An overview of confounding. Part 2: how to identify it and
PENELOPE P. HOWARDS
Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, GA, USA
Bias, causality, confounding factors
(epidemiology), data analysis, epidemiologic
methods, epidemiologic research design
Penelope Howards, Department of
Epidemiology, 1518 Clifton Rd. NE, CNR
3029, Atlanta, GA 30306, USA.
Conﬂict of interest
The author has stated explicitly that she has
no conﬂicts of interest with this article.
Please cite this article as: Howards PP. An
overview of confounding. Part 2: how to
identify it and special situations. Acta Obstet
Gynecol Scand. 2018
Received: 21 December 2017
Accepted: 28 December 2017
Confounding biases study results when the effect of the exposure on the
outcome mixes with the effects of other risk and protective factors for the
outcome that are present differentially by exposure status. However, not all
differences between the exposed and unexposed group cause confounding.
Thus, sources of confounding must be identiﬁed before they can be addressed.
Confounding is absent in an ideal study where all of the population of interest
is exposed in one universe and is unexposed in a parallel universe. In an actual
study, an observed unexposed population represents the unobserved parallel
universe. Thinking about differences between this substitute population and
the unexposed parallel universe helps identify sources of confounding. These
differences can then be represented in a diagram that shows how risk and
protective factors for the outcome are related to the exposure. Sources of
confounding identiﬁed in the diagram should be addressed analytically and
through study design. However, treating all factors that differ by exposure
status as confounders without considering the structure of their relation to the
exposure can introduce bias. For example, conditions affected by the exposure
are not confounders. There are also special types of confounding, such as time-
varying confounding and unﬁxable confounding. It is important to evaluate
carefully whether factors of interest contribute to confounding because bias can
be introduced both by ignoring potential confounders and by adjusting for
factors that are not confounders. The resulting bias can result in misleading
conclusions about the effect of the exposure of interest on the outcome.
DAG, directed acyclic graph.
Confounding is one of three types of bias that can distort
the results of epidemiologic studies and potentially lead
to erroneous conclusions. In the companion paper in this
journal (1), we discuss how confounding occurs and how
to address it. In short, confounding can be considered the
confusion of the effect of the exposure on the outcome
with the effects of other risk and protective factors for
the outcome that are present differentially by exposure
status (2). The ideal study design would avoid confound-
ing through the use of parallel universes where all study
participants are exposed in one universe and unexposed
in the other universe. Only exposure status and
consequences of being exposed would differ across the
universes. Under these conditions, if the outcome were
more common in one universe than another, this differ-
ence would be caused by the exposure and would not be
the result of other risk or protective factors for the
Diagramming the relations among the exposure, out-
come, and other factors affecting the outcome or
exposure can help identify sources of confounding.
Confounding can be addressed, but bias can occur
from treating a non-confounder as a confounder.
ª 2018 Nordic Federation of Societies of Obstetrics and Gynecology, Acta Obstetricia et Gynecologica Scandinavica 97 (2018) 400–406