Interaction: A word with two meanings creates confusion

Perhaps more than any other word in epidemiology,interaction presents a challenge to clinical and epi-demiological researchers. The problem stems from itsapplicability to describe two dierent phenomena. Onthe one hand, interaction refers to the biologicinteraction of two or more causes of disease thattogether assert their inuence on disease risk. On theother, interaction refers to statistical interactionwhich is the necessity for a product term in a linearmodel. In this editorial, we have two related goals: (1)To provide authors with a common terminology fordiscussing biological and statistical interaction; and(2) To emphasize statistical interaction in the additivemodel as the basis for assessing biological interaction.Most causes of disease are dependent on the pres-ence of other factors to assert their eects and this isbiological interaction. Examples of how risk factorsinteract are ubiquitous in medical and public healthsciences. One of the more striking examples isphenylketonuria (PKU), a metabolic disorder inwhich the combination of a genetic mutation and anenvironmental factor, in this case dietary exposure toa particular amino acid gives rise to mental retarda-tion in children. Because exposure to both factors isnecessary for PKU to occur, infants with the geneticdefect are put on a life-long restricted diet and areable to prevent the development of the disease.Another example is human papilloma virus (HPV)and cervical cancer. HPV appears to be a necessarycause, since its presence is observed in >99