Oral Contraceptive Use and BRCA Penetrance: A Case-Only Study

Women with deleterious mutations in BRCA genes are at increased risk of breast cancer and ovarian cancer, and often develop cancer at a young age. Estimates of the lifetime cumulative risk (penetrance) of breast cancer associated with BRCA mutations range from ∼80% in studies on high-risk families (1-5), to around 45% in population-based studies (6-12). A sizable proportion of mutation carriers, however, does not develop breast cancer at all or develop it only late in life. Therefore, the penetrance of the genetic trait may be regulated through other genetic or nongenetic factors. The hypothesis of an interaction of genetic and environmental risk factors is supported by the increasing penetrance and declining age at onset over succeeding generations that several authors have described in BRCA-positive families (13-16).

This multinational Case Only Study (C.O.S.; refs. 17, 18) aimed to investigate the interaction between genetic susceptibility, nutrition, and environmental factors in the occurrence of breast cancer in young women. We hypothesized that among environmental factors (19, 20) affecting the penetrance of hereditary breast cancer genes, oral contraceptives use may play a significant role (13, 21).

Long-term use of oral contraceptives is an established risk factor for sporadic breast cancer (22). Combined oral contraceptive drugs may also modify the risk of early onset breast cancer, especially in women who started using pill very young (23-27). Oral contraceptives have been found to increase the risk of early-onset breast cancer among BRCA1 mutation carriers but might protect them against ovarian cancer (21). A case-control study involving 1,311 pairs of women with BRCA mutations (28) reported that, among BRCA1 mutation carriers, women who first took oral contraceptives before 1975 (when drugs were likely to contain high doses of hormones), who used them before 30 years of age, or who stayed on them for 5 or more years have a statistically significant increased risk of early onset breast cancer. A retrospective cohort of 1,593 BRCA mutation carriers (29) confirmed an increased risk for oral contraceptive users especially for those starting use at 21 to 24 years of age without, however, any difference for calendar time of starting use. Jernstrom and colleagues (30) showed an interaction between teenage oral contraceptive use and BRCA carrier status in early-onset breast cancer cases. The same group has recently suggested an association between BRCA mutation status, CYP17 genotype (gene involved in steroid hormone metabolism), and early oral contraceptive use (31).

The aim of the present work is to evaluate the interaction between oral contraceptive use and genetic susceptibility using a case-only approach.

The C.O.S. study (17, 18) was supported by the European Community. Between January 2001 and April 2004, 3,123 women were recruited from nine centers in seven countries (Estonia, France, Germany, Israel, Italy, Scotland, and Slovenia). Eligible study subjects were women with a breast cancer diagnosis up to the age of 40 y in France, Germany, Italy, and Scotland and, given the small size of countries, up to the age of 45 years in Estonia, Slovenia, and Israel. The number of recruited patients by country is showed in Table 1.

Collaborating centers used different procedures to recruit patients into the study. In Scotland, Israel, and Germany, recruitment was mostly based on hospitals or family clinics; in France, Slovenia, and Estonia, on the local cancer registries; and in Italy, by advertising the study in the media. Such procedures were approved by each center institutional review board. All study subjects received information about the study and provided written consent.

Participants were requested to produce a copy of their pathology report and to fill in self-administered questionnaires on family pedigree and on family history of breast and ovarian cancer (to compute probability of BRCA mutation), dietary habits, and other life-style aspects before diagnosis. We also inquired if one or more family members had been tested for BRCA1 or BRCA2 mutation, and on the results of the test. However, test results were available in a minority of cases and for the purpose of this analysis their classification was only based on the “a priori” probability of mutation estimated from the family history.

Two hundred seventy women were excluded from the study either because they lacked a complete family history, age at diagnosis, or information whether they had used oral contraceptive or not.

The C.O.S. study was based on the collection of a few key information from participants (32). Women were classified according to their probability of carrying a high-penetrance mutation as estimated from their family history of breast and ovarian cancer. The odds of a given exposure among breast cancer cases who most likely carry a high-penetrance mutation because of a highly predictive family history (“genetic cases”) were compared with the odds computed for breast cancer cases who most likely are not mutation carriers because no other case occurred in an otherwise informative family (“sporadic cases”).

A major requirement of a case-only study is an unbiased classification of “genetic” and “sporadic” cancers. Several genetic risk assessment methods are available to estimate the probability of BRCA mutation in individuals to select them and their families for molecular diagnosis (33). Empirical methods based on the number of breast cancer and ovarian cancer cases in the family, age at diagnosis, and occurrence in subsequent generations, ignore data from unaffected relatives (34-37), and may overestimate the probability of mutation in large families with few affected members. By contrast, methods based on genetic models consider information from all relatives, whether affected or not. Berry et al. (38) and Parmigiani et al. (39) developed a method and a software (BRCAPRO) based on Bayes' theorem that requires data on all first- and second-degree relatives of the proband, and incorporates as prior probabilities incidence rates in the United States population, and allele mutation frequencies and penetrances estimated from studies in families with several breast cancer or ovarian cancer cases (4, 39, 40).

The C.O.S. study was carried out in countries with breast cancer incidence ranging from 50 to 100 per 100,000 women per year (41). In such a context, the application of a Bayesian model requires country-specific assumptions of sporadic breast cancer and ovarian cancer incidence; furthermore, breast cancer incidence has increased over generations, both in the general population (42) and in mutation carriers (13-16) and single age-specific incidence curves, as available from cancer registries, and penetrance curves estimated from high-risk families, do not accurately describe the disease risk in succeeding generations.

In our study, we developed a computer program (C.O.S. software) to estimate the risk of mutation when incidence and penetrance are increasing over generations. The software is based on the same Bayesian logic as Parmigiani's method (38, 39) but is also able to evaluate third and if necessary fourth degree relatives, and allows incorporation of a hypothetical third BRCA gene. To estimate BRCA mutation probability, the C.O.S. software requires sex, age or age at death, age at breast cancer or ovarian cancer diagnosis, and age of diagnosis of second or contralateral breast cancer, in each participant's family member. We estimated country-specific general population breast cancer incidence by birth cohort from cause-specific mortality data (43) and population-based cancer survival data (44, 45), using a mathematical model of the relationship between incidence, survival, mortality, and prevalence (46). To estimate breast cancer and ovarian cancer penetrance in those with a deleterious BRCA mutation, we used pedigrees from families identified at the Milan National Cancer Institute with a BRCA mutation (47). We simulated various age- and birth cohort–specific curves for the incidence of genetic breast cancer and ovarian cancer, and chose a set that produced estimates closest to the observed family histories. We examined the performance of the COS software for Italy in predicting mutations in 131 Italian high-risk families in which direct sequencing or combination of direct sequencing and denaturing high performance liquid chromatography of amplicons found 25 deleterious mutations (17 BRCA1 and 8 BRCA2). Considering a probability of 0.1 and over as indicating a positive prediction, we compared sensitivity, specificity, and area under the receiver operator characteristic curve for the COS and BRCAPRO (Version 4.0) programs. COS had higher sensitivity (76% versus 72%) but lower specificity (48% versus 58%) than BRCAPRO. However, the area under the receiver operator characteristic was numerically better at 0.71 [95% confidence interval (CI), 0.65-0.78] for C.O.S., than the 0.67 (95% CI, 0.60-0.73) for BRCAPRO, suggesting that COS was better able to distinguish between families with and without a deleterious mutation but overestimated the probability of mutation. Also the COS software for Scotland was compared with other genetic models to estimate the probability of BRCA mutation, showing higher sensitivity and slightly lower specificity (48).

The aim of the study was to investigate environmental exposures before breast cancer diagnosis. We used self-administered questionnaire addressing oral contraceptive use, menstrual and reproductive history, active and passive smoking, radiation exposure in childhood, body weight, and physical activity at different ages, usual dietary intake, and some medical conditions. Participants were also instructed to measure their height, weight, and hip and waist circumferences in a standard way.

Referring to oral contraceptive, women were asked about their past and current use, at what age they began using pill and what age they stopped, and about the total duration of use.

The same data entry system was available in different languages and allowed the computation of summary variables useful to statistical analysis. Consistency and validity checks were done at the coordinating center (Department of Preventive and Predictive Medicine of the Milan National Cancer Institute).

The statistical analysis aimed to test the interaction between oral contraceptive use and genetic susceptibility. According to a case-only methodology, the frequency of exposure among “genetic” cases was compared with the frequency among “sporadic” cases. We carried out the basic analysis comparing breast cancer genetic cases (with probability of BRCA mutation of ≥45%) and sporadic breast cancer cases (with probability of BRCA mutation of <5%). We choose these cut off points to guarantee in the genetic group a probability of mutation >80% (83% on average), and in the sporadic group a very low probability of mutation (2% on average).

The means for all continuous variables in patients with high probability of mutation were compared with those of patients with a low probability by using Student's t test. χ² test was used to compare frequencies and percentages in relation to probability of mutation. An unconditional logistic regression model was used to compute the interaction odds ratio (OR) and 95% CIs; in our model, the dichotomous dependent variable was the genetic or sporadic nature of the case.

The following covariates were considered as potential confounders according to a priori hypotheses: center, age at diagnosis (in quintiles), education (none or primary school, high school, degree, or more), parity (defined as zero, one, two, three, or more), breastfeeding (no, 1-6, 7-12, >12 mo), and age at first live birth (in quintiles).

In Italy, we contacted again all patients whose questionnaires containing missing values on age at starting pill, duration of use, and on potential confounders such as parity, age at first live birth, and breast feeding. As missing values on confounding factors may bias the results, we also produced a statistical analysis restricted to the Italian series of patients, 96% of whom had complete information. A P value of <0.05 was taken to be significant. All statistical tests were two sided. The analyses were carried out using the STATA 8.0. statistical package.

Table 1 describes the distribution of cases by countries and probability of mutation. We analyzed key information from 1,715 C.O.S. women, 382 classified as genetic cases (probability of mutation, ≥45%) and 1,333 classified as sporadic cases (probability of mutation, <5%).

Characteristics of the study population and comparison of genetic and sporadic cases are reported in Table 2. Genetic cases were significantly younger and more educated; they showed a higher frequency of nulliparity and, among parous women, of those with three or more children. There were no statistically significant difference in mean age at menarche, age at first live birth, months of breast feeding, and usual body mass index. Oral contraceptive has been used by 74.9% of the genetic cases and by 71.6% of the sporadic cases. Among women who used oral contraceptive, the age at start was slightly earlier and the average duration of use was slightly shorter in genetic than in sporadic cases.

We examined the interaction between oral contraceptive and genetic susceptibility by a multiple logistic regression model (Table 3). The adjusted interaction OR of breast cancer comparing ever users with women who had never used oral contraceptive was 1.3 (95% CI, 1.0-1.7). Table 3 reports the interaction OR of genetic breast cancer for each quartile of age at start using pill. This analysis by age at first use included the 1,515 participants with complete data. The higher risk was for women who started using oral contraceptive at age 18-20 (interaction OR, 1.6; 95% CI, 1.1-2.3), but women who started before the age of 18 years were not at increased risk (interaction OR, 1.1; 95%CI, 0.6-1.8). Women probably carrying a BRCA mutation who began taking pill after the age of 25 years were not at increased risk (interaction OR, 1.0; 95% CI, 0.6-1.4). Due to missing values, adjustment for reproductive variables was possible only for a subset of 1,457 cases. Results however showed only trivial changes (interaction OR, 1.3; 95% CI, 1.0-1.8 comparing ever versus never users).

C.O.S. participants took oral contraceptive for an average of 6 years (72.8 months) without significant difference between genetic and sporadic cases (P = 0.30; Table 2). The analysis on duration of oral contraceptive use showed an interaction OR slightly higher for short duration of use (≤5 years) than for longer duration, but the difference was not statistically significant (P = 0.32; data not shown).

This analysis include 846 Italian C.O.S. participants, 184 genetic, and 662 sporadic cases.

The interaction OR of breast cancer comparing ever versus never users oral contraceptive after controlling for age at diagnosis, education, parity, age at first live birth, and months of breast feeding was 1.4 (95% CI, 0.9-2.0; Table 4). Table 4 reports the interaction OR of genetic breast cancer for the same categories of age at start using oral contraceptives of the main analysis. The interaction OR was higher for women starting at the age 18-20 (interaction OR, 2.0; 95% CI, 1.2-3.3), but women starting in their early teens were not at increased risk. However, the trend of decreasing OR for increasing age of starting oral contraceptive use was statistically significant. There was no difference according to the duration of oral contraceptive use (data not shown).

The multinational C.O.S. study on gene-environment interaction in the occurrence of breast cancer in young women was designed to test several specific hypotheses based on the available knowledge on the function of BRCA1 and BRCA2 proteins, namely their role in DNA repair and their role in controlling the effect of estrogen exposure (49). C.O.S. secondary aim was the development of primary preventive recommendations for high-risk families.

In a case-only study, an interaction OR significantly different from one indicates a gene-environment interaction, i.e., that the factor under study affects the incidence of genetic cancer by a greater (or lesser) magnitude than the incidence of sporadic cancer. The major contribution of the present study is that the relative risk of oral contraceptive use is greater for women with high probability of BRCA mutation than for women with a low probability.

This positive interaction was expected, because among the multiple functions of BRCA gene products, there is the regulation of estrogen signaling (50-52). We therefore corroborated one of the C.O.S. a priori hypothesis that oral contraceptives may increase breast cancer risk with a greater effect in presence of BRCA mutation. Women with a genetic predisposition may be more vulnerable to exogenous hormones. The Narod and collegues' case-control study (28) involving 1,311 pairs of women with BRCA1 or BRCA2 mutations reported that, among BRCA1 mutation carriers but not among BRCA2 mutation carriers, women who first took oral contraceptives before 1975, who used them before age 30 years, or who stayed on them for 5 or more years have a statistically significant increased risk of early onset breast cancer. The Brohet and collegues' retrospective cohort (29) confirmed the association with an increasing trend with the duration of use and the greater risk for women starting use before the age of 25 years. Those investigations, however, was not designed to analyze whether oral contraceptive risk is greater among carriers or non carriers. In the C.O.S. study, almost all the patients started using oral contraceptives after 1975 and the risk of carriers was increased with respect to noncarriers. The risk associated with oral contraceptive use in mutation carriers may be greater than our estimate because women with positive family history may be recommended not to use oral contraceptives or may decide to stop their use when a relative develops breast cancer. This may cause an underestimation of the interaction OR measured in a case-only study.

The relationship between oral contraceptive and genetic breast cancer was greater in selected subgroups. When we examined the risk by quartile of age at start using pill, the higher interaction OR was observed for women who started between 18 to 20 years of age (OR, 1.6; 95%CI, 1.1-2.3). Among women who started earlier (before 18 years) there was no difference between genetic and sporadic cases both in the main analysis and in the Italian subset. The interaction with genetic susceptibility may change with age on the basis of different hormonal responses or pathologic conditions. We have to consider that oral contraceptive pill before the age of 18 years is frequently prescribed for reasons other than avoidance of pregnancy, such as to reduce irregularity in menstrual cycle after menarche, or to reduce pain, acne and hirsutism, or for polycystic ovary. These conditions may be associated with breast cancer risk and hormonal treatments might contribute to reduce the risks.

The results, however, suggest that the excess risk of probable mutation carriers with respect to noncarriers is not apparent until age 18 to 20 years and after that decreases with increasing age at start. There may be no excess risk in women starting oral contraceptive use at age 25 years or later as previously shown by Narod and collegues (28).

The strength of our study is the case-only design. It allows to express how greater is the association of the environmental exposure with genetic cancers than with sporadic cancers, which is the statistical interaction between the exposure and the genetic trait. Referring to the classification of cases, the estimated average probability of high probability group—the genetic cases—was 83%; the average probability of mutation of the low probability group—the sporadic cases—was 2%. This implies that 17% of patients classified as genetic may have been false positive, and that 2% of cases classified as sporadic false negative, i.e., actually carriers of BRCA mutation. Such misclassification implies an underestimation of the interaction OR and the risks associated with oral contraceptive exposures for BRCA1 or BRCA2 mutation carriers are likely somewhat higher. Genetic test results were available for 143 cases, 52 positive, and 91 negative. History of oral contraceptive use was most frequent among the former with an age adjusted interaction OR of 6.1 (95% CI, 1.2-30.9), suggesting that the overall results of the study underestimated the effect.

We also tested the relationship with oral contraceptive use comparing the intermediate mutation probability group (5-44%) versus the low probability group (<5%). The interaction OR was 1.0 (95% CI, 0.8-1.3), suggesting a fairly high specificity of our classification.

Knowledge of breast cancer susceptibility genes, along with the introduction of predictive genetic testing, has made it possible to identify women at increased risk for inherited breast and ovarian cancer. Options currently available for these women include surveillance programs aimed at early detection, prophylactic bilateral mastectomy, and prophylactic oophorectomy, whereas nonsurgical primary prevention options are not yet firmly established. Oral contraceptive use has been considered as a preventive measure against ovarian cancer in BRCA mutation carriers. However, the effect of oral contraceptive use on ovarian cancer in BRCA mutation carriers is still unclear (53). Narod and collegues (54-56) reported a protective effect of oral contraceptives on the risk of ovarian cancer among BRCA1 or BRCA2 carriers, but Modan and collegues (57) did not find any effect.

Whatever the effect on ovarian cancer, C.O.S. results suggest that it would be prudent to avoid starting oral contraception before the age of 25 years. Further studies are warranted to clarify the modifying effect of age in the interaction between oral contraceptive use and BRCA penetrance.

No potential conflicts of interest were disclosed.

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.