Last update:


Arch Hellen Med, 21(2), March-April 2004, 137-147


Statistical interaction and effect measure modification

Laboratory of Clinical Epidemiology, School of Nursing, University of Athens, Athens, Greece

The term “interaction” is used in statistics, epidemiology and pharmacology. In statistics, “interaction” is used to refer to departure from the underlying form of a statistical model (additive or multiplicative). In epidemiology the expression “effect measure modification” is usually used instead of the term “statistical interaction”. Εffect measure modification is said to occur when the magnitude of the association between one variable and another differs according to the level of a third variable. For example, obesity increases the risk for breast cancer in postmenopausal but not premenopausal women. Thus menopausal status is a modifier of the effect of obesity on breast cancer. A modifier is a characteristic of the individual which modifies the effect of an exposure on the incidence of disease. In this way the modifier is the characteristic that changes the relation between exposure and disease. The presence or absence of effect measure modification depends on the arbitrary choice of an effect measure. The effect measures usually used are the risk ratio (or incidence density ratio) and the risk difference (or incidence density difference). When the use of one effect measure (for example risk ratio) does not indicate modification, then the use of another effect measure (for example risk difference) may demonstrate modification. Thus blanket statements about the presence or absence of effect measure modification cannot be made based on a single effect measure. Εffect measure modification should be distinguished from confounding which is defined as distortion in an effect measure introduced by an extraneous variable, occuring only in the context of a particular study. Confounding in a specific study is a systematic error (bias) which the investigator should locate and eliminate, in order for the results of the study to be valid. In contrast, effect measure modification is a finding to be reported rather than a bias to be avoided. Apart from above interaction with the meanings reported there may also be biological interaction, which refers to a mechanical interaction that either may or may not occur. Biological interaction can be described more accurately as biological co-action.

Key words: Confounder, Confounding, Effect measure modification, Modifier, Statistical interaction.

© Archives of Hellenic Medicine