A family history of breast cancer will not predict female early onset breast cancer in a population-based setting

We studied the predictive value of a family history of cancer with regard to early onset of breast cancer in a large group of women with and without breast cancer. These women were unselected for having breast cancer, but they all had a family history of breast cancer, having at least one sister with breast cancer, as they were selected via their affected sister. These affected sisters (the index patients) had recently been diagnosed with breast cancer and originated from a cohort of breast cancer patients who were unselected for age of breast cancer or family history.

Index patients (affected sisters) had a recent diagnosis of primary invasive breast cancer or DCIS (ductal carcinoma in situ) and were consecutively asked to participate in a population-based study "PROSPECT" from 1996 to 2002. This population-based study was carried out at the Leiden University Medical Center, Department of Surgery, at the Diaconessenhuis, Leiden, and at the Erasmus Medical Center-Daniel den Hoed Cancer Center in Rotterdam, Departments of Medical Oncology, Radiotherapy and Surgery. At the Leiden University Medical Center and the Diaconessenhuis, all patients recently diagnosed with breast cancer (n = 514) were asked to participate in this study, irrespective of age. The participation rate was 92% (n = 471). At the Daniel den Hoed Cancer Center only patients under the age of 70 were asked to participate. Of the invited patients, 681 (96.5%) were willing to participate. From 1996 to July 2002, 1,152 patients participated and were entered into the database. The Medical Review Board of the Centers approved the study.

Apart from age of diagnosis, these index patients were unselected for age of breast cancer diagnosis, family history for cancer and history of breast cancer screening. The mean age of the index patients was 52.7 (SD 11.1). Index patients younger than 70 years in the study were comparable with those registered in the Comprehensive Cancer Center of The South-West Netherlands during the years 1996–1999, with regard to age distribution (data not shown). For index patients older than 70 years, our sample was comprised of few cases when compared to the almost complete cancer registry of the region. This is mainly due to the exclusion of these patients in the Daniel den Hoed Cancer Center. In first-degree relatives of index patients in our series, the overall breast cancer incidence was twice as high as in the general population, agreeing with the excess risk reported in other population-based studies [2, 3]. For 980 index patients, information regarding breast cancer screening history was available. Overall, 632 (64.5%) patients participated in any kind of screening program. 541 (85.6%) participated in the National Breast Cancer Screening Program (biannual mammography), 42 patients (all above the age of 50; 6.6%) participated in both the National Breast Cancer Screening Program and a High Risk Screening Program (annual mammography), and 49 (7.8%) participated in a High Risk Screening Program. In the majority of these screened women, breast cancer was diagnosed based on symptoms and not based on a finding in the screening program, namely in 286 (52.7%), 33 (67.3%) and 31 (73.8%) cases, respectively.

For each index patient the following disease characteristics were recorded: age at diagnosis of first breast cancer, the occurrence of bilateral breast cancer, and any other diagnosis of cancer. Based on interviews with the index-patients, for each index patient a detailed family history of breast cancer regarding the occurrence of malignancies in first- and second-degree relatives was assessed, and a pedigree was drawn. For each relative, the current age or, if deceased, age of death was assessed. For each relative affected with cancer, the age at diagnosis and the type of cancer (e.g. breast, ovary, and prostate cancer) was registered. Medical records of relatives were not checked, so we were not able to assess types of breast cancer in the family, or to discern bilateral breast cancer from breast cancer recurrences in family relatives besides the index patient. Therefore, the distinction between invasive breast cancer and DCIS, or unilateral and bilateral breast cancer was only assessed for the index patients.

The original cohort contained 1,152 index patients of whom 286 patients did not have sisters. These patients and their families could not be included in the study. The remaining 866 index patients had 1,987 sisters. These 1,987 women with and without breast cancer formed the study population.

We analyzed the impact of a family history of cancer on the early onset of breast cancer in 1,987 women (sisters). For each of these women it was determined whether or not they had been diagnosed with breast cancer and, if so, at what age. If they were not diagnosed with breast cancer and still alive, their age at the time of the interview with the index patient was assessed. If they were not diagnosed with breast cancer and deceased, their age of death was assessed. Additionally, their family history was analyzed. The following were considered to be first-degree relatives of these women: the affected sister (index patients), mothers, and daughters. To avoid dependency between women in families with more than one sister, sisters with breast cancer other than the index patient were excluded from the family history of the women under study. To explain, suppose a family with three sisters, in which the index has a diagnosis of breast cancer as well as one of the other sisters. All sisters, so both a sister affected with breast cancer and a sister unaffected with breast cancer, were included in the analysis as an outcome, with only the information regarding breast cancer in the index patient (the third sister), mother, daughters and 2nd degree relatives as predictors. The following were considered to be second-degree relatives: grandmothers, aunts (sisters of mother or father), and daughters of the affected sister. Early onset of breast cancer was defined as breast cancer diagnosed before the age of 50 or breast cancer diagnosed before the age of 30.

Family histories included the number of cases of breast cancer in first and second-degree relatives, the age at diagnosis of the first breast cancer, ovarian cancer, the combination of ovarian and breast cancer in one person, prostate cancer under the age of 60, and bilateral breast cancer or DCIS in the index patient.

First, we examined the effect of the various family histories on age at breast cancer diagnosis in the women by using a Kaplan-Meier model. Log rank tests were performed. Age of the women was taken as time variable. Follow-up time was censored at their age of the diagnosis of first breast cancer, their age at the moment of the interview with the index patient or the age when dying.

Second, by using Cox-regression, the age of first breast cancer diagnosis was univariate modeled as a function of the various types of family histories (covariates) yielding hazard ratios. Due to low prevalence, the hazard ratios could be considered as relative risks. To test the assumption of proportional hazards, an interaction term of a covariate and a time-dependent covariate was added [10]. A significant effect of that interaction term denotes the presence of a time-dependent effect and thus a violation of the proportional hazards assumption. As the estimated hazard ratios were not constant over time, we performed Cox-regression with a time-dependent effect, with age at breast cancer diagnosis in the women under study as the dependent variable. For each covariate, a model was built including this covariate and an interaction term of a covariate and a time-dependent covariate. Hazard ratios for the women's chances of developing breast cancer at age 30 and an estimate of the relative change with increasing age (a period of 10 years) of these hazard ratios were calculated.

Then, a multivariate analysis was performed, by using a Cox-regression model again with a time-dependent effect, and with age at breast cancer diagnosis in the women under study as the dependent variable. The model was built in a stepwise fashion, including variables that improved the overall fit of the model with statistical significance (p < 0.05). Again, a hazard ratio for the women's chances of developing breast cancer at age 30 and an estimate of the relative change with increasing age (a period of 10 years) of this hazard ratio was calculated.

The developed model was used to calculate a model score, based on the family history characteristics included in the multivariate model. Each family history characteristic yielded one point in this model. Then we verified how many women with breast cancer were identified correctly, by using different cut-off values of the model score. Furthermore, positive and negative predictive values of the model, with regard to an early onset of breast cancer in the women under study, were calculated.

Of the 1,987 women under study, 136 (6.8%) had been diagnosed with breast cancer. The incidence of breast cancer increased with age (Figure 1).

The 1,987 women included in the study had 32,791 first or second-degree relatives in total. A higher risk of breast cancer among these women was statistically significantly related to having at least 2 cases of female breast cancer in first or second-degree relatives, at least 1 case of early female breast cancer (i.e. under the age of 50 or 40), or bilateral breast cancer (Table 1). These effects were stronger for women under the age of 50 with breast cancer. The presence of one of these characteristics more than doubled the breast cancer risk at age 30 (Table 2). This risk decreased with increasing age.

In this population, having at least 2 cases of female breast cancer under the age of 40, having at least 1 case of female breast cancer under the age of 30, or having male breast cancer or prostate cancer under the age of 60 was rare (Table 1). Ovarian cancer and DCIS were not statistically significant associated with an early age at onset of breast cancer in the women under study (Table 1).

Four variables (at least 2 cases of female breast cancer in first-degree relatives (yes/no), at least 2 cases of breast cancer in first or second-degree relatives under the age of 50 (yes/no), at least 1 case of breast cancer under the age of 40 in a first or second-degree relative (yes/no), and any case of bilateral breast cancer (yes/no)) contributed to the presence of breast cancer at 30 years of age, both independently and with statistical significance (Table 3). Of these four family characteristics, the highest hazard ratio for developing breast cancer was related to bilateral breast cancer (HR: 3.47) and the lowest was related to having at least one case of female breast cancer under the age of 40 (HR: 2.06). All the hazard ratios decreased with increasing age, except the one related to having at least one case of female breast cancer under the age of 40 years.

Based on family history characteristics included in the multivariate model, a model score was calculated; each family history characteristic yielded one point in this model. As scores of three and four were rare (seven and three cases respectively), the score was recoded into '0', '1', and '2 or more'. Scores of 1 and higher were registered for about one-quarter of all (affected and disease-free) women, for about one-third of the women with breast cancer (regardless of age), for 40% of the women with breast cancer under the age of 50, and for 60% of the women with breast cancer under the age of 30 (Table 4).

As the model score increased, the age of breast cancer diagnosis decreased (Figure 2). The hazard ratio for developing breast cancer at 30 years of age was notably related to a model score of 2 or more (10.62; Table 3). With increasing age of the women, this hazard decreased sharply (hazard ratio at age 40: 4.56; at age 50: 1.96).

The positive predictive values for model scores of '1 or more' and '2 or more' for breast cancer before the age of 30 were 0,5% and 0.9% respectively, 4% and 11% respectively for breast cancer before the age of 50, and 8% and 13% respectively for breast cancer before the age of 70 (Table 5). The negative predictive value of the model scores varied from nearly 100% to 94%. The concordance index varied from 77% tot 94%.