A cause should not be automatically taken as an effect modifier of other causes: author's reply

摘要: We appreciate Diaz-Quijano’s1 clear elaboration of the scale-dependence of effect measure modification, illustrating that two factors that each influence a given outcome will exhibit interaction on either the additive or multiplicative scale, and possibly both. We agree that, although this is mathematically true, the implications vary on a case-by-case basis.To begin, the analyst must decide if evaluating for interaction is necessary. There are scenarios for which the marginal effect without accounting for heterogeneity across subgroups is of substantive interest. For example, if it is very early in the investigation of a putative cause or the marginal effect is very large, the overall effect estimate alone may be an important scientific contribution. On the other hand, evaluating effect heterogeneity is especially useful if the exposure in question is a treatment, is expensive, has limited availability, or may harm some while benefiting other individuals.