Foundations in Causal Inference

Description:
Certain areas within pharmacoepidemiology, particularly comparative effectiveness and safety studies, rely on estimating causal effects of medications (or vaccines or medical devices) on outcomes using observational data, even if the objective is not explicitly stated as such. Over the past four decades, a formalized language and discipline of causal inference has been developed that can be useful in designing, implementing, and communicating observational research, including pharmacoepidemiology studies. With a mix of didactic teaching and hands-on activities, this course will introduce attendees to the terminology and concepts of causal inference and equip them with tools to apply causal thinking to pharmacoepidemiology questions.

Requisites Statement:
Entry Level

Objectives:
Objective 1: Develop fluency in the terminology, notation, and basic concepts of causal inference
Objective 2: Become proficient in using directed acyclic graphs (DAGs) to represent and communicate structural biases in observational studies
Objective 3: Apply the target trial paradigm to the design and implementation of a pharmacoepidemiology study

Outline:
Presentation 1: The first session will introduce the attendees to basic concepts and terminology of causal inference (e.g., association versus causation, counterfactuals), and the assumptions of causal inference, including exchangeability, positivity, and consistency, in the context of pharmacoepidemiology research.
Presentation 2: The second session will introduce directed acyclic graphs (DAGs) and the notion of d-separation as a way to depict structural biases (confounding, selection bias and information bias) in observational data. Attendees will engage in a hands-on activity to practice discussing and drawing DAGs to represent several real-world pharmacoepidemiology study questions.
Presentation 3: The third session will describe the target trial paradigm as an approach for designing and analyzing pharmacoepidemiology studies to emulate a hypothetical randomized controlled trial. In an interactive exercise, attendees will translate elements of such a hypothetical target trial into parameters for an observational study that could be implemented using real-world healthcare databases.