Causal mediation analysis offers a sophisticated method to understand the mechanisms through which an intervention or treatment exerts its effects on an outcome. This type of analysis is crucial for identifying the indirect effects mediated through specific variables, allowing researchers to dissect the causal pathways and potentially target these mechanisms more effectively in future interventions.

Causal mediation analysis extends beyond conventional mediation analysis by providing a framework to evaluate potential causal roles of mediators. It enables researchers to estimate both direct effects of a treatment or exposure on an outcome and indirect effects that operate through one or more mediators. This is particularly important in fields like public health, education, and social sciences, where understanding the underlying processes can lead to more effective interventions and policies.

This seminar will focus on some of the recent developments in causal mediation analysis and will provide practical tools to implement those techniques. We will discuss the relationship between traditional methods for mediation in the biomedical and social sciences and new methods of causal inference for dichotomous, continuous, and time-to-event outcomes.

We will approach concepts and methods for mediation from the perspective of the counterfactual framework. Definitions, theoretical identification results, and statistical techniques related to mediation analysis will be covered.

In the first part, we will clarify the no-confounding assumptions required for the estimation of direct and indirect effect and will also consider when standard approaches to mediation analysis are valid and when they are not valid. These approaches will be extended to more complex settings, such as in the presence of interactions, non-linearities and time-varying exposures.

The second part will cover selected advanced topics: multiple mediators, multiple exposures, time-to-event outcomes.

Upon successful completion of this course, you should be able to:

• • Explain when traditional methods for mediation fail.
  • Define the concepts about mediation from causal inference.
  • Conduct regression methods for mediation with single mediators and time-to-event outcomes and interpret results of such analyses.
  • Discuss strategies to conduct mediation analysis with multiple mediators and multiple exposures.