Stephen Burgess – email@example.com
Causal inference, evidence synthesis
About the supervisor
I am a group leader in the MRC Biostatistics Unit and a senior scientist in the Cardiovascular Epidemiology Unit (part of the Department of Public Health and Primary Care). Our research group investigates questions about cause-and-effect relationships, particularly using an approach known as Mendelian randomization. We work at the interface between statistics and epidemiology, and aim to develop methods that are used to answer clinically relevant questions in innovative ways. Our group has a strong focus on applied analysis and methods dissemination as well as methods development, and we regularly run short teaching courses. More details on our group activities can be found at http://www.mendelianrandomization.com/.
We welcome applicants from a range of academic disciplines – students from more traditional backgrounds, such as mathematics, as well as students with other backgrounds wanting to transition into biostatistics. I hope that our group is a great place to join as a PhD student, but to be honest, all the other prospective supervisors have interesting projects too – I hope you can find a project that suits your interests!
Potential PhD projects
Investigating the shared genetic basis of different cancer types
Fifteen years of genome-wide association studies in cancer have identified hundreds of inherited genetic variants that associate with the risk of developing cancer but these studies have largely focused on one cancer type at a time. There is emerging interest and an urgent need for statistical methods development in the area of multi-cancer genetics, with far-reaching scope for application to multi-cancer prevention strategies and early detection tests. The aim of this project is to apply (and potentially develop) methodology for investigating inherited genetic associations that overlap across different cancer types, to provide insights into the shared genetic and biological mechanisms underpinning apparently diverse cancer types, as well as to identify overlaps with genetic associations for potentially modifiable risk factors for cancer. This could involve methodology for clustering, dimension reduction, Mendelian randomization, colocalisation, cross-trait linkage disequilibrium score (LD score) regression, and multi-trait polygenic scores (PGS). We currently have access to case-control studies involving over 500,000 cancer cases from international consortia, and prospective cohorts such as the UK Biobank, providing a rich substrate for statistical methods development, application, and validation. The starting point of the investigation will likely be open-ended and determined in partnership with the student; specific further questions of interest will evolve depending on findings.
This project will be co-supervised by Stephen Burgess (MRC Biostatistics Unit) and Siddhartha Kar (Early Cancer Institute, Department of Oncology, Cambridge).
Mendelian randomization and high-dimensional imaging datasets
Identifying causal relationships is of vital importance for understanding disease processes, and for prioritizing risk factors as potential therapeutic targets. It is well-known that “correlation is not causation”. But how then can causation be demonstrated? One potential approach is instrumental variable analysis. An instrumental variable is a factor that is not controlled by an experimenter, but behaves similarly to randomization in a randomized controlled trial to provide a natural experiment in an observational dataset. Naturally-occurring genetic variation represents a fertile source of plausible instrumental variables. The use of genetic variants as instrumental variables is known as Mendelian randomization.
The majority of Mendelian randomization analyses have considered a single risk factor and a single outcome. However, datasets with concomitant measurements on genetic variants, risk factors, and outcomes are increasing in terms of size and scale. This allows the possibility of performing more complex analyses to understand causal relationships between related variables, and to try to unpick complex relationships between risk factors and diseases. This project will involve developing and applying data-adaptive methods, such as those in the machine learning literature, to large-scale datasets with information on imaging phenotypes (primarily brain scans).
There are two directions that this PhD project could take. The first direction is more methodological, investigating and developing generally applicable statistical methods and advice for the analysis of imaging data in a Mendelian randomization framework. The second direction is more applied, considering a substantive question from the field of psychiatry that requires a complex analysis plan and advanced statistical techniques. The two directions are not mutually exclusive; a project may involve elements from both directions.
This project will be supervised by Stephen Burgess, with input from an applied perspective from Anya Topiwala (Oxford).
How to apply
For details of the MRC BSU application process please see How to apply
To be considered for funding applications need to be submitted to the University of Cambridge application system by 23:59 (GMT) on January 5th 2023