EPI 320 Lecture Notes - Lecture 3: Observational Error, Confounding
Wednesday, 4.4
Review of previous lecture
• Epidemiologic research goal: identify risk factors for adverse health
outcomes (like death, cancer, etc.)
• How? Identify risk factor of interest (exposure) and calculate measures of
association (RR, OR, RD)
• When you find an association → is the association true?
Why would we see an association where none exists?
• Random error (chance)
• Bias (systematic errors): from the way we conduct studies/ask Q’s (non-
random samples, non-participation, problems w measurement)
• Confounding
What is confounding?
• Mixing of the effects of exposure and a third factor, the confounder, that’s
associated with both
Conditions necessary for confounding (confounder checklist)
• a factor “C” is a confounder when:
o it is a risk factor for the outcome (C → O)
o it is associated with the exposure within the study population (C → E)
o it is not caused by the exposure
• Refer to Clallam County example
The effect of confounding
• Does lots of damage to epidemiologic studies
o Exaggerate the truth, create spurious associations, diminish or hide
true association
Adjustment (standardization)
• Aims to make confounding go away
• We know: what happened in the real world – exposure and confounder
traveling together
• We ask: what would have happened if the exposed population looked like
the unexposed?