Among cancers in American women, breast cancer (BC) has the second highest incidence and mortality. The association of BC with diet has been inconsistent. Studies that evaluate associations with dietary patterns are less common and reflect an individual’s whole diet. We associated dietary patterns with the risk of BC in American women of the Adventist Health Study-2 (AHS-2), a prospective cohort of 96 001 subjects recruited between 2002 and 2007. Answers to a previously validated FFQ were used to classify subjects to vegan, lacto-ovo-vegetarian, pesco-vegetarian, semi-vegetarian and non-vegetarian dietary patterns. Incident BC were identified by matching AHS-2 subjects to data from forty-eight state cancer registries. Statistical analyses used proportional hazard regression analyses with covariates that were chosen a priori. From 50 404 female participants (26 193 vegetarians), we identified 892 incident BC cases, with 478 cases among vegetarians. As compared with non-vegetarians, all vegetarians combined did not have a significantly lower risk (hazard ratio (HR) 0·97; CI 0·84, 1·11; P=0·64). However, vegans showed consistently lower (but non-significant) point estimates when compared with non-vegetarians (all cases: HR 0·78; CI 0·58, 1·05; P=0·09). In summary, participants in this cohort who follow a vegetarian dietary pattern did not experience a lower risk of BC as compared with non-vegetarians, although lower risk in vegans is possible. These findings add to the very limited literature associating vegetarian diets with BC risk and can assist nutritionists when evaluating the impact of these diets. The findings will also motivate further evaluation of vegan diets and their special characteristics.

Past dietary patterns may be more important than recent dietary patterns in the aetiology of chronic diseases because of the long latency in their development. We developed an instrument to recall vegetarian dietary patterns during the lifetime and examined its reliability of recall over 5·3 and 32·6 years on average. The short-term/5-year recall ability study (5-RAS) was done using 24 690 participants from the cohort of the Adventist Health Study-2 (mean age 62·2 years). The long-term/33-year recall ability study (33-RAS) included an overlap population of 1721 individuals who joined the Adventist Health Study-1 and Adventist Health Study-2 (mean age 72·5 years). Spearman correlation coefficients for recall of vegetarian status were 0·78 and 0·72 for the 5-RAS and 33-RAS, respectively, when compared with ‘reference’ data. For both time periods sensitivity and positive predictive values were highest for the lacto-ovo-vegetarian and non-vegetarian patterns (vegans, lacto-ovo-vegetarians, pesco-vegetarians, semi-vegetarians and non-vegetarians). In the 5-RAS analyses, male, non-black, younger, and more educated participants, lifetime Adventists, and those with more stability of consumption of animal products generally showed higher recall ability. Somewhat similar tendencies were shown for the 33-RAS analyses. Our findings show that the instrument has higher reliability for recalled lacto-ovo-vegetarian and non-vegetarian than for vegan, semi- and pesco-vegetarian dietary patterns in both short- and long-term recalls. This is in part because these last dietary patterns were greatly contaminated by recalls that correctly would have belonged in the adjoining category that consumed more animal products.