Mediation analysis

In epidemiology and related fields it is often of interest to understand to what extent a variable which lies in the pathway between an exposure and an outcome explains the association between them, which is often referred to as mediation analysis. This helps more generally to disentangle pathways via which risk factors cause disease. For example, socioeconomic status is associated with risk of cancer  and cardiovascular disease, and these associations may be partly explained by mediators such as smoking and other lifestyle factors; certain genetic variants are associated with risk of certain cancers, acting through changes in the function of proteins or other intermediate variables. Different approaches have been developed to assess mediation, such as methods based on path analysis, methods based on counterfactuals, interventional and separable effects.  

The aim of this project is to conduct a comprehensive review of different approaches to mediation, investigate common concepts and differences between the different approaches theoretically, and evaluate any differences in practice. 

The aspects considered are motivated by current epidemiological questions and data from large prospective cohort studies will be used to illustrate and compare the different approaches considered. Particular applications may be related to the associations of socioeconomic factors with risk of disease, and biological pathways associated with risk of cancer. 

Considerations related to multiple mediators, measurement error, and confounding will be explored, as well as aspects related to study design and type of outcome (e.g. continuous, categorical, time-to-event). 

 The project will involve developing skills in statistical theory and methods, analysis of large and complex datasets, literature review, and epidemiology.