We observed that adherence to recommended breast healthy behaviors (physical activity, alcohol moderation, and body weight maintenance) did not modify the breast cancer risk attributed to a family history of late-onset breast cancer among postmenopausal women. Therefore, women who participate in breast healthy behaviors appear to derive essentially the same benefits regardless of a family history of late-onset breast cancer. Neither confounding nor exposure misclassification apparently explains our null findings. To our knowledge, no earlier research has evaluated whether adherence to a cluster of behaviors modifies the relation between family history and breast cancer.
Some earlier research has addressed the interaction between family history and individual behaviors in their association with breast cancer risk [6
], although interaction is uniformly assessed on the multiplicative scale in these cohort studies. However, three such studies present enough data from which we can make some comparisons with our study, which examines additive interaction.
Sellers and colleagues [17
] conducted a follow-up study among 37,105 women (age range = 55 to 69 years) who were enrolled in the Iowa Women's Health Study Cohort. The authors assessed the interaction between family history and BMI. Their data yield a negligible IC of -0.0006 with a 95% CI of (-0.0017, 0.0030). Tehard and colleagues [18
] evaluated the interaction between family history and physical activity among 90,509 French women (age range = 40 and 65 years) who participated in the French E3N Study cohort. The findings of these authors, as well, demonstrate a negligible departure from additive effects (IC = -0.0009, 95% CI = -0.0022, 0.0004).
Using a case-control study design among postmenopausal women from California, Carpenter and colleagues [6
] evaluated the interaction between BMI and family history on the occurrence of breast cancer. Although the authors assessed interaction only on the multiplicative scale, they provide enough data to estimate interaction on the additive scale by calculating the relative excess risk due to interaction (RERI) [19
]. BMI was categorized as not more than 27 kg/m2
(normal) versus greater than 27 kg/m2
(overweight), and family history was categorized on the basis of the presence or absence of breast cancer in either a mother or a sister. Compared with the RR estimates of the doubly unexposed group (that is, no family history and normal BMI), those of the doubly exposed group, the family history exposed group, and the overweight group are 2.52, 1.54, and 1.27, respectively. This yields an RERI of 0.71, which is greater than the negligible RERI obtained from the cohort of Sellers and colleagues (RERI = -0.15) and the estimate from our study (RERI = 0.02). However, direct comparison with our findings is limited because our exposure classifications (cluster of breast healthy behaviors and FHLBC) differ substantively from theirs.
Our study has important limitations. Although preventive behaviors are dynamic, this analysis is based on assessment at one time point. It is possible that longitudinal assessment of breast healthy behavior would identify more clearly the distinction between women engaging in sustained participation versus sustained non-participation and, thus, present a stronger contrast. However, lifestyle behaviors are habitual and become relatively stable in adulthood as compared with childhood [20
]. Our measure of body weight management takes into account the women's entire adulthood, not just the time at enrollment. We believe, therefore, that it is unlikely that changes in behavior during the relatively short period of follow-up would lead to important differences in the interaction with FHLBC on breast cancer risk.
Similarly, family history is measured only at baseline in the WHI. It is likely that some of our unexposed respondents became aware of a family history during the follow-up period and therefore were misclassified as unexposed. We also expect, however, that the opportunity to influence breast cancer detection is low given the small group of women likely to be misclassified, the short interval during which diagnostic bias might occur, and the thorough diagnostic follow-up on all trial participants. Furthermore, this bias, if present, should occur in both the breast healthy and non-breast healthy strata, thus contributing the same absolute effects on the rate difference estimates and leaving the IC unaffected. We expect, therefore, that any bias due to misclassification of family history is very unlikely to have a substantial impact on our observed IC and null conclusion.
Lastly, FHLBC is only a proxy measure for a complex set of breast cancer-relevant exposures shared among families. The exact distribution of shared behaviors, environments, and genes leading to the presence of an FHLBC is likely to differ substantially from one participant to another in this study. Since habitual health behaviors are often shared within families [21
], it is likely that mothers and sisters of women engaging breast healthy patterns of behavior were more likely to exhibit similar types of behavior as compared with mothers and sisters of women who did not engage breast healthy behavior. This pattern would lead to the possibility that breast cancers arising among relatives of breast healthy participants occurred despite
healthy behavior, thus suggesting greater contribution of genetic and environmental factors among this group. One way to address this concern would be to assess whether a woman's risk attributable to having an FHLBC is influenced by a discordance between her behavior and that of affected relatives. However, the WHI-OS data do not include information about the behavioral norms for women's families or affected relatives and so we could not perform such analyses.