UCLA

Applied epidemiologists are required to not only address causal aims but descriptive and predictive aims as well. There is a lack of guidance on how to approach aims that are not obviously causal with the causal tools and methods that epidemiologists are often trained in. Directed Acyclic Graphs (DAGs) are used in epidemiology and clinical research to clarify assumptions and illustrate causal questions to inform study design and statistical analysis. However, there is little guidance on the use of DAGs outside of causal inference. This dissertation aims to address this gap by walking through the use of DAGs while navigating and adapting previously developed frameworks. In chapter 1, we provide the background and general approach of the dissertation. In chapters 2-4, we adapt an existing framework to provide guidance on the use of DAGs to address descriptive, predictive, and causal aims, respectively. We demonstrate the application of DAGs by working through an example aim using data from the National Health and Nutrition Examination Survey I (NHANES-I) Epidemiologic Follow-up Study (NHEFS) as used in Causal Inference: What If. Lastly, chapter 5 provides a brief discussion of the similarities and differences in addressing these types of aims. We found that the importance of the target population is prevalent in any type of study. Similarly, selection bias, information bias, and missing data issues can arise in any study whereas confounding may not be as much of a concern in descriptive and some predictive studies. DAGs are useful to communicate and address these uncertainties.