Breast cancer biologic and etiologic heterogeneity by young age and menopausal status in the Carolina Breast Cancer Study: a case-control study

The CBCS is a population-based case-control study conducted in 24 (phases I and II) and 44 (phase III) counties of central and eastern North Carolina. Women were eligible for inclusion if they were 20–74 years of age at the time of diagnosis (cases) or study recruitment (control subjects). As described previously [17, 21‐23], the CBCS researchers collected extensive clinical, molecular, and epidemiologic data for in situ (phase II) and invasive (phases I–III) breast cancer cases diagnosed from 1993 to 1996 (phase I), 1996 to 2001 (phase II), and 2008 to 2013 (phase III). Cases were identified through rapid case ascertainment by the North Carolina Central Cancer Registry; control subjects were recruited (for phases I and II only) through North Carolina Department of Motor Vehicles records and Health Care Financing Administration records for Medicare enrollment. Phase III was conducted in 44 counties and recruited only patients with invasive breast cancer. In all phases, randomized recruitment was used to oversample younger (<50 years) and African American cases as well as to frequency-match control subjects to cases by age (<50 years vs. ≥50 years), 5-year age group, and self-reported race (African American vs. non-African American) [21]. The response rates for cases and control subjects in phases I and II have been reported previously [24]. For phase III cases, the overall contact rate (contacted/eligible) was 95.5 %, the cooperation rate (enrolled/contacted) was 75.8 %, and the overall response rate (product of contact and cooperation rates) was 72.0 %. The present analysis included all cases (n = 5309) from phases I–III for case-case analyses as well as phases I and II cases (n = 2311) and control subjects (n = 2022) for case-control analyses. All study protocols were approved by the University of North Carolina (UNC) School of Medicine Institutional Review Board, and all participants gave their written informed consent.

Breast tumor tissue and medical records were obtained from local hospitals for all in situ and invasive breast cancer cases. Stage at diagnosis (in situ, stages I–IV), lymph node status (positive vs. negative for metastasis), estimated tumor size (>2 cm, ≤2 cm), nuclear grade (marked, slight/moderate pleomorphism), and histologic grade (poorly, moderately/well-differentiated) were obtained from medical record reviews, although nuclear and histologic grade data were unavailable for all phase II cases. Immunohistochemistry (IHC) assays conducted at the UNC Immunohistochemistry Core Laboratory were completed to define human epidermal growth factor receptor 2 (HER2) status (positive defined as ≥10 % cells staining vs. negative) [24]. Estrogen receptor (ER) and progesterone receptor (PR) status information was obtained from medical records when available (88 %) or by IHC at UNC (12 %), with ER/PR positivity defined as ≥5–10 % cells with nuclear staining [25]. A 10 % random sample of cases showed high agreement between ER status obtained from medical records and IHC assays at UNC (k-statistic = 0.62, concordance = 81 %) [25]. Breast cancer subtype was defined on the basis of ER, PR, HER1, HER2, and cytokeratin (CK) 5/6 positivity as previously described [25]: basal-like (ER⁻, PR⁻, HER2⁻, HER1⁺, and/or CK 5/6⁺), luminal (ER⁺ and/or PR⁺), HER2⁺/ER⁻ (ER⁻, PR⁻, HER2⁺), and unclassified (negative for all five markers). Collection of molecular subtype data is ongoing for phase III cases; only phases I and II cases were included in analyses of these characteristics.

In-home interviews were conducted by study nurses for all cases and control subjects [17]. Participants were asked questions regarding their reproductive and medical histories as well as exogenous hormone exposures. Nurses also measured body weight, height, waist circumference, and hip circumference during the interview. All breast cancer risk factors were categorized as previously reported [17]. Briefly, menopausal status was defined as pre- or postmenopausal on the basis of self-reported cessation of menstruation for women over age 50 years; women under age 50 years were defined as premenopausal unless they had reported undergoing menopause-related cessation of menstruation, bilateral oophorectomy, or ovarian irradiation. Body mass index (BMI) was defined as body weight/height ratio in kilograms per square meter using categories of the National Heart, Lung, and Blood Institute (normal/underweight <25 kg/m², overweight 25.0–29.9 kg/m², and obese ≥30 kg/m²) [26]. Waist-to-hip ratio (WHR) was calculated as the waist/hip circumference ratio in centimeters and categorized as <0.77 cm, 0.77–0.83 cm, and ≥0.84 cm. Classifications for other risk factors were consistent with previous CBCS analyses [17]: age at menarche (<13 years, ≥13 years), parity (nulliparous, 1–2 births, ≥3 births), age at first live birth (<26 years, ≥26 years), history of breastfeeding (never, ever), lifetime duration of breastfeeding (never, >0–3 months, ≥4 months), and oral contraceptive use (never, ever).

Case-control analyses were conducted for breast cancer risk factors, stratifying by age at diagnosis or menopausal status. Unconditional logistic regression, with an offset term to account for the sampling probabilities of cases and control subjects, was used to estimate case-control ORs and 95 % CIs for each risk factor. Young (<40 years) vs. older (≥40 years) groups were evaluated among premenopausal women and pre- vs. postmenopausal status was evaluated among women ≥40 years. All analyses were adjusted for age (5-year age categories) and race (African American vs. non-African American). Similar to the case-case analyses described above, we restricted our analyses of risk factor associations by menopausal status to pre- and postmenopausal women aged 40–59 years to address residual confounding by age and to control for nonpositivity (i.e., the lack of premenopausal women above age 59 years). Heterogeneity in risk factor associations by age or menopausal status was evaluated by conducting likelihood ratio tests in which the estimated log-likelihood of adjusted models was compared with that of the adjusted model including a multiplicative interaction term for age (or menopausal status) and the corresponding risk factor. Statistically significant heterogeneity was defined with α = 0.1. To explore racial differences in etiologic heterogeneity by age and menopausal status, case-control analyses were further stratified by race. Models for age at first live birth, history of breastfeeding, and lifetime breastfeeding duration were restricted to parous women. Statistical significance was defined with α = 0.05. All analyses were performed using SAS version 9.3 software (SAS Institute, Cary, NC, USA).

Tumor characteristics for in situ and invasive cases were examined together in this study, as sensitivity analyses did not reveal any difference in results when excluding in situ cases (results not shown) nor any difference in the prevalence of in situ cases by age (Table 2). In logistic regression analyses of the relationship between young age at diagnosis and tumor characteristics, young women had more aggressive tumors (Table 2); specifically, young women’s tumors were more likely to be higher stage and basal-like breast cancers. Young women also had significantly more ER and PR negative disease with a greater frequency of marked pleomorphism, positive lymph nodes, and larger tumor size. Although not significantly different, HER2 positivity and more poorly differentiated histologic grade were more common among young women.

To evaluate whether young- and older-onset breast cancers are etiologically distinct, we estimated associations with risk factors stratified by age (Table 3). These analyses were restricted to premenopausal women to avoid confounding by menopausal status. Age appeared to modify associations for several risk factors. Ever-breastfeeding and a lifetime breastfeeding duration ≥4 months were associated with reduced risk of young-onset (<40 years of age) breast cancer and no change in risk of older-onset (≥40 years) disease. Likelihood ratio tests showed significant effect measure modification by age for both ever-breastfeeding (interaction p = 0.003) and duration of breastfeeding (interaction p = 0.04). Parity and longer time since last term pregnancy (10–19 years) were protective among older women and had either null or weakly increased risk for young women, although this heterogeneity was not statistically significant. High WHR was protective among younger cases and either null (BMI) or associated with elevated risk (WHR) among older women. Oral contraceptive use was more strongly, but not significantly, associated with risk among younger women, while associations with BMI, age at first birth, age at last birth, and age at menarche were similar between young and older premenopausal cases. Thus, patterns of risk factor associations differed substantially between young and older premenopausal women.

Relative to the age patterns described above, associations with menopausal status were attenuated. After restricting to cases ≥40 years of age, we assessed the association between menopausal status and tumor characteristics. In crude analyses, premenopausal status appeared to be associated with poor-prognosis tumor characteristics similar to those observed when we compared young and older premenopausal women. Premenopausal cases were significantly less likely to have in situ disease and ER positivity and more likely than postmenopausal women to have stage II, III, or IV disease; higher histologic and nuclear grade; lymph node positivity; and greater tumor size (Additional file 2: Table S2). However, after adjusting for age and/or restricting the age range of women to control for nonpositivity (i.e., the lack of premenopausal women above age 59 years), we observed that few changes persisted. Among women aged 40–49 years, only PR status and nuclear grade showed differences by menopausal status; premenopausal women had higher PR positivity and less marked pleomorphism. Among women aged 40–59 years, premenopausal women had a greater likelihood of stage II and PR⁺ disease as well as larger tumors than postmenopausal women (Additional file 2: Table S2); however, strata for premenopausal women over age 50 years were very sparse, leading to some instability of estimates. These results suggest that, while menopausal status may be associated with some tumor characteristics, associations are weaker than those by age.

Similarly, stratification on menopausal status showed limited evidence of etiologic heterogeneity (Additional file 3: Table S3). Only age at last birth showed significant modification by menopausal status (p = 0.01), in that age at last birth ≥30 years was not associated with increased risk among premenopausal women but did increase risk among postmenopausal women. No other risk factors were differentially associated with pre- or postmenopausal breast cancer after adjusting for age and race.

In race-stratified analyses, associations between young age and tumor characteristics were more precise among white women but were similar in direction (results not shown). Only HER2 positivity, lymph node positivity, and histologic grade showed a suggestion of heterogeneity by race: Young age was associated with increased HER2 positivity (OR 1.72, 95 % CI 1.08–2.72) and increased lymph node positivity (OR 1.35, 95 % CI 1.04–1.76) among premenopausal white women but not African American women (HER2 OR 1.03, 95 % CI 0.56–1.92; lymph node OR 1.07, 95 % CI 0.82–1.40). Additionally, among white women, younger cases had a higher prevalence of poorly differentiated histologic grade (OR 1.63, 95 % CI 1.13–2.36); this association was qualitatively different among African American women (OR 0.86, 95 % CI 0.57–1.30). No strong heterogeneity by race was observed in the associations between tumor characteristics and menopausal status after controlling for age at diagnosis (results not shown).

Likewise, little heterogeneity was observed by race for associations between breast cancer risk factors and menopausal status. With regard to breast cancer risk factors, higher WHR increased risk among African American women regardless of age, while age appeared to modify the association between WHR and risk in white women (results not shown). History of breastfeeding and longer breastfeeding duration strongly reduced risk among both young and older African American women (history of breastfeeding: young OR 0.50, 95 % CI 0.25–1.00; older OR 0.59, 95 % CI 0.38–0.92; breastfeeding duration: young OR 0.47, 95 % CI 0.20–1.11; older OR 0.56, 95 % CI 0.32–0.98), while the protective effect of breastfeeding in white women appeared to be restricted to young women. No other risk factors were associated with breast cancer risk in race- and age-stratified analyses.