A Bayesian Neural Network Approach to Study the Systemic Structure of Obesity in Great Britain

Globally obesity has more than doubled since 1980. In the UK, obesity represents a major public health challenge. The prevalence of adult obesity has increased from 7% in 1980 to 25% in 2013 (Ng et al. 2014). The high prevalence and long-term latency in development of health conditions places a considerable burden on health care. While a series of policies have been put in place which have attenuated the rate of increases in obesity prevalence, they have yet to reverse trends (Jebb et al. 2013). This situation is partly because an individual's body weight is influenced by a complex multi-factorial aetiology: dietary behaviours, psychology, genetics, food production, physical activity and environmental factors. While we have a sound understanding of their independent influences on obesity, empirical work has remained fragmented and often ignores how they interact..

A comprehensive study undertaken for the UK government (Vandenbroeck et al. 2007) highlighted the complex underlying structure of factors associated with obesity and the need for a whole systems approach to obesity reduction. The study developed the Obesity Systems Map (OSM) which identified the links between all known factors of positive energy balance (i.e. risk of obesity) in seven clusters, comprising factors of influence at the individual-level and local contextual levels. While OSM represents a useful conceptual framework to guide the formulation of policy responses, it has remained descriptive in nature.

No concerted attempt has been made to translate the OSM into an empirically testable model. This is important to analyse the multiple causal relationships between the various determinants of obesity and determine their relative importance on increasing the risk of obesity. Drawing on recent developments in machine learning and linking separate surveys dedicated to individual OSM components, this project seeks to develop an integrated analytical approach to better understand the systemic structure of obesity.

Specifically, this project aims to build a unified statistical model to better understand the multiple systems of risk factors that influence obesity in Great Britain. The project has the following objectives:
Develop an empirically testable model of the OSM;
Build a synthetic dataset combining official surveys covering distinct components of the OSM;
Determine the systemic causal relationships in the OSM using Bayesian Neural Networks (BNNs);
Identify the key risk factors of obesity in each of the seven OSM clusters;
Determine the system-wide relative importance of key risk factors in influencing obesity.

By addressing these objectives, the project will produce the first whole systems analysis of the causes underlying the risk of obesity. The novelty of assessing the interrelationships between multiple systems rarely considered will provide theoretical insights into their causality. The project will also deliver an innovative statistical framework to guide policy development and identify critical areas of intervention for obesity reduction. The use of machine learning techniques will provide a novel application of advanced methods often absent from population studies into the drivers of obesity.


Bibliography
Vandenbroeck, P, Goossens, J, Clemens, M, 2007. Foresight. Tackling Obesities: Future Choices -Building the Obesity System Map, Government Office for Science, London, UK.
Jebb S, Aveyard P, Hawkes C. 2013. The evolution of policy and actions to tackle obesity in England. Obesity Reviews, 14:42-59.
Ng M, Fleming T, Robinson M, et al. 2014. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet, 348:766-781.