Introduction

More than 42 % of US women aged 60 years and older are obese (as measured as a body mass index (BMI) >30 kg/m2) [1]. Larger body size is an established risk factor for postmenopausal breast cancer. Most studies examining associations with body size used BMI as a proxy of overall adiposity and found that among women who do not use postmenopausal hormones, obese women had a 1.5- to 2-fold higher postmenopausal breast cancer risk than women in the normal range of BMI [2].

With aging, women lose lean body mass and gain weight as visceral fat [3]. Metabolically active visceral fat releases substantial amounts of insulin-like growth factors (IGF), inflammatory markers, free fatty acids, locally produced estrogen, and adipokines to the liver [4] that might reach the breast through systemic circulation. In epidemiologic studies, people with larger amounts of visceral fat, as measured by larger waist circumferences, have higher risk of hyperinsulinemia and type II diabetes [5], both breast cancer risk factors. Therefore, among postmenopausal women, waist circumference, which is more strongly correlated with visceral fat than BMI [6], might be a better indicator of breast cancer risk [7].

Prior cohort studies have found that women with large waist circumferences have higher risk of postmenopausal breast cancer [8–16]; however, not all studies were restricted to women not taking postmenopausal hormones, which are known to modify relations between measures of body size with postmenopausal breast cancer, and not all studies adjusted for BMI. Adjustment for BMI is potentially informative because an association with waist circumference after adjustment for BMI would suggest that excess abdominal weight increases risk of breast cancer more than an equivalent amount of excess weight located elsewhere on the body [17].

To clarify the association between waist circumference and risk of postmenopausal breast cancer, we examined the association, with and without adjustment for BMI, among postmenopausal women not using menopausal hormones in the Cancer Prevention Study (CPS)-II Nutrition Cohort. We also examined associations by estrogen receptor (ER) status of the breast tumor. This cohort is well suited to examine this association because it includes large numbers of postmenopausal women with information on waist circumference who were not using hormones.

Methods

Description of cohort

Women in this analysis were drawn from the 97,785 female participants in the CPS-II Nutrition Cohort, a prospective study of cancer incidence and mortality established in 1992 as a subgroup of a larger mortality study initiated in 1982 [18]. All participants completed a mailed questionnaire in 1982 and 1992/3 that included information on demographic, medical, and behavioral factors. Follow-up questionnaires were sent to cohort members every 2 years starting in 1997 to update exposure information and ascertain cancer outcomes. Waist circumference was ascertained once as part of the 1997 follow-up survey. With the 1997 survey, study participants were provided with a tape measure and instructed to measure their waist circumference just above the navel to the nearest quarter inch, while standing, and to avoid measuring over bulky clothing. Hip circumference was not collected. BMI was calculated from weight reported on the 1997 survey and height reported on the 1982 survey. The response rate for each of the follow-up questionnaires through 2009 was at least 86 %. Informed consent for participation was assumed based on completion and return of study questionnaires. The Emory University School of Medicine Institutional Review Board approves all aspects of CPS-II.

Cohort for analysis

Of the 87,257 women in the CPS-II Nutrition Cohort who completed the 1997 questionnaire, we excluded from the analytical dataset women who reported a history of cancer prior to enrollment (except non-melanoma skin cancer, n = 14,273), were not postmenopausal or had missing age at menopause (n = 2,081), had missing or outlier data (<39 cm or >139 cm for waist circumference (n = 11,510)), or had missing data for BMI (n = 4,809). In preliminary analyses, we found that the associations of waist circumference and BMI with breast cancer risk did not differ for women who were classified as never and former postmenopausal hormone users in 1997 (p value for interaction = 0.45 and 0.92, respectively); therefore, only women who reported current postmenopausal hormone use in 1997 were excluded (n = 25,475).

Breast cancer cases

In the analysis, we included invasive breast cancer cases (ICD-9 code: 174 or ICD-10/O code: C50) that were diagnosed between the return of the 1997 survey and June 30, 2009. Most incident breast cancer diagnoses were first self-reported on follow-up questionnaires and then were verified through medical records (n = 836) or linkage with state registries (n = 221) [21]. An additional 22 cases, not self-reported, for which breast cancer was listed as an underlying or contributory cause of death on the death certificate, were initially identified through linkage with the National Death Index (NDI). Finally, nine more cases of breast cancer were identified during registry linkage conducted to verify a self-report of, or NDI-ascertained death from, a different type of cancer [19]. A total of 140 women who reported a breast cancer diagnosis that could not be verified were excluded from analyses. Clinical characteristics of the tumor were obtained from state registries or abstracted from medical records.

Statistical methods

Waist circumference was converted to centimeters and categorized into approximate quintiles that were rounded and incorporated the World Health Organization (WHO) cut-points of 80 and 88 cm [20]. BMI was categorized into <25.0, 25.0–29.9, and ≥30.0 kg/m2. Waist circumference and BMI were normally distributed so we also examined continuous versions of the variables (per 10 cm and per kg/m2 unit, respectively). Age-adjusted Spearman’s correlation coefficient between continuous waist circumference and BMI were calculated. Participants contributed person-time to the analysis from the return of the 1997 questionnaire until they were censored at the date of any cancer diagnosis, date of death, date of last survey returned, or the end of follow-up June 30, 2009. Age-adjusted breast cancer incidence rates were calculated for categories of waist circumference and BMI. Cox proportional hazards regression was used to estimate the associations of waist circumference and BMI with breast cancer risk. All Cox models were stratified on single year of age in 1997 by including age in the STRATA statement. Multivariable-adjusted models included known breast cancer risk factors: age and other known or suspected breast cancer risk factors, including race (white, black, other/missing), education (<high school graduate/missing, high school graduate, some college, college graduate), parity and age at first birth (no children, 1–2 births with first birth <25 years of age, 1–2 births with first birth at 25 years or older, three or more births with first birth <25 years of age, three or more births with first birth at 25 years or older), age at menopause (<50, 50–54, ≥55 years, missing/unknown), height (0–159, 160–164, 165–169, ≥170 cm), first-degree family history of breast cancer (0, ≥1 female members, missing), personal history of benign breast disease (yes, no, missing), diabetes (yes, no, missing), physical activity (0–6.9, 7–17.4, 17.5–20.4, and 20.5 + MET/hours, missing), alcohol use (never, former, current drinker <1, 1, and ≥2 drinks per day, missing), smoking status (never, former, current, missing), use of oral contraceptives (never, ever, missing), former use of postmenopausal hormones (never, former), and recent mammogram (yes, no, missing). Mutually adjusted models included all covariates as well as continuous BMI variable in the waist circumference models and continuous waist circumference variable in the BMI models. Tests of trend were examined by assessing every 10 cm for waist circumference and every 1 kg/m2 for BMI. The statistical significance of the interaction between waist circumference and BMI was estimated comparing the −2 log likelihood of models with and without an interaction variable created using continuous variables for waist circumference and BMI. We also examined interaction by age (<60, 60–69, and ≥70 years) and years since menopause (<10, 10–14, 15–19, ≥20 years). We evaluated whether associations differed by ER status of the tumor using a joint Cox proportional hazards model [21]. The proportional hazard assumption was evaluated for associations of BMI and waist circumference with risk; no violations were observed. All analyses were conducted in SAS, version 9.3 (SAS Institute, Inc, Cary, NC).

Results

During a median follow-up time of 11.58 years, 1,088 invasive breast cancer cases were diagnosed among 28,965 women at risk. Former users of menopausal hormones comprised 38.8 % of the study population. Many women (44.6 %) had a waist circumference of 88 cm or larger, whereas only 18.5 % of the women were obese (BMI ≥30 kg/m2). The age-adjusted correlation between waist circumference and BMI was 0.80 (data not shown in tables). Women with smaller waists tended to be younger, more educated, exercise participant, and a current smoker and drinker at baseline, than women with larger waists (Table 1). Women with larger waists were more likely to be 20 or more years since menopause and to report type 2 diabetes mellitus.

Table 1 Distribution of breast cancer risk factors by categories of waist circumference in women from the CPS-II Nutrition Cohort who were never/former HRT users in 1997
Full size table

In multivariable-adjusted models without adjustment for BMI, waist circumference was statistically significantly positively associated with risk of postmenopausal breast cancer; for every 10 cm increase in waist circumference, there was a 13 % higher risk (Table 2). Upon further adjustment for BMI, the association with waist circumference was eliminated. Without adjustment for waist circumference, BMI was statistically significant positively associated with risk [per 1 kg/m2 hazard ratios (HR) = 1.04, 95 % confidence intervals (CI) 1.03–1.05] and controlling for waist circumference did not attenuate the association (per 1 kg/m2 HR = 1.04, 95 % CI 1.02–1.06).

Table 2 Age-adjusted, multivariable-adjusted, and mutually adjusted associations of invasive breast cancer risk with waist circumference and body mass index among never and former postmenopausal hormone users, Cancer Prevention Study-II Nutrition Cohort (1997–2009)
Full size table

There was no evidence of statistical interaction between waist circumference and BMI (p value for interaction = 0.95; Table 3). The association between waist circumference and risk of breast cancer, controlled for BMI, did not differ by age in 1997 or by years since menopause (p value for interaction >0.05; data not otherwise shown).

Table 3 Multivariable-adjustedb HR and 95 % CI of invasive breast cancer with waist circumference by body mass index among never and former users of postmenopausal hormone therapy, Cancer Prevention Study-II Nutrition Cohort (1997–2009)
Full size table

In analyses stratified on ER status, waist circumference, after adjustment for BMI, was not associated with ER+ or ER− breast cancer risk (Table 4). Adjusting for waist circumference, obesity (BMI ≥ 30.0 kg/m2) was associated with higher risk of ER+ breast cancer (HR = 1.41, 95 % CI 1.08–1.85), but not ER− breast cancer risk (HR = 0.61, 95 % CI 0.35–1.08; p value for tumor heterogeneity = 0.02).

Table 4 Age-adjusted, and mutually adjustedb,c associations of ER+ and ER− breast cancer risk with waist circumference and body mass index among never and former postmenopausal hormone users, Cancer Prevention Study-II Nutrition Cohort (1997–2009)
Full size table

Discussion

In this large prospective study of predominantly white, postmenopausal women, we found a statistically significant positive association between waist circumference and postmenopausal breast cancer risk; however, the association was eliminated after adjusting for BMI. The positive association between BMI and risk was statistically significant even after adjusting for waist circumference and was limited to tumors expressing the ER.

The World Cancer Research Fund/American Institute for Cancer Research concluded that there is “probable” evidence that central obesity is associated with risk of postmenopausal breast cancer [22]. However, few cohort studies have investigated whether central obesity contributes to risk of postmenopausal breast cancer beyond its contribution to overall obesity. While nine prospective studies [8–16, 23, 24] presented associations of waist circumference and BMI with risk, only five studies presented results for waist circumference and/or BMI after mutual adjustment [10, 12, 13, 15, 24]. Consistent with the results from the CPS-II Nutrition cohort, larger waist circumference was associated with higher risk of breast cancer but attenuated toward the null in the mutually adjusted models in the Iowa Women’s Health Study (IWHS), the Women’s Health Initiative (WHI), and the European Prospective Investigation into Cancer and Nutrition (EPIC) cohorts [10, 12, 24]. In the Nurses’ Health Study (NHS), the association between waist circumference and postmenopausal breast cancer was unaffected with the inclusion of BMI in the model (Q5 vs. Q1: RR = 1.88–1.83) [13]. Possible reasons for these discrepancies are unclear. BMI was significantly positively associated with postmenopausal breast cancer incidence even after controlling for waist circumference or waist-to-hip ratio in the NHS, the New York University Women’s Health Study, and the WHI [10, 13, 15], as we found in the CPS-II Nutrition cohort. In a subset of the WHI participants with dual energy X-ray absorptiometry (DXA) measurements, similar positive associations with postmenopausal breast cancer risk were reported for whole body fat mass (HR = 1.88) and fat mass of the trunk (HR = 2.05); however, the HRs were not mutually adjusted [25].

A limited number of cohort studies have examined the statistical interaction between waist circumference and BMI. Early results from the IWHS suggested a statistically significant multiplicative interaction between age, BMI, and waist-to-hip ratio [14]. Recently interaction results from larger cohorts, including the NHS [13] and the California Teachers Study [16], were not statistically significant, consistent with the results from the CPS-II Nutrition cohort.

There is consistent evidence that larger BMI is a risk factor only for ER+ breast tumors [11, 16, 24, 26, 27]. Our results further support this evidence for BMI, and we showed that this association persists after controlling for waist circumference. Waist circumference also appears to be associated with ER+ breast cancer in postmenopausal women [11, 16, 24, 25, 27]; however, adjusting for BMI attenuates the associations with risk [24, 27], as we observed in the CPS-II Nutrition cohort.

Both systemic and local biologic mechanisms have been hypothesized to underlie the association between obesity and postmenopausal breast cancer risk. The most widely accepted systemic effects related to breast cancer risk are those due to higher levels of circulating free estradiol resulting from the conversion of androgens to estrogens by aromatase from adipocytes; these effects may work in concert with or independently of insulin resistance leading to hyperinsulinemia and perturbations of the insulin/IGF axis [7]. In the WHI, 23.8 % of excess breast cancer cases attributed to obesity are due to elevated estradiol levels and 65.8 % due to perturbations in the insulin pathways [28]. Low circulating levels of adiponectin might also contribute to the increased risk of breast cancer [29, 30], although in at least one study, the association with adiponectin was substantially attenuated after controlling for estradiol [30].

Because the amount of fat in the breast is proportional to total body adipose tissue mass [31–33], the association between BMI and postmenopausal breast cancer might reflect the local microenvironment of adipocytes in the breast. Obesity is associated with low grade, chronic inflammation that, at the local level, leads to the recruitment of macrophages around necrotic adipocytes, visualized as crown-like structures (CLS) [34]. In human breast tissue, higher proportion of CLS was observed in overweight and obese women with breast cancer than in normal weight patients [35] or women without breast cancer [36]. As a paracrine and autocrine organ, mammary adipose tissue also produces estradiol, adipokines, and factors involved in the insulin/IGF axis [37]. In summary, the association between BMI and postmenopausal breast cancer might be mediated by local and/or systemic mechanisms. Much of the research in the local breast environment has been conducted in premenopausal women; however, the breast tissue of postmenopausal women who undergo only partial age-related lobular involution [38] might experience the same obesity-related mechanisms.

The strengths of this study include the prospective collection of anthropometric data and the large number of cases diagnosed over a long follow-up time. Limitations include the use of anthropometry as indirect measures of body fatness; however, similar positive associations with postmenopausal breast cancer risk were reported for BMI and DXA, a direct measure of body fatness [25]. The self-assessment of waist circumference in the CPS-II Nutrition cohort also might have limited our conclusions. However, in other studies of older women, waist circumference was measured with good validity (Pearson r = 0.87 for BMI, r = 0.85 for waist circumference) [10, 16]; thus, we expect random measurement error would have only modestly attenuated the observed associations. Moreover, we were only able to examine waist circumference at one point in time.

Conclusions

Our study of predominantly white women provides further evidence that waist circumference is associated with risk of postmenopausal breast cancer but not beyond its contribution to overall obesity. Whether these observations are valid across the age range (i.e., premenopausal women) is unclear. Pathways driven by central obesity do not appear to completely explain the biologic mechanisms through which obesity increases risk of breast cancer. To better understand the local influence of obesity on the development of postmenopausal breast cancer, research on the breast microenvironment and improved techniques to non-invasively sample non-malignant breast tissue in postmenopausal women are necessary. Our data support the value of measuring BMI to capture the increased risk of postmenopausal breast cancer associated with larger body size.