The proportion testing positive for each gene with Manchester scores [14] above and below 20 and 23 are presented in Table ​Table2.2. This shows a substantial effect of cancer burden for BRCA2 with high-risk families (scores above 20 points) having a much lower proportion of positive predictive tests after 50 years.

Overall, of the FDRs of unknown mutation status, 92/92 with ovarian cancer, 220/244 with breast cancer and 234/648 unaffected FDRs were included in the analysis. In total this amounted to 839 actual and presumed carriers for BRCA1 and 603 actual and presumed carriers for BRCA2. There were 243/839 (29%) BRCA1 individuals with ovarian cancer compared to 64/603 (11%) female BRCA2 carriers. 411/839 (49%) BRCA1 carriers and 355/603 (59%) BRCA2 carriers had developed breast cancer. Penetrance estimates for each gene are shown (Table ​(Table3;3; Figures ​Figures1,1, ​,2)2) for breast and ovarian cancer. The curves were remarkably similar for each gene, with breast cancer penetrance to 70 and 80 years of 68% (95%CI 65–71%) and 79.5% (95%CI 75–84%) for BRCA1 and 74% (95%CI 71–77%) and 88% (95%CI 85–91%) for BRCA2. Ovarian cancer risk to 70 and 80 years was 60% (95%CI 65–71%) and 65% (95%CI 75–84%) for BRCA1 and 30% (95%CI 25.5–34.5%) and 37% (95%CI 31.5–42.5%) for BRCA2. The penetrance for ovarian cancer was significantly higher for BRCA1 (p < 0.0001), but breast cancer incidence for BRCA2 was borderline significantly higher than for BRCA1 (p = 0.09). Indeed breast cancer penetrance estimates for BRCA2 after 60 years were significantly higher as was overall penetrance including the index case (p = 0.02). There was no significant effect of ovarian cancer cluster regions (OCCR, nucleotides 2401–4190 BRCA1 and nucleotides 3035–6629 in BRCA2) for either gene with lifetime ovarian cancer risks (to 80 years) of 65% for 573 BRCA1 carriers outside the OCCR and 70% for 266 women with mutations within the OCCR (p = 0.18). Similarly there was no effect of position for BRCA2 with lifetime risks of 37% for 373 BRCA2 carriers outside the OCCR and 41% for 230 BRCA2 women with OCCR mutations (p = 0.17). There was a 10% higher cumulative incidence at most ages for breast cancer in those outside the BRCA2 OCCR, although lifetime risk was little different at 90% and statistical significance was not reached (p = 0.07). No such difference was seen for BRCA1 with virtually identical incidence curves (p = 0.25). DCIS was included as breast cancer. However, this only amounts to 1% of BRCA1 breast cancers and 2% for BRCA2. It is likely that nearly all of these would have become invasive as only 1/16 occurred after 60 years of age. Tamoxifen is not licensed in the UK for prevention. Only 23 mutation carriers took tamoxifen as part of the IBIS1 prevention trial and this is unlikely to have materially changed the penetrance estimates.

An estimate for breast cancer penetrance was also made for each 10–20 year birth cohort. A highly significant difference was identified with those born after 1960 having a breast cancer risk to 40 years of age of 40% compared to only 7.5% for those with a year of birth between 1900 and 1920 (Figure ​(Figure3:3: p < 0.00001). However, after exclusion of the index case the cumulative risk to 40 years dropped to between 21–23% for the birth cohorts after 1940. This was, nonetheless still a highly significant trend (p = 0.0005). After exclusion of the index case there was no significant birth cohort effect observed for ovarian cancer (p = 0.086). To assess the earlier age at breast cancer diagnosis on life expectancy we carried out a Kaplan-Meier survival analysis on the birth cohorts, again excluding the index case. There was no significant difference in survival from birth (Table ​(Table4;4; log rank df 6, p = 0.07), although there was a trend to better survival in the earlier birth cohorts. Indeed if the index case was included 21/83 (25%) index cases post birth year 1960 had died by 45 years of age, equivalent to a cumulative mortality of 35% to that age.

Breast cancer incidence was also assessed after family ascertainment. Incidence figures for breast/ovarian cancer are shown in Table ​Table5.5. These reflect the incidence in unaffected women at the time of family ascertainment and follow up was censored at the time of risk reducing surgery (oophorectomy/mastectomy). As the index case was used to identify the mutation usually on surveillance the incidence rates for these cases are artificially high. Excluding the index cases there was an incidence of 2.5–2.7 per thousand for breast cancer in proven carriers. Even including 40% of the follow up time and 80% of the breast cancers from the FDR unknown category (Tables ​(Tables11 and ​and2),2), this still gave an annual incidence of breast cancer of 1.98% for both BRCA1/2 mutation carriers (38/1917; 35/1763.6). An annual rate of 2% averaged over the risk period of 30–79 years would if anything indicate a higher risk than those indicated by the Kaplan-Meier analysis.

The previously reported positional effect of mutations for both BRCA1 and BRCA2 is not borne out by our analysis. No substantial effect of increased risk of ovarian cancer was seen in the respective ovarian cluster regions of each gene and only a borderline significant reduction of breast cancer risk was seen for BRCA2. Much of the OCCR association has been based on ratios of breast to ovarian cancer [10] or on the presence or not of ovarian cancer in the family [11]. Even this reliance on the presence of ovarian cancer for BRCA2 has been questioned by the report of 58% of BRCA2 related ovarian cancer families having mutations outside the OCCR [12]. Although the BCLC study on BRCA1 positional effect [10] included 356 families compared to our 223 families no absolute estimate of penetrance was made. Whilst the breast cancer incidence was lower in the central portion of the gene (nucleotides 2401–4190) (RR 0.71) in their analysis it was not possible to derive absolute risk figures for each portion of the gene. Additionally it is likely that our more extensive testing of unaffected relatives may provide a more accurate overall picture as reported here. Accurate estimates of cancer risk are essential for families and individuals undertaking genetic testing. Based on our analysis, it is questionable whether any account should be taken of the OCCR in each gene or indeed any substantial positional effect in genetic counselling.

It is also clear that for individuals undertaking predictive genetic testing in the context of families ascertained from cancer genetic clinics as opposed to population testing that risk figures similar to those derived in our study or the BCLC is quoted in our own clinics and we recommend that penetrance estimates are derived for the population being counselled. Our data are nonetheless at variance to a similar analysis carried out in North America [15]. A series of 1948 families were tested for mutations in BRCA1/2 in eight centres. 283 families with BRCA1 mutations were identified and 143 in BRCA2. The authors used statistical modelling to arrive at penetrance figures by 70 years of 46% (95%CI 39–54%) for BRCA1 and 43% (95%CI 36–51%) for BRCA2. The authors did not appear to take advantage of any further testing of relatives in the family. Whilst they corrected for potential ascertainment bias, they did not allow for the effects of modifier genes in these families and purely looked at attributable risk from BRCA1 and BRCA2 mutations alone. This was based on the apparent lack of heterogeneity in another study of Jewish families from North America [16]. What is particularly concerning is the risk attributed to "non mutation carriers" to 70 years. A figure of 5% as a general population risk for breast cancer may have been correct 20–30 years ago, but is certainly not the risk faced by women in the US or the UK today. Breast cancer risk to age 70 is 7.6% in the UK [17] and nearer 8% in the US. A correction for this difference might give penetrance figures of nearer 74% for BRCA1 and 69% for BRCA2. The decision not to include any adjustment in these families for the effects of modifier genes is questionable. The difference in penetrance obtained from the BCLC and from population studies strongly suggests the presence of additional genetic factors in high-risk families. We have recently reported that those testing negative for a family BRCA mutation are still at 3-fold relative risk of breast cancer [13]. This phenocopy effect was also seen in the Iceland data for their founder BRCA2 mutation, although to a lesser extent given the strong population based element of their analysis [18]. However, it is possible that modifier genes are more prevalent in some populations and that penetrance in North America is less affected by modifier genes than in the UK. The presence of these modifier alleles is now indisputable from recent genome wide association studies [19-21].

A potential criticism of our study is that we have not taken enough account of ascertainment bias and that additional adjustment maybe necessary beyond excluding the index case. An analysis using these adjustments was carried out in the North American study [15] and recent reports from the Cambridge group [22]. These studies did not take into account the widespread testing of relatives and as explained above the American study deliberately excluded any effect other than of the BRCA1/2 mutation. Whilst it is clearly interesting to know the effect of BRCA1/2 alone, women undergoing testing will want to know what their own specific risk of breast and ovarian cancer are, including that contributed by other potential "modifier" genes in their family. We must also acknowledge that confidence intervals in table ​table33 should also be wider due to forcing the data on unknown FDRs into a known category.

The high-risk women testing positive is also supported by the prospective part of our study. The 2–2.7% annual risk demonstrated is equivalent to the highest risk in a 10-year period (23% BRCA1; 30% BRCA2-Table ​-Table3).3). Although most of the breast cancers were detected by screening, only one was detected at a prevalence mammogram. These follow up risks are also supported by a similar follow up study in the Netherlands where 8 breast cancers occurred in 63 mutation carriers with a calculated annual risk of 2.5% [23].

Our own study and recent analyses from North America and Iceland demonstrate that women in the most recent birth cohort have a substantially higher risk of developing breast cancer than past cohorts [16,18]. The incidence of breast cancer in BRCA2 carriers has risen 4 fold in 80 years in Iceland (as has breast cancer in the general population) and we have observed a similar increase from <10% risk by 40 years in those born before 1930 to a 40% risk on those born after 1960, although this was less significant after allowing for ascertainment bias. It is, therefore, inappropriate to quote risks as low as 43–46% (based on population studies) for lifetime breast cancer risk to women in their twenties or early thirties if they test positive for a mutation in a high-risk family. Another potential effect of earlier breast cancer might be a reduction in life expectancy. With increasing survival from birth in the general population and improved survival from diagnosis of breast cancer we might have expected to see improved life expectancy. However, it would appear that these elements almost completely cancel each other out and there is no evidence for improved survival from birth in modern BRCA birth cohorts.