About the Project
For informal enquiries, please contact Dr James Bentham at [Email Address Removed].
Aim: The student will develop investigate patterns in multiple risk factors for non-communicable diseases (NCDs), and develop statistical models of their joint distributions.
Background: NCDs include cancers and heart disease, and are responsible for 70% of global deaths. There are various major factors that affect risk of developing these diseases, including how a person’s weight compares to their height (e.g. their body-mass index), their blood pressure and cholesterol levels, their height, and whether they have diabetes or smoke.
Bayesian hierarchical models of single risk factors have been developed and fitted using MCMC (1), and we have published estimates by country over the past 40 years for body-mass index (BMI) (2), diabetes (3), and blood pressure in adults (4), and BMI in children (5). The next step is to understand how multiple factors are distributed within populations. This is important because risk of developing NCDs depends on the interaction of these variables. For example, risk may be higher in someone with moderate levels of all the factors than in a person with high levels of one factor and low levels of the others.
Student’s expected contribution: The student will explore the data describing these multiple risk factors, to develop an understanding of how they relate to one another, and how this changes over time. In the early part of the project, the student will investigate and model the relationship between mean BMI and mean height in children. This is likely to involve development of extensions to existing Bayesian hierarchical models. Later, the student will select appropriate model types for fitting whole joint distributions. One option would be to apply mixture models using MCMC, which can produce flexible fits to non-normal data using weighted combinations of Gaussian distributions. Another possibility would be to apply deep learning methods to the data, which have been used successfully in other fields, and which are appropriate for this type of complex system.
Skills required: The student will require an MSc in Statistics or equivalent, and a strong interest in Bayesian hierarchical modelling and MCMC methods. A strong interest in programming is essential, and excellent programming skills will be developed during the project: the existing models are fitted using R, but the student will need to explore other platforms, such as Stan or NIMBLE, or could code their own samplers using a low-level language.
The project will be funded by SMSAS, and the student will be based in the School. The student will be co-supervised by Professor Majid Ezzati at Imperial College, who leads the NCD Risk Factor Collaboration, which will provide the data for this project.
(1) Finucane, M and colleagues. Statistical Science 2014; 29: 18-25.
(2) NCD Risk Factor Collaboration. Lancet 2016; 387: 1377-1396.
(3) NCD Risk Factor Collaboration. Lancet 2016; 387: 1513-1530.
(4) NCD Risk Factor Collaboration. Lancet 2017; 389: 37-55.
(5) NCD Risk Factor Collaboration. Lancet 2017; published online 10th October.
Continue with Facebook
