Discussion
Our results suggest that neither EP-HT nor E-HT increase the risk of young-onset breast cancer and that E-HT might be associated with a reduced risk. Although the low rate of P-HT usage resulted in imprecise estimates, we also found some evidence that P-HT use increased the risk of young-onset breast cancer.
Our finding that E-HT use is associated with a decreased risk of young-onset breast cancer is consistent with the WHI findings for postmenopausal breast cancer. Although Shantakumar et al. and Palmer et al. previously found evidence that E-HT use increased the risk of young-onset disease, the estimates from both studies were highly imprecise (3 and 9 E-HT users, respectively).
Our results are also consistent with the analysis by Nelson et al. of 40–49-year-old women, in which they reported that HT users who did not have a uterus (probable E-HT users) had a reduced risk of young-onset breast cancer compared with former or never users and that HT users who did have a uterus (probable EP-HT users) had risk similar to those of former or never users. However, both Chlebowski et al. and Shantakumar et al. reported elevated breast cancer risk with EP-HT use, and we had only limited power (approximately 43%) to detect a difference between the estimate of 1.26 reported in the WHI and our estimate of 0.8 (Web Appendix 2, Web Figure 4 http://aje.oxfordjournals.org/content/181/10/799/suppl/DC1). No other studies have examined the relationship between P-HT and young-onset breast cancer, but several studies targeting older women also found evidence of increased risk.
Although young women who take HT could be a medically distinct group, most of our HT users were in their 40s and thus likely to be perimenopausal. Nonetheless, use of hormones can serve as a surrogate for a range of treatment indications that are independently associated with a reduced risk of breast cancer, such as primary ovarian insufficiency, endogenous estrogen deficiency, gynecologic surgery, menopausal symptoms, or premature menopause. Although we adjusted for these factors and tested for interaction by oophorectomy status, some residual confounding by indication may persist.
HT users might also be more likely to utilize the medical system in general. Although we did not have good information on mammography use before the index age, we adjusted for health care utilization using a recent dental visit as an indicator. Any residual confounding due to surveillance bias would likely result in elevated estimates of breast cancer risk among HT users.
Menopausal status is a potential source of confounding in many existing studies of HT. For women at a given age, those who are premenopausal or perimenopausal are at greater risk of breast cancer, but they are less likely to be using HT than are postmenopausal women. In an attempt to minimize bias due to confounding by menopausal status, investigators might exclude premenopausal and perimenopausal women, women who have undergone a hysterectomy, and women who began HT before menopause. However, because many women initiate HT after a hysterectomy or at the onset of menopausal symptoms, we felt that an investigation of the risk implications of HT in young women was of public health importance. By including pre- and postmenopausal women, we were also able to study associations with the timing of HT initiation, which may be key to identifying critical susceptibility periods. An additional strength of the Two Sister Study is the sister-matched design, which presumably controls for unmeasured confounders that are similar across sisters.
Study limitations include our small numbers of exposed women, possible healthy-participant bias, possible surveillance bias, and possible recall bias. Exact participation rates for the Two Sister Study are hard to establish, as some contacted cases might have elected not to contact us because they realized they were ineligible. Only a small fraction of participants were explicitly excluded because of death or poor health, and more than half of the identified cases participated. Although we cannot exclude healthy-participant bias, the magnitude of such bias would be mitigated by the sister-matched design. As noted earlier, there might be some surveillance bias due to our inability to capture mammography usage around the index age. However, differential surveillance would presumably make HT users more likely to be diagnosed, with the implication that the use of HT could be even more protective than we estimated.
Recall bias may be present, because controls were more likely to have a large gap between their interview age and index age and those with longer recall periods were less likely to report HT use before their index age. We adjusted for recall time in both the multivariate and propensity score analyses, but residual confounding might exist. Nonetheless, because of their shorter recall times, we would expect cases to report HT use more often than controls, and therefore recall bias would not explain the evident protective association of E-HT use seen here. Our sensitivity analyses that excluded sister pairs with large age differences also served to reduce recall bias. The similarity of these results with the main findings suggests that recall bias was not influential.
The fact that control sisters tended to be slightly older than their case sisters is also a concern. Although this tendency helped ensure that controls were cancer-free at the age that their sister was diagnosed, it produced a potentially biasing birth cohort effect related to the sharply declining use of HT in the early 2000s. We controlled for this effect by adjusting for individualized propensity scores. Propensity scores are not typically applied to case-control studies, but we were able to model well-informed, stable propensity estimates by using a large subcohort of the Sister Study. This subcohort represents the source population for the Two Sister Study. We further equalized the opportunity for exposure by evaluating all covariates at an index age that occurred before either sister was diagnosed. The use of an index age both controlled for age and eliminated effects due to any behavioral changes brought about by having a sister diagnosed with breast cancer. We do acknowledge, however, that this novel application of propensity scores to a matched case-control study has not been validated.
To our knowledge, this is the largest study to examine the association between HT use and young-onset breast cancer and the first conducted in the post-WHI era. This research informs the trade-offs between risks and benefits of HT for young women, who might take it to help manage symptoms of early or surgically induced menopause. Our findings suggest that for women under 50 years of age, EP-HT use does not increase risk but that E-HT, indicated only for women who have undergone hysterectomy, might be associated with a reduced risk of young-onset breast cancer.