Confounding, Interaction, and Mediation Flashcards
What is a mediator?
Associated with outcome and associated with exposure and ON causal pathway.
How do you distinguish between mediators and confounders?
Through THEORY.
Cannot be done through analysis.
Mediator is meaningful piece of causal pathway.
What is effect modification?
When magnitude of effect of exposure on outcome depends on levels of a third varaible.
When effect of exposure on outcome is different across strata of a third variable.
What is biological interaction?
Interdependent operation of two or more biological causes to produce, prevent, or control an event.
What is statistical interaction?
Assess effect modification quantitatively with a joint interaction term of the exposure and third variable in the statistical analysis.
What is the most common type of modification in epi?
Effect measure modification.
When measure of association changes due to a third variable.
Why is it important to study?
- Know how to allocate resources.
- Tailor and target interventions.
What happens when we use Mantel-Haenszel method?
Pool OR/RR in order to hold the third variable constant.
When do we not want to use MH method?
When effect modification or mediation in place. We only want to use to control for confounding.
How do we see if effect modification present in continuous variables?
We use an interaction term in linear regression.
What is an absolute versus relative measure of association?
Risk, rate, prevalance, and odds are measures of association.
When 1/another then relative (ratio).
When 1-another, then absolute.
When is effect modification additive versus multiplicative?
Multiplicative = relative measure.
Additive = absolute measure.
Opposite of how you measure.
What is additive/multiplicative expected risk versus observed risk?
Expected risk is when one exposure is experienced (i.e. smoking and no asbestos or no smoking and asbestos ).
Observed risk is when both are experienced (smoker with asbestos risk (add risks together).
If observed is greater than expected, then statistical interaction on additive scale.
What happens when observed risk is less than expected risk?
Could be protective factors.
Modifies.
What is the statistical test for interaction in OR?
Breslow-Day test.
Tests for homogeneity of strata-specific odds ratio.
