12.6 - Why study interaction and effect modification?

  1. Increasing precision of description.

    Example: Lung Cancer Rates

    Below are lung cancer rates per 100,000 person years in males and females stratified by smoking status.

      Male Female
    Smokers 400 200
    Nonsmokers 50 50

    The relative risk of smoking in females: 200/50 = 4.00
    The relative risk of smoking in males: 400/50 = 8.00

    There is a effect modification by sex! It looks like smoking is twice as bad for males as it is for women. Males are twice as likely to develop along cancer from smoking.

    • Interpretation 1: Women appear to be less susceptible to the carcinogenic effects of cigarette smoking; or,
    • Interpretation 2: Women smokers did not inhale as deeply as male smokers, and thus received a lower dose of actual exposure and resulted in lower risk.

    This is an example of why we consider strata effects to give us a clue as to the biological or behavioral factors. The above illustration of the different strength of the smoking-lung cancer association between men and women may lead to an appreciation of a need to increase precision in the measurement and specification of the exposure variable.

  2. Indicating the need to control for effect modifiers:

    Since an effect modifier changes the strength of the association under study, different study populations may yield different results concerning the association of interest. For instance, you might need to present separate models for men and women. This is important because, unlike potential confounders, modifying variables cannot create the appearance of an association where none exists, nor obscure an association where one does. But the proportion of the study population that has a greater susceptibility will influence the strength of the association. Therefore, to achieve comparability across studies, it is necessary to control for the effect of the modifying variables.

  3. Developing etiological hypotheses:

    Attention to effect modification in the data may lead to the formulation of etiologic hypotheses that advance our understanding of the pathogenetic processes involved. For example, the interaction of smoking and asbestos suggests the possibility of impairment of lung clearing processes and/or of mechanical injury from asbestos particles increases susceptibility to carcinogens in cigarette smoke.

  4. Defining subgroups for preventive approaches:

    The synergistic relationship between smoking and asbestos in the etiology of lung cancer suggests particular efforts to convince asbestos workers not to smoke. The preventive approach for men from women may be different based upon the results of this study.