We assessed interaction between the additive effects of family history and estrogen plus progesterone replacement therapy on the risk of breast cancer. We find no important interaction.
Previous studies have examined interaction as a departure from multiplicativity. Two studies provide data about stratum-specific absolute risks, from which we can assess departure from additivity. Olsson et al12
examined interaction among a cohort of 29,508 Swedish women aged 25–65 years. They defined estrogen replacement use as ever versus never and did not provide a definition of breast cancer family history. Reanalysis of their published data yields an interaction contrast of 0.005, similar to the small IC in our data.
Sellers et al18
conducted a follow-up study among 35,919 women aged 55–69 years. They defined family history as any breast cancer ever diagnosed in a mother, sister, or daughter, and estrogen replacement use as ever versus never. From these data, we estimate an interaction contrast of 0.007, which is also similar to the clinically insignificant IC in our study.
The other published works on this topic were based on case-control study designs,7-11,15-17
for which stratum-specific absolute risks are not observed. Nevertheless, we can evaluate departures from additivity by assessing the relative excess risk for interaction (RERI).23
Four studies provided enough data for us to estimate the RERI. For each study, we classified estrogen replacement use as ever versus never, and family history as present or absent, based upon breast cancer in a mother or sister. The resulting RERI values for these 4 studies are 0.58 (Pesch et al15
), 0.60 (Magnusson et al16
), 0.20 (Newcomb et al10
), and 0.94 (Nomura et al11
). There are no established standards for a clinically meaningful RERI. We propose that a RERI that exceeds half the risk in the doubly unexposed (ie, RERI >0.50) implies a potentially important degree of interaction. Although the data of Magnusson et al showed a RERI above this threshold, their data demonstrated no departure from multiplicative effects and thus Magnusson et al concluded that no interaction existed.
The RERI for our study (0.12) is lower than that observed by others. It is possible that the categorization of estrogen replacement therapy as ever versus never (as in all of the studies to which we compare our results) influences the estimates of interaction. The Pesch et al15
data allow for categorization by current versus not current, which comes closer to the definitions in the present study. Such reclassification strongly reduces the RERI estimate (0.02), thus supporting the potential for estrogen exposure categorization to explain the higher estimates among earlier studies. A second potential explanation is that our estimates of the other studies’ IC or RERI are based on unadjusted data extracted from the published results. None of these studies randomized estrogen replacement as in the present study. Thus, confounding remains a potentially important factor in explaining the discrepancy between the present findings and previously reported results. For these reasons, we believe that the evidence from the extant literature does not substantially challenge our conclusion of an absence of interaction between estrogen plus progesterone replacement therapy and family history on the incidence of breast cancer.
Our study has limitations. First, this study had both a short period of exposure to hormone treatment and a short follow-up period. It is possible that longer exposure and longer follow-up would demonstrate a different association between postmenopausal hormonal therapy and breast cancer, and be characterized by more interaction between this therapy and family history. The extant literature is inconsistent regarding a dose-dependent relation between duration of estrogen use and breast cancer risk.6-10,12,24-26
Second, family history is measured only at baseline in the Women’s Health Initiative. It is likely that some women categorized as not having a family history became aware of such a history during the follow-up period. If the occurrence of a new family history influenced the woman’s adherence to the trial medication or led to more frequent self-examinations for breast lumps, then this misclassification could be differential and dependent. This misclassification would bias the hormonal therapy risk difference estimate among the family history negative group in an unpredictable direction, thus causing unpredictable bias of the IC. Nevertheless, we expect that the opportunity to influence the diagnosis of breast cancer either by study adherence or by detection behavior is low, given the small group of women likely to be misclassified with regard to family history and the short interval of follow-up. Therefore, we expect that any bias due to misclassification of family history is unlikely to have substantial impact on the observed interaction contrast. Finally, the Women’s Health Initiative trial population represents a more educated and a somewhat healthier population with fewer racial minorities than the US population as a whole. While this is likely to influence the absolute rates of breast cancer, it is unknown how these differences might influence the interaction between family history and estrogen plus progesterone replacement therapy exposure on breast cancer. We posit that the biology of unrepresented groups is likely to be sufficiently similar as to yield similar conclusions. However, this remains to be examined.
Our study has unique and important strengths. First, this study takes advantage of randomized hormonal treatment. Prior work demonstrates that family history influences postmenopausal women’s choices to use estrogen,19
creating an important degree of confounding when examining the potential interaction between these 2 clinically-important breast cancer risk factors. Thus, randomized hormonal therapy allows a better estimate of the individual—and combined—effects of hormones and family history on breast cancer risk. Second, our approach to interaction on the additive scale provides clinically useful information for postmenopausal women and their physicians who are considering the use of postmenopausal hormonal therapy. Since family history is immutable and taking hormones is a personal choice having potential clinical benefits, the absolute risks associated with choosing hormone therapy can help women decide whether the incremental breast cancer risk is acceptable.