How to evaluate interaction between causes: a review of practices in cardiovascular epidemiology.

To increase the knowledge of interaction or synergy between risk factors in an important task in medical research. Still, current literature in cardiovascular epidemiology reflects major misconceptions as how to evaluate interaction. This paper presents Rothman's model of causation from which strict empirical criteria of interaction can be derived. In principle, the method to apply consists of comparing risk differences for one risk factor of interest across strata of the other. Commonly used but incorrect approaches are exemplified and discussed. These include reporting risk of disease among those with combined exposure, comparing relative risks for one exposure after stratification by level of the other, and including an interaction term in the regression model and drawing conclusions from its P-value.