Use of Causal Diagrams in Variable Selection for Causal Observational Studies

Deciding which variables to adjust for when addressing causal questions in observational studies can be challenging. For example, lack of adjustment for some variables might lead to sub-optimal control for confounding whereas overadjustment for other variables can in fact introduce bias to a study. In recent years, causal diagrams have become popular tools in epidemiology that can guide researchers in better understanding the causal structure of a study question. Causal diagrams can act as blue prints for variable selection for causal questions and can help researchers better understand the role of different variables and when, if at all, to adjust for them. This presentation will introduce causal diagrams, how they work and how they can guide researchers in identifying variables such as confounders, colliders and meditators. A number of examples and scenarios will be presented for each type of variable. Strengths and limitations of causal diagrams will also be discussed.