Abstract
Mendelian randomisation (MR) is an observational epidemiological method that uses genetic predisposition to an exposure as a proxy for randomisation. In recent years, methodological developments and advances in data accessibility have made MR easier to conduct, and have thus made it increasingly popular. However, this means it is more likely that MR may be conducted by those with limited understanding of its assumptions, and very little is known about the current levels of reproducibility in MR studies.I conducted four studies to assess reproducibility in the field of MR, using data from the UK Biobank (a prospective cohort of half-a-million, middle-aged, British individuals, which contains genetic, lifestyle and medical data). I outline approaches and methods for improving reproducibility in MR. I found that the quality of 116 published MR studies conducted in the UK Biobank ranged from low to moderate across 25 items of reporting and methodological quality, and this was not influenced by journal characteristics (Chapter 2). I explored how varying statistical decisions impact the reliability of MR results, and outlined how a multiverse approach can reduce the impact of this (Chapter 3). I then outline a pair of novel sensitivity measures to explore bias in MR due to correlated pleiotropy, i.e., when the genetic instrument acts on the exposure via confounders of the exposure-outcome relationship (Chapter 4). Finally, I develop another sensitivity analysis method which uses MR to address regression dilution bias, i.e., when measurement error in covariates inflates the causal estimate of the exposure on the outcome, to be used when standard MR methods cannot be applied due to correlated pleiotropy and weak instrument bias (Chapter 5).
I conclude that reporting transparency and reproducibility in MR research needs to be improved, and I outline potential methods we can use to improve the robustness of MR results.
| Date of Award | 23 Jan 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Marcus R Munafo (Supervisor), Rebecca Richmond (Supervisor) & Jasmine N Khouja (Supervisor) |