Mediation Analysis Workshop
11 March 2016
An extremely popular event sponsored by the Society for Social Medicine in conjunction with the INTERSTELA Project held in Bristol.
This workshop outlined the challenges of traditional methods for mediation analysis and some of the novel approaches based on counterfactual theory or instrumental variable methods including Mendelian Randomization.
Mediation analysis is widely used to understand the pathways and mechanisms linking exposures and outcomes. Traditional methods for mediation analysis have several limitations including:
- susceptibility to confounding and bias when the mediator is measured with error
- interactions between the exposure and mediator are not accounted for
- collider bias may be induced by conditioning on the mediator in the presence of mediator-outcome confounders
- the models cannot be estimated if a descendent (consequence) of the exposure confounds the mediator-outcome association (‘intermediate confounding’).
The sessions included in the workshop were:
- Mediation analysis and pitfalls of traditional methods
- Novel mediation methods based on counterfactual theory
- Instrumental variable methods for mediation analysis, with applied examples from epigenetic epidemiology
- Facilitated discussion.
- Richard Emsley from University of Manchester
- Rhian Daniel from London School of Hygiene and Tropical Medicine
- Rebecca Richmond from University of Bristol
- Kate Tilling from University of Bristol