Questions: Effect Modification and Interaction in Epidemiology

Question 1

A study finds the overall RR for lung disease from asbestos exposure is 2.5. When stratified by smoking status, the RR in non-smokers is 1.8 and the RR in smokers is 12.0. What is the correct epidemiologic interpretation?

• A. Smoking is a confounder; the stratified results should be combined into one adjusted estimate to control for smoking's distortion
• B. Smoking is an effect modifier; the association between asbestos and lung disease genuinely differs across smoking strata, and the stratum-specific RRs should be reported rather than pooled
• C. The stratified results are less reliable than the overall RR because of reduced statistical power in smaller subgroups
• D. Both the overall RR and the stratified RRs should be reported, but the overall RR is the primary finding

Answer: (B)

When the exposure-disease association differs substantially across strata of a third variable, that variable is an effect modifier — and the heterogeneity is a finding, not a nuisance. Effect modification should be *reported* as stratum-specific estimates, not controlled for (which would obscure the real biological difference). Confounders distort associations and should be controlled; effect modifiers reveal genuine heterogeneity and should be reported separately. The practical test: if stratifying reveals real difference in the association, it's modification, not confounding.

Question 2

Two risk factors each independently double disease risk (RR = 2). Together they produce RR = 4. Which statement correctly characterizes this finding?

• A. There is multiplicative interaction, because the joint effect (4) exceeds either factor alone (2)
• B. There is no multiplicative interaction, because RR = 4 is exactly the product of the two individual RRs (2 × 2), which is the expected joint effect under independence
• C. There is additive interaction, because the absolute risk difference from the combination must exceed each factor's individual contribution
• D. The two factors are confounders of each other, since they have similar effect sizes

Answer: (B)

Under multiplicative independence, the joint RR equals the product of the individual RRs: 2 × 2 = 4. Finding RR = 4 means the two factors are multiplicatively independent — no multiplicative interaction. However, this does NOT mean there is no additive interaction. If each factor produces an absolute risk increase of 10 per 1000, but together they produce a 30 per 1000 increase (rather than the additive-independent expectation of 20), there IS additive interaction even without multiplicative interaction. The scales are distinct.

Question 3

Effect modification, like confounding, is a source of bias in epidemiologic studies and should be controlled for to obtain an unbiased overall estimate of effect.

• T. True
• F. False

Answer: False

This is the most important distinction in this topic. Confounding is a *bias* — it distorts the true association, should be identified, and should be controlled for. Effect modification is a *finding* — it reveals genuine biological heterogeneity in how the exposure affects different subgroups. Controlling for an effect modifier obscures a real phenomenon. The correct response to effect modification is to stratify and report subgroup-specific estimates, not to adjust them away.

Question 4

Additive interaction between two exposures can exist even when there is no multiplicative interaction, because the two scales measure different aspects of the joint effect.

• T. True
• F. False

Answer: True

The additive scale asks: does the joint absolute risk increase exceed the sum of each factor's individual absolute risk increase? The multiplicative scale asks: does the joint RR exceed the product of the individual RRs? These are mathematically independent questions, and factors can show interaction on one scale but not the other. The choice of scale is not arbitrary — for public health planning, additive interaction is often more relevant because it reveals the *number of preventable cases* attributable to the combination.

Question 5

Why do epidemiologists often prefer additive interaction over multiplicative interaction for public health decision-making, even though relative risks are the more commonly reported measure?

Model answer: Additive interaction measures whether the *absolute* risk difference caused by two exposures together exceeds the sum of their individual effects. This is directly relevant to public health because it quantifies the excess cases attributable to the combination that would not occur if either exposure were removed. Multiplicative interaction (comparing joint RR to the product of individual RRs) is useful for understanding biological mechanisms but does not directly answer 'how many cases would we prevent by addressing both exposures?' Decisions about resource allocation and intervention priority require absolute measures, not relative ones.

The practical consequence: two exposures with no multiplicative interaction can still produce substantial additive interaction — meaning their combination causes far more absolute cases than either would alone. A public health intervention targeting only one factor might prevent far fewer cases than expected if additive synergy is ignored. This is why absolute risk scales are essential for policy, even when relative risk scales are standard for reporting.