Effect Modification

Question 1

Definition of effect modification

Answer 1

The magnitude and/or direction of association changes between different groups according to levels of an extraneous factor

Question 2

How do we identify potential modifiers?

Answer 2

1. Knowledge of the biology of the disease under study (Known population subgroups with varying disease risks)
2. Past experience in a similar study you conducted
3. Published findings from other similar studies
4. Data-specific assessment

Question 3

How to do the data-specific assessment to identify potential modifiers?

Answer 3

Step 1: Compare overall association with the association stratified on levels of the potential effect modifier (Stratified analysis: Mantel Haenszel)

If the stratified RRs are very different between subgroups, then no reason to calculate the adjusted RR as a confounder. The factor may be a modifier. We then report stratum-specific estimates and CIs to illustrate modifying effect of third factor

Step 2: Statistical hypothesis test for effect modification (Cross-product interaction test; Multivariable regression)

Question 4

Interpret the modifier result.

Answer 4

The association between obesity and CVD mortality is modified by physical fitness level (interaction P<0.05).

The association between obesity and CVD mortality is attenuated to 1.13 among fit men, and is strengthened to 3.54 among low fit men.

Question 5

T/F: Like confounding, effect modification is a real phenomenon reflecting complex interrelationships of exposures and outcomes

Answer 5

True

Question 6

T/F: Effect Modification is not fully a matter of statistical significance

Answer 6

True. We should not judge effect modification ONLY by statistical test of interaction.

Need to examine distribution of cell counts within stratum

• Insufficient sample size (cell counts) can lead to incorrect

conclusion when interpreting the interaction