A meta-analysis by Rebbeck and colleagues (3), published in 2009, evaluated four case–control or cohort studies with nonoverlapping participants that addressed this question (31, 32, 34, 35). These studies included 3,066 BRCA1, 1,116 BRCA2, and a further 1,669 mutation carriers where the specific gene not stated. The results suggested a statistically significant protective association between rrBSO and breast cancer risk for BRCA1 (HR, 0.47; 95% CI, 0.35–0.64) and BRCA2 mutation carriers (HR, 0.47; 95% CI, 0.26–0.84), and when the specific gene mutated was not stated (HR, 0.49; 95% CI, 0.37–0.65; ref. 3). These findings were supported by several subsequent studies with similar results (36, 37); however, analyzed datasets were overlapping. Not surprisingly, these results impacted clinical practice.

It has recently become clear that the findings of these older studies may be spurious due to the presence of several biases (4, 43). Heemskerk-Gerritsen and colleagues assessed the association between rrBSO and breast cancer risk by analyzing new data from the Hereditary Breast and Ovarian Cancer in the Netherlands (HEBON) nationwide cohort (4). They first replicated the eligibility criteria and analyses of the four major historical studies [30–32, 36; two of which (31, 32) were included in the meta-analysis by Rebbeck and colleagues (3)]. The results were similar to the findings of the original studies with hazard (HR) or odds ratios varying from 0.36 to 0.62, lending support to the intervention. To demonstrate the impact of bias, they reanalyzed the HEBON data using a statistical design that minimized several biases. To reduce cancer-induced testing bias, they started the observation period at receipt of genetic test result or age 30, whichever came last and excluded women diagnosed with breast cancer before the start of observation. To reduce immortal person-time bias, they treated rrBSO as a time-dependent variable, allocating all person-years of observation before rrBSO, as well as the three months following rrBSO, to the non-rrBSO group. Utilizing data from 589 BRCA1 and 233 BRCA2 mutation carriers, with 75 and 14 incident breast cancers respectively, and a median follow-up of 3.2 years, they found no statistically significant association between rrBSO and breast cancer risk for mutation carriers combined (HR, 1.09; 95% CI, 0.67–1.77). The estimates for BRCA1 and BRCA2 mutation carriers analyzed separately were HR, 1.21 (95% CI, 0.72–2.06) and 0.54 (95% CI, 0.17–1.66), respectively. There was also no statistically significant association between premenopausal rrBSO (i.e., before age 51) and breast cancer risk for mutation carriers combined (HR, 1.11; 95% CI, 0.65–1.90). The median age at rrBSO was 45 years (range, 31–67 years). The use of hormone replacement therapy (HRT) was not reported, and data related to other breast cancer risk factors, including parity, were missing (41%), which may have introduced confounding by other risk factors. In addition, confounding by indication, survival bias from competing risk of tubo-ovarian cancer and informative censoring may have been present. Regardless, it was the publication of this pivotal study in 2015 (4) that initiated the ongoing debate and controversy regarding whether rrBSO reduces breast cancer risk for BRCA1 and BRCA2 mutation carriers.

Since Heemskerk-Gerritsen's analysis (4, 43), six further, larger cohort studies have been published that address this question (see Table 3; refs. 6–11). These studies attempted to minimize bias; however, most have potential residual methodologic issues (see Supplementary Table S1). Taken together, these studies do not help reach consensus on whether rrBSO is associated with reduced breast cancer risk.

Characteristics of contemporary studies of rrBSO and breast cancer risk in BRCA1 and BRCA2 mutation carriers.

Kotsopoulos and colleagues published a prospective cohort study of 2,969 BRCA1, 725 BRCA2, and 28 BRCA1 or BRCA2 (specific gene unknown) mutation carriers with no prior breast cancer diagnosis, to evaluate the effect of rrBSO on breast cancer risk (5). Of the 3,722 women studied, 857 underwent rrBSO before cohort enrolment and 695 underwent rrBSO after enrolment. The observation period commenced either at completion of the baseline questionnaire or receipt of genetic testing result, whichever was later, to limit cancer-induced testing bias. The mean age at rrBSO was 46.3 years (range, 13–78). With 350 incident breast cancers observed during a mean follow-up period of 5.6 years, there was no statistically significant association between rrBSO and breast cancer risk for BRCA1 (HR, 0.97; 95% CI, 0.73–1.29; P = 0.85) or BRCA2 mutation carriers (HR, 0.68; 95% CI, 0.38–1.21; P = 0.19). rrBSO was also not statistically significantly associated with reduced risk for breast cancer diagnosed under age 50 years for BRCA1 mutation carriers (HR, 0.84; 95% CI, 0.58–1.21; P = 0.34); however, rrBSO was associated with an 83% lower risk of breast cancer diagnosed under age 50 years for BRCA2 mutation carriers (95% CI, 0.05–0.61; P = 0.006). It is unclear why this analysis was limited to breast cancers diagnosed before age 50, rather than examining the effect of rrBSO before age 50 on risk of breast cancer over the entire follow-up period. In light of Heemskerk-Gerritsen and colleagues' publication, Kotsopoulos and colleagues treated rrBSO as a time-dependent variable, mitigating immortal person-time bias and adjusted for parity and other well-described risk factors in a multivariable analysis. They attempted to reduce potential confounding by indication, informative censoring, and survival bias by adjusting for cancer family history (i.e., number of first-degree relatives affected by breast cancer). However, this approach fails to consider more subtle components of family history that affect cancer risk, such as age at breast cancer diagnosis and affected status of more distant relatives (of particular importance where there is paternal inheritance) and therefore provides only partial mitigation.

Following on from Kotsopoulos and colleagues, Terry and colleagues analyzed data from 716 BRCA1 and 573 BRCA2 mutation carriers from the Prospective Family Study Cohort, encompassing the Breast Cancer Family Registry (BCFR) and the Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer (kConFab; ref. 6). In their sample, the median age of rrBSO was 44 years for BRCA1 and 46 years for BRCA2 mutation carriers. Incident breast cancer was diagnosed in 116 BRCA1 and 80 BRCA2 mutation carriers during a median follow-up of 10.7 years. To demonstrate the importance of treating rrBSO as a time-dependent variable, Terry and colleagues first treated it as a fixed exposure and observed a statistically significant association between rrBSO and reduced breast cancer risk similar to Rebbeck and colleagues' meta-analysis (BRCA1: HR, 0.40; 95% CI, 0.26–0.67; BRCA2: HR, 0.32; 95% CI, 0.17–0.60). However, there was no statistically significant association when rrBSO was treated as a time-dependent variable (BRCA1: HR, 1.20; 95% CI, 0.67–2.12; BRCA2: HR, 0.86; 95% CI, 0.43–1.72). This supported the conclusions of Heemskerk-Gerritsen and colleagues and emphasizes the potential effect of this bias on these observational studies. The study design reduced cancer-induced testing bias by only including women who were unaffected at the start of the observation period; however, the observation period did not start at the time of genetic testing, so residual cancer-induced testing bias may have been present. Other potential sources of bias may also have been present including confounding by indication, survival bias from competing risk of tubo-ovarian cancer, informative censoring and confounding by other risk factors. Furthermore, Terry and colleagues did not report on the effect for mutation carriers undergoing pre-menopausal rrBSO, although an analysis of women in the upper tertile of breast cancer risk (inclusive of mutation carriers and other high-risk women), showed no difference in risk based on age at rrBSO (<45, 45–49, ≥50 years).

More recently, Choi and colleagues (7) reported findings from further analyses of BCFR data, including 746 women with BRCA1 and 576 with BRCA2 mutations, of which 483 and 373 had breast cancer, respectively. This newer study had the same methodologic issues as Terry and colleagues (6). Some of these potential biases were likely exacerbated by the apparent inclusion of prevalent breast cancer cases at cohort recruitment, and of additional retrospective data back to age 16 years (47). The estimated HRs for BRCA1 and BRCA2 mutation carriers overall were 0.57 (95% CI, 0.38–0.84) and 0.62 (95% CI, 0.41–0.96), respectively, and 0.28 (95% CI, 0.10–0.63) and 0.19 (95% CI, 0.06–0.71), respectively, in the first 5 years following rrBSO. No HRs were estimated for the relationship between premenopausal rrBSO and breast cancer risk (7).

Mai and colleagues have also recently addressed this question using data from the US Gynecologic Oncology Group-0199, a multi-institution, prospective cohort study of women at high risk of tubo-ovarian cancer (8). Considering only women in this study without a personal history of breast cancer (minimizing cancer-induced testing bias), there were 242 BRCA1 and 189 BRCA2 mutation carriers included in the analysis between rrBSO and breast cancer risk, of whom 120 BRCA1 and 102 BRCA2 mutation carriers had rrBSO. rrBSO was treated as a time-dependent variable, mitigating immortal person-time bias. Thirty-eight incident breast cancers were observed during follow-up: 29 in BRCA1 and 9 in BRCA2 mutation carriers. There was no statistically significant protective association between and rrBSO and breast cancer for BRCA1 or BRCA2 mutation carriers combined or separately (HR, 1.15; 95% CI, 0.52–2.54; P = 0.72; BRCA1 HR, 1.22; 95% CI, 0.50–3.00; P = 0.66; and BRCA2 HR, 1.09; 95% CI, 0.20–6.06; P = 0.92, respectively). This held true when the analysis was limited to premenopausal rrBSO (combined HR: 0.84; 95% CI, 0.40–1.77; P = 0.64; BRCA1: HR, 0.84; 95% CI, 0.37–1.91; P = 0.68; BRCA2: HR, 0.73; 95% CI, 0.11–4.82; P = 0.75); however, that analysis also included women with a personal breast cancer history, which, if anything, would lead to an overestimate of any association, through cancer-induced testing bias. Despite being a prospective cohort study specifically designed to address this question, the study had a small number of incident cancers and remained subject to several important biases. The authors recognized potential confounding by indication, especially as women in the rrBSO group were less likely to have a first- or second-degree relative diagnosed with premenopausal breast cancer (P = 0.03). Like other contemporary studies, survival bias from competing risk of tubo-ovarian cancer, informative censoring and confounding by other risk factors may have been present.

Mavaddat and colleagues recently published the largest study addressing this issue, using international, multi-center prospective pooled cohort data (9). It included 2,272 BRCA1 and 1,605 BRCA2 mutation carriers from three large consortia – the International BRCA1/2 Carrier Cohort Study (IBCCS), the kConFab Follow-up Study and the BCFR. Notably, the IBCCS cohort overlaps with that analyzed by Heemskerk-Gerritsen and colleagues (4) and the kConFab and BCFR cohorts overlap with those in Terry and colleagues (6) and Choi and colleagues (7). Cancer-induced testing bias was minimized by excluding women affected with breast cancer at the start of the observation period and by commencing observation after mutation testing (in 97% of enrolled women). rrBSO was treated as a time-dependent variable, with the addition of a latency period immediately after rrBSO (and at commencement of observation). During 5.4 and 4.9 years of follow-up respectively, a total of 269 and 157 incident breast cancer cases were diagnosed in BRCA1 and BRCA2 mutation carriers, respectively. In the primary analysis, there was no statistically significant association between rrBSO and breast cancer risk in BRCA1 (HR, 1.23; 95% CI, 0.94–1.61) or BRCA2 (HR, 0.88; 95% CI, 0.62–1.24) mutation carriers. For women with BRCA2 mutations, the HR for those who underwent rrBSO prior to the age of 45 was 0.68 (95% CI, 0.40–1.15), whereas that for rrBSO after age 45 was 1.07 (95% CI, 0.69–1.64). There was some evidence of a stronger association with increasing time since rrBSO for BRCA2 mutation carriers (Ptrend = 0.011), with a HR, 0.51 five years after rrBSO (95% CI, 0.26–0.99; P = 0.046) overall, and HR, 0.39 (95% CI, 0.16–0.97) in women undergoing rrBSO ≤45 years. These findings should be interpreted with caution as there was substantial variation in this HR between individual cohort studies included in the analysis (Pheterogeneity = 0.005; ref. 9). Like the others, this study is subject to possible bias from informative censoring. Women undergoing rrBSO were more likely to have a family history of tubo-ovarian cancer (P < 0.001), suggesting potential confounding by indication, although no statistically significant difference was observed in their family history of breast cancer among first- and second-degree relatives and a statistical adjustment was made to account for this. The authors also adjusted for parity, age at first birth and HRT, limiting confounding by other risk factors.

Stjepanovic and colleagues (10) conducted an analysis of data from five prospectively maintained registries in Spain and the United States, including 444 BRCA1 and 409 BRCA2 mutation carriers aged ≤51 years, 337 of whom underwent rrBSO before age 51. During the median 4.3 years of follow-up, 96 women developed incident breast cancer (54 with BRCA1 mutations and 42 with BRCA2 mutations). The median age of premenopausal rrBSO was 42 years (range, 30.5–50.9) in BRCA1 and 43.5 years (range, 33.7–50.9) in BRCA2 mutation carriers. In contrast to some of the other recent studies, a statistically significant protective association between rrBSO and breast cancer risk was reported for BRCA1 mutation carriers (HR, 0.45; 95% CI, 0.22–0.92; P = 0.03), but there was no statistically significant association for BRCA2 mutation carriers (HR, 0.77; 95% CI, 0.35–1.67; P = 0.51). They concluded that this evidence was sufficient to continue to recommend premenopausal rrBSO for BRCA1 mutation carriers (10). Stjepanovic and colleagues reduced several biases, including immortal person-time bias by treating rrBSO as a time-dependent variable and adding a 3-month latency after rrBSO. Cancer-induced testing bias was removed by commencing the observation time at receipt of mutation results, or at age 30, whichever occurred later and excluding women with a prior cancer diagnosis (10). However, the authors were unable to control for differences in family history or other potential confounding and therefore, the threat of confounding by indication and other risk factors persists. A sensitivity analysis that excluded women undergoing rrBM yielded similar results to the primary analysis, although this does not completely exclude the possibility of informative censoring (4). The authors went on to conduct a meta-analysis of findings from theirs and four published studies (4–6, 41) to determine the association between premenopausal rrBSO and breast cancer risk. Utilizing the two studies that distinguished rrBSO before or after age 50 (5, 41) alongside their own data in a subsequent analysis, Stjepanovic and colleagues observed a HR, 0.61 (95% CI, 0.36–1.02) for BRCA1 and HR, 0.43 (95% CI, 0.18–1.01) for BRCA2 mutation carriers (10).

rrBM is the most effective breast cancer risk reduction intervention for mutations carriers. rrBM was associated with an 87% and 82% reduction in breast cancer risk for BRCA1 and BRCA2 mutation carriers, respectively, in a meta-analysis of four studies (36, 37, 48–50). Similar to studies of rrBSO, all rrBM studies were observational and subject to bias; however, there is clear biologic plausibility that rrBM may reduce breast cancer risk. Discussion of rrBM is recommended by NCCN, NICE, ESMO, ACOG, and Australian eviQ guidelines; however, uptake is variable (51–53), so alternatives are desirable.

The use of chemoprevention in women with BRCA1 and BRCA2 mutations is also endorsed in guidelines. Unlike rrBM or rrBSO, it has the advantage of being a reversible intervention if women experience side-effects or change their mind.

Despite high-quality data supporting the efficacy of chemoprevention for noncarriers at high risk of breast cancer (54–57), and evidence that the risk reduction persists for many years after cessation of the medication (55–56), data pertaining to mutation carriers are extremely limited. The only randomized trial of aromatase inhibitors for primary breast cancer prevention in carriers was underpowered but reported no protective association between letrozole and breast cancer in postmenopausal women (HR, 1.29; 95% CI, 0.4–3.9; ref. 58). The NSABP-P1 study of tamoxifen for breast cancer prevention estimated a risk ratio for breast cancer of 1.67 (95% CI, 0.32–10.7) for BRCA1 and 0.38 (95% CI, 0.06–1.56) for BRCA2 mutation carriers randomized to tamoxifen (59). This study had limited power due to only 8 BRCA1 and 11 BRCA2 mutation carriers identified among 288 incident breast cancer cases. Given that the point estimate for BRCA2 was considerably less than 1 however, these findings are often interpreted to indicate that tamoxifen may be efficacious for risk reduction in this population, although there remains considerable uncertainty. Enrolment of women onto randomized clinical trials of new potential chemopreventive agents is encouraged (https://www.breastcancertrials.org.au/current-clinical-trials/brca-p).

A detailed discussion of lifestyle factors is beyond the scope of this article, but population recommendations for healthy living, such as maintaining a healthy weight, participating in regular moderate-intensity exercise, minimizing alcohol intake and minimizing exposure to combined exogenous estrogen and progesterone, should be applied (14).