Letters Section Editor: Robert M.
Golub, MD, Senior Editor.
In Reply: We agree with Dr Pinsky that statistical interaction is scale dependent. While this is well recognized in the epidemiological literature, in genetic epidemiology there is a very strong default option in the design of the studies. Thus the typical study for binary outcomes (phenotypes) generally uses a case-control design, and main effects are presented in a multiplicative scale in the majority of investigations.1 Thus,
testing for multiplicative interaction makes sense. Similarly, for quantitative (continuous) traits most analyses use an additive scale.
Regardless, in our empirical evaluation, we tried to replicate the analyses using the same scale and model as had been used by the original investigators of each study. We agree that even when formally significant interaction was found, evidence of statistical interaction should not be considered identical to evidence of biological interaction.2
Patsopoulos NA, Tatsioni A, Ioannidis JPA. Claims of Sex-Gene Interactions—Reply. JAMA. 2007;298(23):2741-2742. doi:10.1001/jama.298.23.2742