Confounding
Defintion?
How Do You Look For it?
What do you do about it?
Definition: bias in the estimation of effect due to lack of comparability between exposure groups
How to Look for it:
- stratify data and compare stratum specific effect measures to the crude effect measures
- Strata with similar OR’s/RR’s but different from crude OR/RR = implies confounding
What do you DO about it?
- adjust for it
- must always check for interaction before attempting to adjust for confounding of the same factor
Interaction
Defintion?
How Do You Look For it?
What do you do about it?
Definition: Heterogeneity of effect, increases knowledge of biological mechanism
How to Look for it:
- stratify data and compare stratum-specific effect measures
- Different OR’s/RR’s within strata = implies interaction
- Summary OR (or crude OR) is not appropriate representation of effect
What do you DO about it?
- Interaction effect is a finding to be reported, not avoided
Interaction
Dichotomous Variables
Continuous Variables
Dichotomous Variables - interaction means that the effect of the exposure variables on the outcome variables differes depending on whether another variable (effect modifier) is present
Continuous Variables - interaction means that the effect of exposure on outcome depends on the level of another variable (rather than presence/absence)
Effect Modification:
Synergistic
Antagonistic
Effect Modifier and Exposure are synergistic if the presence of the effect modifier strengthens the effect of the exposure
(positive interaction)
Effect Modifier and Exposure are antagonistic if the presence of the effect modifier diminishes/eliminates the effect of the exposure
(negative interaction)
2 Ways to Define Interaction
- Definition based on heterogeneity of effects:
- -interaction occurs when the effect of E on O is not homogenous in strata formed by EM - Definition based on comparison between observed and expected joint effects of the exposure and effect modifier
- -interaction occurs when the observed and expected joint effect of EM and E differ.
- -expected joint effect is based on EM and E’s independent effects
How to measure Interaction
- Additive Interaction Assessment
- Multiplicative Interaction Assessment
- Additive Interaction Assessment - effect is measured on an additive or absolute scale (AR) - based on absolute differences between CI or rates
- Multiplicative Interaction Assessment - effect is measured on a relative (ratio) scale (RR, OR)
Evaluating Interaction in Cohort Studies
Strategy 1: Homogeneity / Heterogeneity of Effects
Interaction is evaluated by comparing the strata-specific AR’s / RR’s
–If they are heterogenous, The effect of E is modified by the presence of Effect Modifier = interaction is present
additive - AR
multiplicative - RR
Evaluating Interaction in Cohort Studies
Strategy 2: Comparison of Joint Effects
Additive Interaction
Expected Joint Effect = AR(E+Z-) + AR(E-Z+)
(independent effect of E + independent effect of Z)
Observed Joint Effect = CI(E+Z+) - CI(E-Z-)
(doesn’t have to be CI, can be I, IR)
Multiplicative Interaction
Expected Joint Effect = RR(E+Z-) x RR(E-Z+)
Observed Joint Effect = RR (E+Z+) = CI(E+Z+) / CI(E-Z-)
Evaluating Interaction in Case Control Studies
Strategy 1: Homogeneity / Heterogeneity of Effects
Multiplicative Interaction is assessed by comparing the OR(Z-) to OR(Z+) , or the effect of E in the absence of Z and the effect of E in the presence of Z
*no additive interaction
Evaluating Interaction in Case Control Studies
Strategy 2: Comparison of Joint Effects
Additive Interaction
Expected OR = OR(E+Z-) + OR(E-Z+) - 1.0
Observed OR = OR++
Multiplicative Interaction
Expected OR = OR+- x OR -+
Observed OR = OR ++
Difference between Effect Modification and Confounding
Effect Modification refers to heterogeneity of effect
Confounding refers to bias in the effect estimate due to lack of comparability between exposure groups
