Premium
Directed acyclic graphs (DAGs): an aid to assess confounding in dental research
Author(s) -
Merchant Anwar T.,
Pitiphat Waranuch
Publication year - 2002
Publication title -
community dentistry and oral epidemiology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.061
H-Index - 101
eISSN - 1600-0528
pISSN - 0301-5661
DOI - 10.1034/j.1600-0528.2002.00008.x
Subject(s) - confounding , observational study , directed acyclic graph , medicine , causal inference , outcome (game theory) , computer science , mathematics , algorithm , pathology , mathematical economics
– Confounding, a special type of bias, occurs when an extraneous factor is associated with the exposure and independently affects the outcome. In order to get an unbiased estimate of the exposure–outcome relationship, we need to identify potential confounders, collect information on them, design appropriate studies, and adjust for confounding in data analysis. However, it is not always clear which variables to collect information on and adjust for in the analyses. Inappropriate adjustment for confounding can even introduce bias where none existed. Directed acyclic graphs (DAGs) provide a method to select potential confounders and minimize bias in the design and analysis of epidemiological studies. DAGs have been used extensively in expert systems and robotics. Robins (1987) introduced the application of DAGs in epidemiology to overcome shortcomings of traditional methods to control for confounding, especially as they related to unmeasured confounding. DAGs provide a quick and visual way to assess confounding without making parametric assumptions. We introduce DAGs, starting with definitions and rules for basic manipulation, stressing more on applications than theory. We then demonstrate their application in the control of confounding through examples of observational and cross‐sectional epidemiological studies.