Clinical relevance guided the choice of covariates (Hernan et al. 2002). The multivariate models were adjusted for time varying confounders using simple updating information at each 2-year and 4-year questionnaire cycle, including high alcohol consumption (≥30 g/day, <30 g/day), caffeinated coffee consumption (continuous, cup/d), body-mass index (<25.0, 25.0 to 29.9, ≥30.0 kg/m2), physical activity (quintiles), marital status (married/partnered or widowed/separated/divorced/single), and reported regular use of antidepressants (yes or no) and minor tranquilizers such as benzodiazepines (yes or no). In NHS II, hormonal status (post-menopausal with or without hormonal therapy, pre-menopausal or never used hormonal therapy) was also included. Sensitivity analyses including factors that can mediate the effects of smoking, such as self-reported high blood pressure, myocardial infarction or angina, stroke, and cancer (all yes/no) were preformed. All analyses were performed with SAS software, version 9.2 (SAS Institute Inc., 2003). All P values reported are 2-sided.
