Download printable overview of the workshop here!

Directed Acyclic Graphs (DAGs) applied to veterinary epidemiology

Session organisers: Locksley Messam1, BSc, DVM, PhD  and Hsin-Yi Weng2, BVMS, MPH, PhD

1 Section: Herd Health and Animal Husbandry, School of Veterinary Medicine, University College Dublin, Dublin, Ireland (locksley.messam@ucd.ie)
2 Department of Comparative Pathobiology, College of Veterinary Medicine, Purdue University, West Lafayette, Indiana, USA (weng9@purdue.edu)


Introduction

Directed Acyclic Graphs (DAGs) are widely accepted by the epidemiologic community as useful tools for combatting confounding in observational studies. They can be used in both the design and analytic phases of a study and provide a visual, transparent and logically coherent way of identifying potential confounders of exposure-outcome relationships. This includes the identification of variables which meet traditional criteria for confounding but will, if adjusted for, cause bias in estimates. 

Assumed knowledge

We assume that participants will be familiar with: 

  • The definition of a confounder and 
  • The interpretation of the effects of an exposure on an outcome, based on a multiple linear or logistic regression equation.

Workshop structure: Two parts

Part I: Participants will initially be introduced to the role of causal assumptions in making inferences from observational studies, the terminology of DAGs, their construction, interpretation and applications. 

Part II: We will then split into groups of 4-5 persons and work on the construction, interpretation and solution of DAGs using provided examples. The focus will be on practice in identifying appropriate subsets of variables for confounder control.

Outcomes

At the end of this workshop, each participant should:

  1. Be familiar with terminology relevant to the construction of a DAG.
  2. Be able to independently draw a DAG as a means of articulating causal assumptions, when estimating the effect of an exposure on an outcome.
  3. Be able, given a DAG, to solve it in order to identify different sets of variables for statistical control, depending on the exposure of interest. 
  4. Be able to use a DAG to identify variables which will cause biased estimates when adjusted for.

Materials (to be provided)

  • Workshop lecture notes.
  • Exercises (and solutions) for group practice and discussion.
  • A supplementary reading list for DAGs

Maximum number of participants

30


Instructors

Locksley Messam

Locksley’s research interests include: Human-Animal interactions and their effects on human health and wellbeing, principles and applications of diagnostic test interpretation in veterinary medicine, and the application of  epidemiologic methods and approaches used in other fields to veterinary medicine. He enjoys teaching and is currently a Lecturer at the University College Dublin where he teaches a course on epidemiology to veterinary students and participates in epidemiology instruction in the MPH programme.

Hsin-Yi Weng

Hsin-Yi‘s main research interests are in applying epidemiologic methods to studies promoting animal health and welfare, human-animal interactions, and public health. She is an Assistant Professor of Clinical/Analytical Epidemiology at Purdue University and currently teaches a DVM epidemiology course and a graduate level course focusing on design and analysis of epidemiological studies.