ABSTRACT
According to one textbook, the de‰nition of interaction in epidemiology is ‘to describe a situation in which two or more risk factors modify the effect of each other with regard to the occurrence or level of a given outcome’ (Szklo and Nieto 2004). This phenomenon is also known as effect modi„cation (Szklo and Nieto 2004), as the effect of one risk factor on the health outcome is different across the levels of another risk factor, where the latter is often called an effect modi‰er of the former. To test statistically whether or not there is interaction between two risk factors, the most frequently used method is to incorporate a product term of the two risk factors; that is, multiplying the exposure variable by the effect modi‰er variable, in a regression model (Selvin 1994; Kirkwood and Sterne 2003; Szklo and Nieto 2004). If the regression coef‰cient for the product term is signi‰cantly different from zero, this indicates that the effect of one exposure is dependent upon (i.e., modi‰ed by) the levels of the effect modi‰er. Another approach is to stratify the effect modi‰er and to compare the effect sizes of the exposure variables across a different stratum of the effect modi‰er (Rothman 2002). These two approaches are equivalent mathematically if the effect modi‰er is a categorical variable. When the effect modi‰er is a continuous variable, a common practice is to categorise it into two or more groups and then to test whether or not the effect sizes are different across these groups.
