The EMBRACE cohort is one of the largest prospective studies reporting cancer risks in proven BRCA1 and BRCA2 mutation carriers. Follow-up rates were high (>75% were followed up by questionnaire, and all participants were flagged for notification of death or cancer through the Office for National Statistics). We included individuals diagnosed with breast cancer in ovarian cancer analyses because a diagnosis of breast cancer was not associated with risk of ovarian cancer in Cox regression (P = .30 for BRCA1, and P = .60 for BRCA2 carriers). Survival from breast cancer could, however, potentially affect incidence of ovarian cancer. The study population is enriched for families that meet high or moderate risk screening criteria presenting to genetic clinics; therefore the estimates will be most relevant to similar families. Because of this selection bias, the risk estimates are likely to be higher than would be obtained in a population-based study, albeit such a study would be infeasible because of the low prevalence of mutations in the general population.
The average cumulative risks of breast cancer by age 70 years were estimated for BRCA1 and BRCA2 carriers. Risks in BRCA1 carriers were similar to those derived from retrospective models based on complex segregation analysis but slightly higher in BRCA2 carriers. The latter observation is consistent with the aggregation of genetic modifiers in families because carriers in EMBRACE were identified through clinical genetic testing of individuals with stronger family history. The average cumulative risks of ovarian cancer by age 70 years for BRCA1 and BRCA2 carriers were also somewhat higher than estimated through segregation analysis, particularly for BRCA1 carriers. Because model-based estimates of ovarian cancer penetrance were derived from population-based studies and apply to women with weaker family history than those recruited in EMBRACE, these results are again consistent with the influence of genetic modifiers or other factors that cluster in families and modify cancer risks for mutation carriers.
The analyses for breast cancer were censored at ovarian cancer. As an alternative, we also performed competing risk analyses for breast and ovarian cancer in which the cumulative probabilities of each cancer were estimated simultaneously (Supplementary Table 5, available online). These estimates were somewhat lower. For example, cumulative incidence of breast cancer by age 70 years was 55% (95% CI = 34% to 72%) for BRCA1 and 52% (95% CI = 34% to 67%) for BRCA2 carriers.
A few prospective studies have reported cancer incidence in unaffected mutation carriers. Kramer et al. reported breast cancer risk by age 70 years of 76% among 98 BRCA1 carriers from multiple-case families. Moller et al. published a larger study, but breast and ovarian cancer incidences were not reported separately. Recently Metcalf et al. published risks of breast and ovarian cancer similar to ours, in a large series of mutation carriers. This study also confirmed the influence of family history on disease risks.
We also estimated the average cumulative risks of CBC for BRCA1 and BRCA2 carriers, respectively. These results cannot be directly compared with previous studies. However, in a retrospective analysis in our dataset, 10-year risks of CBC after a first breast cancer were 33.5% for BRCA1 carriers and 19.5% for BRCA2 carriers (Supplementary Table 6, available online). Metcalf et al. reported a combined 10-year actuarial risk of 29.5% (95% CI = 20.6% to 38.3%) using pedigrees segregating BRCA1 and BRCA2 mutations that were "retrospectively" ascertained. In a subsequent cohort study, these investigators reported risks of 24% for BRCA1 and 19% for BRCA2 carriers. Pierce et al. reported a 10-year risk of 26.0% (95% CI = 22.0% to 30.0%) among 71 BRCA1/2 carriers. Graeser et al. reported a lower risk (16.6%, 95% CI = 13.3% to 19.9%). This study differed from ours in several respects: index patients were excluded from analyses; only 17% of relatives were proven mutation carriers; and ascertainment of cancer occurrence was incomplete. Malone et al. reported lower CBC risk in a nested case–control study, which may reflect use of a population-based design. An increased CBC risk has been associated with decreasing age at diagnosis of the first cancer and with family history of breast cancer. Our results confirm high risks of CBC for both BRCA1 and BRCA2 mutation carriers. For BRCA1 carriers, the risks were higher than the corresponding risks for the first cancer. This higher risk presumably reflects risk modification by other genetic factors or other risk factors enriched in women with breast cancer.
Cancer rates may have been slightly underestimated if there were underreporting of prophylactic surgeries. Assuming similar rates of prophylactic surgery among women who did not respond to follow-up questionnaires as those responding, breast cancer incidence may have been underestimated by approximately 7%, CBC by approximately 10%, and ovarian cancer by approximately 20%. This would correspond to a cumulative breast cancer risk by age 70 years in BRCA1 carriers, for example, of approximately 63% rather than 60% and an ovarian cancer risk of approximately 67% rather than 60%. There was a suggestion of a cohort effect in cancer risks in our study. Both breast and CBC incidence appeared to be increased in birth cohorts after 1950 compared with those before 1950 (data not shown), as has been observed previously. On the other hand, the incidence of ovarian cancer appeared to be reduced in later cohorts. This could be the result of oral contraceptive use which became widespread in the United Kingdom in the 1970s. The number of individuals enrolled from earlier birth cohorts was, however, insufficient to definitively establish these effects.
We also investigated the effect of oophorectomy on cancer risks. There is considerable evidence that prophylactic oophorectomy reduces cancer risks in mutation carriers. One meta-analysis reported 50% reduction in breast cancer risk associated with oophorectomy. However, as the authors of this meta-analysis pointed out, studies varied widely with respect to methodology and inclusion criteria. For example, some studies examined only unaffected women, whereas others included women with previous breast cancer. Fewer studies have reported gene-specific effects. In this study, we stratified analyses by genotype and by first breast cancer or CBC. We observed a trend toward reduction in breast cancer risk for both BRCA1 and BRCA2 carriers; in BRCA1 carriers, breast cancer risk was halved. Although not statistically significant, the effect size is consistent with previous estimates. As has been observed previously, oophorectomy carried out at younger ages had greater impact on breast cancer risk. There was a suggestion that oophorectomy reduces risk of CBC for BRCA1 carriers but has a larger and statistically significant effect on risk for BRCA2 carriers. Kauff et al. reported a similar risk reduction in BRCA2 carriers. In their study, women with and without a history of previous breast cancer were included in analyses and hazard ratios were adjusted for differences in history of breast cancer between the oophorectomy and surveillance groups.
In a collaborative study, Domchek et al. reported a reduction in breast cancer risk associated with oophorectomy in unaffected BRCA1 carriers (HR = 0.63, 95% CI = 0.41 to 0.96) and for BRCA2 carriers (HR = 0.36, 95% CI = 0.16 to 0.82) but did not observe any effect on CBC risk. There is some overlap between centers included in the PROSE collaboration reported by Domchek et al. and EMBRACE. After excluding potential overlapping centers, however, our results were essentially unchanged. Although the estimated relative risks in the two studies are consistent, it is important to note that the analytical approaches were different. Domchek et al. considered only ooophorectomy occurring after ascenrtainment and used women not having oophorectomy as a historical control group, whereas we considered oophorectomy both before and after recruitment and analyzed oophorectomy as a time-dependent covariable.
This study also had some limitations. Whether our results reflect true differences in the effect of oophorectomy in BRCA1 and BRCA2 carriers, differences in the timing of oophorectomy and follow-up in different subgroups, or random variation due to small numbers remains to be tested in larger cohorts. The results may have also been confounded if, for example, women with family history of ovarian cancer were more likely to undergo oophorectomy and a family history was associated with breast cancer risk or if factors such as parity, oral contraceptive, or hormone receptor therapy use, which may be related to both oophorectomy uptake and cancer risk, were inadequately adjusted for. In addition therapies associated with the first breast cancer may be responsible for risk reduction, rather than oophorectomy per se. A potential shortcoming of this study is lack of data on tamoxifen, other therapies, and surgical procedures carried out for unilateral breast cancer. In addition, there may have been some underreporting of prophylactic oophorectomy in women without cancer, resulting in underestimation of the effect of oophorectomy on cancer risks.
We further investigated the role of common breast cancer susceptibility alleles and their associations with breast cancer risk in this cohort. A number of genetic modifiers of BRCA1 and BRCA2 have been identified. The relative effect of each individual locus is small (per-allele HR < 1.3). However, because the absolute risk of breast cancer conferred by mutations in BRCA1 and BRCA2 is already high, the effects of genetic modifiers on the absolute risk of disease are much greater than in the general population. In this study, we constructed a risk score based on the joint distribution of the associated loci and tested the effect on breast cancer risk of tertiles of the risk score in our cohort of unaffected mutation carriers. The variants were assumed to act multiplicatively on risk. For the risk score based on the combination of seven BRCA2-associated variants, the third of BRCA2 carriers with the highest risk score are at more than threefold increased risk of breast cancer compared with the third of carriers at lowest risk. The association between the risk score and breast cancer risk in BRCA1 carriers was in the expected direction but was not statistically significant. However, only four risk alleles were tested for BRCA1. To our knowledge, this is the first study to evaluate the effects of SNPs on cancer risk in carriers prospectively. These results confirm findings based on retrospective analysis from the CIMBA consortium and suggest that genetic profiles may be useful for improving risk prediction in mutation carriers, but the confidence intervals surrounding the estimates are wide, and larger studies are needed to provide more accurate prospective estimates.
The results from our prospective study provide absolute estimates of cancer risk in carriers and of the modifying effects of genetic polymorphisms and oophorectomy. Clearly, larger prospective studies with longer follow-up are required to provide definitive estimates—collaborations such as the International BRCA1/2 Carrier Cohort Study (IBCCS) will provide a mechanism to generate such estimates. Incorporating these factors into risk prediction models should improve the accuracy of these models and guide clinical management of BRCA1 and BRCA2 carriers.