A collider is a variable that arises independently from a cause and a purported effect.
Collider bias:
(1) can arise during study design or statistical analysis
(2) can indicate an association between the cause and effect when none exists, or distort the association if one exists
Alternative terms:
(1) selection bias in sampling
(2) admission rate bias (2 conditions that can cause a patient to be hospitalized may appear to be related if a study is performed on hospitalized patients)
Related biases:
(1) confounder bias: the confounder can cause both the exposure and the outcome
The phenomenon can be visualized by the use of a causal diagram (directed acyclic graph or DAG).
Collider bias can be prevented through proper use of inclusion criteria. Exposure and outcome should not drive inclusion or selective retention of subjects.
To read more or access our algorithms and calculators, please log in or register.
