Family History of Breast Cancer in Relation to Tumor Characteristics and Mortality in a Population-Based Study of Young Women with Invasive Breast Cancer

It has been hypothesized that breast cancers in women with an inherited predisposition to breast cancer are phenotypically and prognostically distinct. Although family history is a well-established etiologic risk factor for breast cancer, its relationship with survival remains unclear. A number of studies have observed improved survival for women with a positive family history (1–8) whereas others report little or no difference (9–15) or worse survival (16–18). Few studies have been population based (6, 8, 11, 15, 18, 19); most have been set within single institutions, specialty clinics, or high-risk family settings and may not include the wider spectrum of breast cancer and family history found in the general population. Studies of tumor characteristics and prognosis have yielded mixed results (20–28), and most prior work has been conducted in selected populations, particularly high-risk families. The lack of cohesion in results has been noted and recommendations have been made to examine these questions in population-based settings with careful consideration of age and comparison groups (29).

This study investigates the extent to which family history of breast cancer relates to tumor features and mortality in a population-based cohort of young women with invasive breast cancer and assesses the extent to which associations are accounted for by differences in screening or treatment according to family history.

A population-based cohort of women with invasive breast cancer was ascertained through the Cancer Surveillance System (CSS) of Western Washington, a participant in the Surveillance Epidemiology and End Results (SEER) Cancer Registry Program of the National Cancer Institute (30). Cases were women with a first primary, invasive breast cancer who originally completed in-person interviews for 2 population-based case–control studies of breast cancer, both of which have been previously described (31, 32). The first study ascertained all breast cancer cases diagnosed on January 1, 1983 to April 30, 1990 in Caucasian women born after 1944 who were residents of King, Pierce, and Snohomish counties at diagnosis. The second study ascertained all cases diagnosed before age 45 from May 1, 1990 to December 31, 1992 among women of all races in the same 3 counties. Both studies used standardized in-person interviews that elicited information in a comparable format. Interviews were completed on 83.6% (n = 747) of those eligible in the first study and 83.9% (n = 539) in the second study. Follow-up of this cohort of 1,286 was initiated to investigate factors affecting survival in young women with invasive breast cancer. Participants signed consent forms for the original and subsequent studies. The Fred Hutchinson Cancer Research Center Institutional Review Board approved all aspects of this research.

Information on risk factors for breast cancer and a detailed family history of cancer for the period preceding diagnosis was obtained through a structured in-person interview which included enumeration of all first (mother, sisters)- and second (aunts, grandmothers)-degree female blood relatives, followed by questions on each relative about birth and death year, and cancer history. Cohort members (or proxies) were recontacted beginning in 1993 to collect information on treatment and selected postdiagnosis exposures. The initial follow-up survey was completed for 1,157 (90.0%) cohort members (212 were proxy interviews). Information on tumor size, nodal status, stage, method of detection, and treatment was obtained through a detailed medical record review which was completed on 1,127 (87.6%) women. CSS data were used as the source of information on clinical factors when medical record review was not possible.

Paraffin-embedded preadjuvant therapy primary breast tumor samples were collected from area pathology laboratories. A single pathologist (PLP) conducted a centralized histopathologic review of all available tumor blocks or slides (n = 1,019 or 79.2%). Tumor tissue sufficient for immunohistochemical (IHC) analyses was available on 907 (70.5%) women. Protein expression of estrogen receptor (ER), progesterone receptor (PR), p53, Ki-67 proliferation–related antigen, HER-2, cyclin E, p27, and BCL-2 was assessed with immunoperoxidase assays as described previously (33). Antibody scoring protocols involved a subjective interpretation of staining intensity and/or the percentage of positive tumor cells and incorporated known expression patterns and cellular locations of proteins. For ER/PR, any staining above negative was considered positive; negative and low intensity BCL-2 stains were classified as “low” and intermediate/high staining was classified as “high;” Ki-67 proliferation index was computed as the average of the ratios of Ki-67–positive cells over 4 high powered fields with 25% or above considered high proliferation and less than 25% classified as “low;” any nuclear staining of tumor cells for p53 was considered positive; negative and “1+” staining for HER-2 were classified as negative, and “2+” and “3+” were considered positive; cyclin E and p27 expression were classified on a scale of 1 (negative) to 7 (highest intensity), with values of 1 to 4 classified as low and 5 to 7 classified as high for cyclin E, and values of 1 to 2 classified as low and 3 to 7 classified as high for p27. Positive and negative controls were included in each assay series; antibody batches were titered against known controls to assure standard testing over time.

BRCA1 and BRCA2 genotyping was conducted as part of a prior study with lymphocytes from women with either a first-degree family history of breast cancer or a diagnosis of breast cancer before age 35, 2 hallmarks of an increased likelihood of carrying a BRCA1/BRCA2 mutation (34, 35). Altogether, 323 cases in this cohort meeting either or both of these criteria were genotyped with single-strand conformational polymorphism (SSCP) followed by direct sequencing of all variant bands occurring in less than 5% of samples and supplemented with novel primers designed to provide redundant coverage in some regions, as previously described (34–36). The proportions genotyped did not vary by age or family history; a higher proportion of genotyped cases were alive at last follow-up and fewer had advanced stage disease than nongenotyped cases, due largely to a lag between interview and blood collection in the first case–control study. Analyses herein included data only on BRCA1/BRCA2 mutations identified clearly as disease associated (34, 35). The pathology and genotyping laboratories were blinded with regard to risk factors, treatment history, and vital status.

The CSS follows all cancer cases for vital status, including those who move outside of the area, through annual contact with follow-up physicians and linkage with databases such as the Washington State death files and the National Death Index. Vital status was also updated through ongoing contacts with the women (or proxies). Follow-up ended in April 2009, by which time 34.1% (n = 438) of the 1,286 cohort members had died. Of 848 women not reported to be dead as of April 2009, 97.4% were verified as alive within the preceding 12-month period and 99.8% were located within the preceding 36-month period. Overall, the mean and median follow-up periods accrued for living cohort members were respectively 19.8 and 19.3 years.

Cox proportional hazard modeling was used to estimate the relative risk (HR) of death associated with clinical characteristics and family history. Analyses considered both all-cause- and breast cancer mortality. Deaths not attributable to breast cancer were rare (88.4% of 438 deaths were confirmed as breast cancer related) and results were similar using both outcomes; herein, we present all-cause mortality. Cohort members were followed until the earliest of the date of death, date last known to be alive, or the end of follow-up. Results were similar using 2 endpoints, 10-year survival and survival through April, 2009; we present results for survival through April, 2009.

To account for the left truncation of survival times (women had to survive a minimum interval postdiagnosis to complete the original interview), we conditioned on the probability that a woman survived to time of interview (37). Unconditional logistic regression was used to obtain estimates of the relative risk (as estimated by the ORs and 95% CIs) for associations between indicators of inherited predisposition and tumor characteristics. Analyses were conducted with the statistical packages Stata (Stata Corporation) and SAS (SAS Institute Inc.).

All analyses accounted for the effects of age (continuous) and diagnosis year (exact). The following factors were evaluated for potential confounding of survival analyses on the basis of a 10% magnitude changes in risk estimates: race (white and nonwhite), smoking (ever and never), alcohol use (average lifetime weekly number of drinks), oral contraceptive use (ever and never), parity (nulliparous and parous), time since last birth, body mass index (kg/m² in quartiles), number of relatives at risk, mammogram history (ever/never, frequency in 5 years before diagnosis), method of detection (screening mammography, clinical breast exam, self/partner, and other), stage, tumor size, and nodal status. Potential confounding by treatment was assessed for surgery, radiation therapy (ever/never), chemotherapy (ever/never), hormone therapy (ever/never), and use of either of 2 commonly prescribed combination chemotherapy regimens [CAF: cyclophosphamide/doxorubicin/5-fluorouracil; CMF: cyclophosphamide/methotrexate/5-fluorouracil (83% of cases who received chemotherapy received one or the other)]. These factors were further tested for potential modifying effects on the association between family history and mortality. The distant and regional stage categories had to be combined for some analyses because of the small numbers of both distant stage cases overall (n = 24) and distant stage cases still alive at end of follow-up (n = 1). Similarly, women with varying numbers of affected nodes had to be combined into a positive node group for some analyses. To assess the effect of family history in the context of other accepted clinical predictors of mortality, a final multivariate model examined the influence of family history on mortality after accounting for age, year of diagnosis, tumor size, nodal status, grade, and ER/PR/HER2 status. Twenty-six adopted women were excluded, resulting in a final analysis cohort of 1,260 women. Fifteen women with at least one first-degree relative with unknown cancer status and no known breast cancer affected relatives (1.2%) were excluded from analyses of family history. In addition, 204 women (16%) without any known affected relatives had at least one second-degree relative with an unknown breast cancer status; for most of these women, cancer history was complete for all but 1 (n = 126) or 2 (n = 36) second-degree relatives. Analyses of family history were conducted 2 ways with identical results: excluding these 204 women and including them in the group with no family history; results presented here include these 204 women in the no family history group. For analyses of family history in relation to mortality and to tumor features, we considered effects both including and excluding women with mutations in BRCA1 or BRCA2, with similar results.

A description of the relationship of clinical and demographic characteristics to mortality is presented in Table 1. Advanced stage, larger tumor size, positive nodal status, bilateral breast cancer, and early year of diagnosis were associated with an increased risk of dying. History of a mammogram was associated with a reduced risk of dying.

The distribution of clinical characteristics according to family history is shown in Table 2. Similar proportions of women with a first-degree and second-degree only family history of breast cancer were diagnosed with local stage disease (60.7% and 61.1%), whereas the proportion was slightly lower (55.1%) in women with no family history of breast cancer. Results for nodal status were similar to those for stage. Women with a first-degree family history had the highest portion of small (≤2 cm) tumors (58.3%), and women with only a second-degree family history and no family history had smaller proportions with a small tumor (51.2% and 50.6%, respectively). A higher proportion of women with a first-degree family history reported a mammogram in the preceding 5 years (56.3%) than women with only a second-degree family history (42.3%) and no family history (31.2%). Women with a family history were more likely to have had their breast cancer detected by screening mammogram (14.5% in first degree and 11.2% in second degree compared with 9.8% in those with no family history).

Women with a first-degree family history of breast cancer had a 40% reduction in the risk of dying compared with women with no family history (95% CI: 0.5–0.8). The presence of only a second-degree family history did not influence the risk of dying (HR: 0.9; 95% CI: 0.8–1.2; Table 3). The risk of dying was similarly decreased for cases with an affected mother (HR: 0.6; 95% CI: 0.4–0.8) or sister (HR: 0.5; 95% CI: 0.3–0.9), and the absence of an effect associated with having only a second-degree family history was consistent across cases with affected aunts versus grandmothers and maternal versus paternal side relatives (data not shown). The risk of dying was lower in women with greater numbers of affected relatives (P_trend = 0.05) and slightly lower in women with a relative diagnosed before age 45. Neither family history of bilateral disease nor family history of ovarian cancer had any effect on the risk of dying (data not shown).

The associations involving family history could not be ascribed to mutations in BRCA1 and BRCA2, as results were unchanged by exclusion of carriers (Table 3). Associations were also unchanged following adjustment for stage, tumor size, nodal status, mammography history, method of detection, body mass index, recency of last birth [the latter 2 factors influence the risk of dying in this cohort, as previously reported (33, 38)], education, income, race, smoking, alcohol intake, surgery type, radiotherapy, chemotherapy, hormonal therapy, or treatment combinations. There was little or no evidence of effect modification by clinical characteristics or treatment (Table 4).

The tumors of women with a first-degree family history had a somewhat more favorable prognostic profile than those in women with no or only a second-degree family history (Table 5). Associations were similar when we included and excluded BRCA1/2 mutation carriers; results presented exclude carriers. Compared with women with no family history, tumors in women with first-degree family history were less likely to be large in size. Tumors in these women were also less likely to be in the higher risk group for almost all other tumor features, although these associations were within the limits of chance.

Last, we considered the risk of dying in relation to family history in conjunction with established tumor features (Table 6). The presence of a first-degree family history of breast cancer was associated with a significant reduction in the risk of dying (HR: 0.7; 95% CI: 0.5–0.9) independent of the prognostic effects of age, year of diagnosis, tumor size, nodal status, grade, and ER/PR/HER2 status. Results were similar upon exclusion of BRCA1/BRCA2 mutation carriers.

Our results indicate that women diagnosed with invasive breast cancer at a young age who have features of inherited predisposition, including a mother or sister with breast cancer, one or more relatives with breast cancer diagnosed at an early age, and/or multiple affected relatives do not have an increased risk of dying and in fact, have a substantively reduced risk of dying when compared with breast cancer cases with a negative family history of breast cancer. This study was unique in its focus on women diagnosed with breast cancer at a young age, and most past studies of family history and survival have had insufficient power to focus on young women, in which mortality rates are so high. In turn, it is worth noting that the results of this study are not necessarily generalizable to women diagnosed with breast cancer at older ages.

The survival advantage afforded in this study for cases with first-degree family history is compatible with findings of improved survival in a number of prior studies (1–5, 7), including our preliminary report on a portion of the cases in this analysis (6), but in contrast to findings in other studies of worse survival (16–18), and no difference (9, 10, 12, 13, 15, 39), although these past studies largely had limited numbers of young women. Only 4 studies reported on young cases, one finding somewhat worse survival (14), 2 finding improved survival (7, 8), and another finding no association (15) between family history and survival, but only one of these studies excluded BRCA1/BRCA1 carriers and another was exceedingly small in sample size. Past studies have varied in their definitions of positive family history. We examined family history in a number of ways, including by the presence of breast cancer in any first- versus only second-degree relatives, age of diagnosis in relatives, numbers of affected relatives, and individual relative type, and found considerable consistency in the observation of a reduced risk of death in relation to first-degree family history and multiple affected relatives. One recent report of a nonstatistically significant reduction in the risk of dying related to first-degree family history of breast or ovarian cancer (HR: 0.86; 95% CI: 0.71–1.05) noted a further improvement in survival for those with multiple affected first-degree relatives (HR: 0.21; 95% CI: 0.03–1.49), but this was based on limited follow-up and only one death in those with multiple affected relatives (8).

The infrequent nature of BRCA1/BRCA2 mutations, even in women with a first-degree family history (35, 40), suggests they likely do not play a major role in any association between family history and survival. A strength of this study was the ability to assess to what extent, if any, an effect of family history on survival was attributable to BRCA1/BRCA2. We genotyped the majority of cases with a first-degree family history (172 of 222) and results were unchanged when we included or excluded mutation carriers.

A primary goal of this study was to consider whether associations between family history and survival might be explained by differences in factors such as mammographic screening, method of cancer detection, and treatment on the premise that these factors could vary by family history. Although cases with a first-degree family history more often had smaller tumors and cases with either a first- or second-degree family history more often had had a prior mammogram, neither factor seemed to account for the observed reduction in mortality. Similarly, there was no evidence that treatment modified the association between family history and mortality, nor does the improved survival we observed seem to be attributable to other potential confounders.

Few studies have considered tumor characteristics in relation to family history and those that have were confined largely to basic features such as nodal status and tumor size. In our study, women with a first-degree family history had tumors with a generally more favorable prognostic profile as assessed by nodal status and size. Smaller tumor sizes and/or less advanced stage for family history–positive cases has been reported in some (1, 5, 39, 41, 42) but not other (2, 4, 10, 12, 14) prior studies. Although past studies have not reported on mammography, a hospital-based study of 2,256 breast cancers found that not only family history–positive tumors were smaller but also that they were more likely to be detected without symptoms, consistent with the heightened proportion with prior mammography histories we observed (42). Despite finding that first-degree family history–positive cases had smaller tumors, the association between family history and the risk of dying did not vary nor was confounded by stage, nodal status, or size.

Compared with women with no family history, the tumors of women in our study with a first-degree family history were more likely to be ER positive and less differentiated. The propensity for BRCA1 (but not BRCA2) mutation carriers to have ER-negative tumors is well established (43, 44). Our tumor analyses exclude most of the expected mutation carriers and thus our finding on ER is broadly compatible with previous findings among noncarriers of BRCA1/BRCA2 mutations with a positive family history (45). A few prior studies have examined ER and reported little or no difference according to family history, but these studies were not focused on young women in which the proportion of ER-negative tumors is much greater (5, 12, 39). Molino and colleagues found that cases with a positive family history were more likely to have ER-positive tumors but this study included few younger women (42). Another recent report in women of all ages found no association between family history and tumor classified by ER, PR, and HER2 (28). In our study, the tumors of women with a first-degree family history had more favorable (though not statistically different) profiles with regard to a number of other prognostic features, including grade and expression of PR, BCL-2, p53, and HER2. Women with only a second-degree family history had tumor profiles that more closely resembled women with no family history.

Strengths of this study include its population-based design, large sample size, and centralized pathology review and tumor analysis. The follow-up period was extensive and longer than most past studies, reducing concern that observed relationships are a short-term phenomenon. Our findings are strengthened by the ability to assess the potential confounding effects of mammographic screening, the method of cancer detection, and treatment regimens.

Our findings must be considered in light of potential limitations. A population-based study offers broader generalizability to the wide spectrum of family histories in the general population and especially for the modest to minimal family histories usually seen outside of high-risk clinics. However, generalizability may be tempered by the extent to which the 15% of women otherwise eligible for the study who never participated (a group which experienced greater 5-year mortality) differ from participants in terms of the association between family history and mortality. Similar to most studies, we relied on self-reported family histories; however, multiple validation studies indicate that self-reported family histories of breast cancer are largely accurate (46–49). Interpretability of studies of survival outside of a clinical trial can be hampered by the unavailability of treatment information and/or the nonrandomized nature of treatment. Our medical record review elicited extensive, comprehensive treatment detail and given the high proportion of reviews completed and the high proportion for which complete therapy information was obtained, it is unlikely that missing information accounts for these results. We saw no evidence that treatment modified or confounded results. Despite having information on mammography and the method of detection, our finding of a reduced risk of dying in women with a first-degree family history may nonetheless be, in part, attributable to unmeasured residual confounding involving behaviors associated with having a family history. This topic warrants further study.

There are also technical considerations for the laboratory analyses. BRCA1/BRCA2 genotyping was carried out in the mid-1990s with SSCP (34–36). Although a widely accepted technology at the time, newer methods detect a higher proportion of mutations (50) and some mutations were likely missed. BRCA1/BRCA2 genotyping was carried out on only 25.6% of the cohort, targeting subgroups most likely to carry a mutation. Prior studies, including our own in this population (34, 35) and others (51–53), have shown relatively low mutation frequencies in women with a family history and/or an early age of diagnosis (<10%–12%) and considerably smaller carrier frequencies outside of these higher risk groups, thus, it is unlikely that many mutations were missed in the untested group. Tumor tissue for IHC analysis was unavailable for 29% of participants. Comparison of women IHC tested and not tested showed no differences in vital status, stage, tumor size, nodal status, and family history. Tissue availability varied slightly by age, in that women with unavailable specimens were younger than those with available tissue (P < 0.01). This is not unexpected because tissue unavailability was greater for the earliest diagnosis years, imposing disproportionate losses in years when a greater proportion of cases was very young (an artifact due to the birth year eligibility criteria in our first study) compared with later years when the age spectrum went up to age 44.

Our results point toward better survival for women diagnosed with breast cancer at a young age that have certain hallmarks of inherited predisposition, including first-degree family history of breast cancer and multiple affected relatives. Larger replication studies are needed. Because the vast majority of women with a first-degree family history do not carry mutations in BRCA1/BRCA2, these findings suggest that other shared genetic factors may contribute not only to an increased risk of developing breast cancer, but also possibly to distinctive tumor features and prognosis. To the extent that these associations are replicated and extended, they could be important in terms of prognostication and treatment strategies. The greatest benefit would come as mechanisms and specific pathways that may underlie disease progression among women with varying inherited predisposition profiles are identified.

No potential conflicts of interest were disclosed.

The authors thank the study participants and area physicians for their generous contributions to this study and the many staff who completed the data collection activities.

This work was supported, in part, by grants and contracts R01-CA-63705, R01-CA-41416, N01-CP-95671, R01-CA-59736, and R21-CA-66186 from the National Cancer Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.