As environmental epidemiologists struggle to find the most appropriate model to answer important etiological questions, directed acyclic graphs (DAGs) provide an excellent tool to find the minimal adjustment set, sources of biases, and an unbiased model for inference, while clearly specifying the assumptions of the model. Model specification based on DAGs allow researchers to clearly address issues of confounding, residual confounding, selection bias, overadjustment, collinearity, and intermediate variables that also act as confounders. However, much of the discussion regarding DAGs is theoretical, and examples of how to utilize DAGs in typical research settings are limited. Therefore, for the proposed workshop, we aim to provide 1) an introduction to DAG theory focusing on recent advancements regarding overadjustment and collinearity (i.e., issues particularly common in environmental epidemiologic settings), 2) specific environmental epidemiological examples in which DAGs have been applied, and 3) details on how to use a well- constructed DAG to specify an appropriate statistical model for causal inference. Specifically, we will show how to construct a DAG based on a well- defined research question, determine the appropriate covariate adjustment set using DAG software, when and how to apply marginal structural models for causal inference, provide SAS code for estimation, and interpret results using examples relevant to environmental epidemiology. Examples will further consider how to evaluate the assumptions of marginal structural models, and how to apply marginal structural models in a longitudinal data setting. This workshop will help bridge the gap between methodological advancements and practical applications in research settings by providing real world examples and tools for implementation.
- To provide a background on recent advancements in directed acyclic graphs regarding overadjustment and collinearity.
- To provide hands-on examples of practical applications of directed acyclic graphs to studies of environmental epidemiology.
- To provide helpful tools to construct and interpret directed acyclic graphs and determine the minimal adjustment set.
- To demonstrate when and how to apply marginal structural models for causal inference.
- To bridge the gap between methodological advancements and practical applications in research settings.