Additive excess risk model for epidemiologic interaction in retrospective studies.