Age-specific breast cancer risk by body mass index and familial risk: prospective family study cohort (ProF-SC)

Body mass index (BMI) is an intriguing risk factor for breast cancer because its association with the disease depends on time of life. Greater BMI has been found to be associated with an increased risk for postmenopausal women [1–9], while for premenopausal women, young women, and even adolescent girls [2, 4, 6, 8, 10–14], greater BMI has been found to be associated with a decreased risk. These findings have been consistent across different racial and ethnic subgroups [2, 14, 15] and across both case–control and cohort designs globally [1–4, 6–13], suggesting they are not a consequence of systematic biases [9].

BMI is an important risk factor because it is potentially modifiable. The fact that greater BMI appears to be protective at young ages, yet has the opposite association in later life, presents a potential problem for simple cancer control messaging; therefore, its consequences need to be quantified. A prospective study and a case-control study have found that the increased risk associated with higher BMI increases with time after menopause but is not evident until 10 years post menopause [16, 17].

A better understanding of how the BMI-associated risk varies with age and menopausal status is needed. It is interesting that, from genome-wide association studies, genetic risk scores based on single-nucleotide polymorphisms (SNPs) that predict higher BMI in childhood or adulthood are associated with lower risk of both premenopausal and postmenopausal breast cancer [18, 19]. Under the assumptions of Mendelian randomization, the authors concluded that these relationships were causal (even though those SNPs explained only a small proportion of the variation in BMI), thus lending additional support to the evidence that the effect of BMI varies by time of life.

Family history is another important risk factor for breast cancer that could not exist without there being a very strong gradient in underlying familial risk. To explain an overall average estimate of a 2-fold increased risk associated with having an affected first-degree relative, there must be at least a 20-fold inter-quartile risk ratio across the underlying familial causes [20]. This gives reason to consider family history not solely as a binary construct but rather as an underlying continuous measure that reflects this large gradient. Underlying familial risk can be predicted from family history using risk models that use pedigree information including age of onset of affected relatives. It is becoming increasingly possible to better differentiate women according to underlying genetic risk using SNP-based scores [21, 22]. Familial risk prediction is likely to improve with larger genome-wide association studies and the use of more informative statistical methods to create better SNP-based and family-history-based risk scores.

With the advent of gene panel testing for high-risk mutations in known breast cancer susceptibility genes [23–25], family cancer clinics will screen increasing numbers of women with a family history of breast cancer. The vast majority of these women will, however, not be found to carry a mutation that can currently be classified as deleterious. Therefore, a key clinical issue is risk management advice for women at familial risk who are found not to carry high-risk mutations. To resolve this issue, it is essential to know if their breast cancer risk factors are the same, and if the risk associations are of the same magnitude, as they are for women in the general population.

This issue of multiplicative and additive interaction with familial risk must be considered for each risk factor. If there is no difference in strength of associations by familial risk and the study is well-powered, advice on that risk factor’s relevance can be confidently given to women across the full spectrum of familial risk. In theory, if there are no interactions between risk factors on the multiplicative scale then there will be additive interactions [26, 27]. Knowledge of the extent of disease association by familial risk will enable prevention and screening measures to be appropriately offered to women.

To address the issues of whether breast cancer risk associated with BMI depends on the age of a woman, her menopausal status, and her underlying familial risk, we conducted a prospective study of women across broad ranges of age and familial risk at baseline.

The Breast Cancer Prospective Family Study Cohort (ProF-SC) comprises baseline and follow-up data from the Breast Cancer Family Registry (BCFR) and the Kathleen Cuningham Foundation Consortium for Research into Familial Breast Cancer (kConFab) (for full details see [28]). These prospective family cohorts are enriched for familial risk of breast cancer and have accumulated up to 20 years of follow up. The BCFR is a collaboration of six breast cancer family studies from the USA, Canada and Australia, and the protocols and data collection have previously been reported for the baseline studies [29] and the follow-up studies [28]. kConFab is an Australian and New Zealand breast cancer family study, and details of the core resource [30] and follow-up study [28, 31] have been previously reported. Ethics approval for the six sites of the BCFR and for kConFab was granted by the applicable human research ethics committees at the participating institutions. All participants in the BCFR and kConFab provided written informed consent before participation.

Figure 1 shows the distribution of predicted lifetime breast cancer risk based on BOADICEA. The second peak starting at 60% lifetime risk is almost entirely due to identified BRCA1 and BRCA2 mutation carriers.

Figure 2 shows that the HR estimates for the association between greater BMI and breast cancer risk change from all being negative to all being positive when moving from age at baseline < 40 years to age > 60 years. The same changes in risk occur in moving from pre- to postmenopausal status.

Table 3 shows the model fits after taking familial risk into account. The most parsimonious best fitting model was model V, which shows that after the BMI association was fitted as a function of menopausal status (HR = 3.36, P < 0.001), there was evidence of an association with familial risk (as represented by the log 1-year BOADICEA estimate in model V; HR = 2.05, P < 0.001). The other models show that there was no evidence for an interaction of 1-year BOADICEA score with menopausal status (model VI, HR = 1.00, P = 1.0), with BMI (model VII, HR = 1.13, P = 0.5), or with the interaction of BMI and menopausal status (model VIII, HR = 0.73, P = 0.4). That is, there was no evidence that the multiplicative interaction between BMI and menopausal status differed by familial risk irrespective of how we modeled the BMI association We also re-analyzed the data by including women with missing menopausal status and putting them in a category of their own. This made no difference to our general findings of no evidence for gene–environment interactions on the multiplicative scale.

Figure 3 shows the overall implications of the study estimates on the predicted age-specific cumulative risk for women with different baseline BMI and familial risk and age 50 years at menopause. In terms of absolute risk, the risk difference for premenopausal women is small when comparing those in the lowest BMI category with those in the highest BMI category. In contrast, the corresponding risk difference for postmenopausal women is much larger and in the opposite direction. The latter difference in absolute risk is even more so for women with a greater familial risk (e.g. for cumulative risk to age 80 years, 8% for women with high familial risk versus 4% for population risk).

Using a large international prospective cohort enriched for women with a family history of breast cancer [28] we have found that the absolute breast cancer risk gradient with BMI increases with age after menopause, and with underlying familial risk. There are three key findings of clinical and biological significance and they are illustrated in Fig. 3.

First, we found that greater BMI at a young adult age is associated with a decreased risk of breast cancer, as have others [44]. We have shown that this negative association with BMI does not translate into a substantial influence on absolute risk of breast cancer.

Second, our modeling confirms that BMI is associated with an increase in risk once a woman becomes postmenopausal. In terms of differences in absolute risk, it is not until a woman is in her mid to late 50s that the risk manifests; the influence on absolute risk then increases with age.

Third, our modeling predicts that the greater a woman’s familial risk, the greater the influence of BMI on her absolute postmenopausal breast cancer risk. We base this on our finding that, in terms of multiplicative risk, the association of breast cancer with BMI did not differ for women at different underlying familial risk (Table 3). Unlike most other cohorts, our enriched cohort has adequate statistical power to examine interactions with underlying familial risk [28, 45, 46]. We also created a continuous measure of familial risk using multi-generational pedigree information and the BOADICEA model to estimate 1-year and lifetime (from birth) risk of breast cancer [35, 36, 43].

As illustrated in Fig. 1, about one third of our cohort has a lifetime risk above the clinically relevant cutoff of 20% [47, 48]. Figure 3 shows that our observed lack of multiplicative interaction means that the difference in absolute risk between women at higher compared with lower BMI is greater for those women who are at higher underlying familial risk. Our finding of a lack of multiplicative interaction between BMI and 1-year BOADICEA score is consistent with the lack of multiplicative interactions with BMI and more than 100 genetic variants found by a large pooled case–control analysis of almost 60,000 women [49].

One potential limitation is that a change in the BMI association with baseline menopausal status could be due to unmeasured confounders. For this to happen, such confounders would have to have a similar menopausal-dependent risk association. As most risk factors related to BMI (e.g., physical activity) do not have clear differences in association by menopausal status, unmeasured confounding is not likely. Other limitations include limited power to address issues specific to mutation carriers and hormone receptor status of tumors. The evidence of whether the protective association of BMI in early life applies solely to estrogen receptor negative disease is less consistent, with contradictory findings from two meta-analyses [50, 51] and a recent study based on a pooled study [52].

Although the negative breast cancer risk association with childhood and adolescent BMI is small in terms of absolute risk, understanding this in the light of the genome-wide association studies results [18, 19], which also support a negative association of BMI with breast cancer risk at a young age, could aid in understanding the role of breast development in breast cancer susceptibility. While increased glucose and other nutrients might alter the ability of BRCA1 to function as a tumor suppressor [53], pre-pubertal estrogen exposure could increase the ability of major breast cancer susceptibility genes to prevent breast cancer through cellular differentiation [54].

An explanation of the negative association with childhood and adolescent BMI might be found in the growth of mammographically dense tissue and changes to the architecture surrounding the mammographically dense tissue, which develops and grows rapidly in adolescence [55]. Importantly, as we show here, although higher BMI is associated with reduced breast cancer risk before menopause, the direction of the association is reversed post-menopause and is of far more consequence in terms of absolute risk.

Laboratory studies have given insights into the mechanisms that might explain why weight gain and metabolically rich environments increase postmenopausal breast cancer risk, with implications for prevention [56]. These mechanisms include conversion of androgens to estrogens in adipose tissue [57], but could also include inflammation and metabolic processes related to cancer risk [58, 59] and changes to epigenetically regulated genes such as BRCA1 [53].

In summary, the negative association with BMI in premenopausal women has a much smaller influence on absolute risk than the positive association with BMI in postmenopausal women. Women at higher familial risk have a much larger difference in absolute risk depending on their BMI than women at lower familial risk.

Our modeling predicted that, for young and premenopausal women, the decrease in breast cancer risk associated with increasing BMI does not have a substantial influence on absolute risk in those periods of life. Our modeling also predicted that the absolute breast cancer risk gradient with BMI increases with age post menopause, and with underlying familial risk. We argue, therefore, that there is no discrepancy between the conclusions of the Mendelian randomization studies [18, 19] and the epidemiological literature (see “Background”). The genetically driven protective role of BMI on breast cancer risk in early adulthood appears to be of little consequence in terms of absolute risk and is overtaken by an environmentally driven deleterious role of greater adult BMI in later life that is even more important for women at increased genetic risk. Given that age-adjusted BMI is correlated between early and mid-adulthood [60], maintaining a healthy weight throughout adult life is of clinical significance for all women, and especially those with a family history of breast cancer.