mortality using data from National Health and Nutrition Examination Survey 2005-06 Cycle. An online platform implementing this and other CoDA methods was established, and used in a federated analysis investigating the association between composition of the 24-hour day and all-cause mortality by combining data from multiple international studies.
Results: The main contribution of this work is to expand CoDA methods in the epidemiology of PA and provide the ability to estimate the joint associations between all 24 hours movement behaviours and health. The most consistent amongst those found is a strong association between the ratio of moderate-to-vigorous physical activity (MVPA) and sedentary behaviour (SB) and positive health outcomes. In terms of communicating results, visualization approaches using traditional graphs and untransformed variables were most successful, and conclusions based on illustrations or plausible scenarios were more readily grasped.
| Date of Award | 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Sebastien Chastin (Supervisor) & Philippa Dall (Supervisor) |