Many apparently robust observational findings in medical research have been refuted by randomized controlled trials (RCTs), largely because common analysis methods are sensitive to unmeasured confounding. Methods to examine causality, even in the presence of unmeasured confounding, include Mendelian randomization (MR) (1, 2), which uses the particular properties of germline genetic variation to strengthen causal inference regarding modifiable exposures. Other approaches include regression discontinuity designs, non-genetic instrumental variables (IVs) analyses, and natural experiments. Each of these approaches require the existence of an instrument for the exposure, and this may not always exist. This PhD will investigate a novel method to strengthen causal inference through using genetic information to estimate the contribution of known and unknown confounders and apply this to estimate plausible causal effects of exposures. Findings from this approach will be triangulated (3) with evidence from other approaches and data synthesised using multi-parameter evidence synthesis (MPES). The research will be both methodological and applied, with the applications being developed and disciplinary area they are within being selected by the student.
Aims and objectives
- Formulate the way in which genetics can be used to estimate underlying confounding by known confounders through directed acyclic graphs (DAGs).
- Choose an exemplar case for which data are available (e.g. in UK Biobank, with available summary data from genome-wide association studies [GWAS], etc.).
- Develop and apply proposed methods for estimating the influence of known confounders on the association between the exposure and outcome in these data.
- Explore whether “GWAS by subtraction” (4) could plausibly be used to investigate unknown confounders.
- Explore whether MPES can be used to synthesise findings from this approach with existing evidence.
Methodology
The methods and technical skills the student will learn and use include:
- Constructing DAGs formally representing the causal structure of variables in epidemiological and related investigations.
- Analyses of longitudinal and/or case-control data using conventional statistical approaches and applying methods intending to improve causal inference in non-genetic epidemiological data, potentially including use of marginal structural models, G-formulae based approaches, and structural equation modelling, amongst others.
- Learning the principles of MR, including that of gene-environment equivalence and the IV conditions required when an IV analysis is used.
- Constructing and using polygenic risk scores and performing LD score regression and other analytical approaches when these can help illuminate the structure of the data.
- Becoming acquainted with the many sensitivity analyses available for MR analyses and applying these.
- Using negative control and zero-relevance point approaches in MR.
- Becoming acquainted with non-MR approaches to strengthening causal inference in observational data.
- Literature searching approaches for an unbiased assessment of the published evidence on a particular topic.
- Data extraction methods for systematic reviewing.
- Fundamentals of MPES.
How to apply for this project
This project will be based in Bristol Medical School - Population Health Sciences in the Faculty of Health Sciences at the University of Bristol.
Please visit the Faculty of Health Sciences website for details of how to apply
Continue with Facebook
