Reproductive risk factors in relation to molecular subtypes of breast cancer: results from the Nurses' Health Studies

Abstract

Several intrinsic breast cancer subtypes, possibly representing unique etiologic processes, have been identified by gene expression profiles. Evidence suggests that associations with reproductive risk factors may vary by breast cancer subtype. In the Nurses' Health Studies, we prospectively examined associations of reproductive factors with breast cancer subtypes defined using immunohistochemical staining of tissue microarrays. Multivariate-adjusted Cox proportional hazard models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Over follow-up, we identified 2,063 luminal A, 1,008 luminal B, 209 HER2-enriched, 378 basal-like and 110 unclassified tumors. Many factors appeared associated with luminal A tumors, including ages at menarche (P_{heterogeneity}=0.65) and menopause (P_{heterogeneity}=0.05), and current HT use (P_{heterogeneity}=0.33). Increasing parity was not associated with any subtype (P_{heterogeneity}=0.76), though age at first birth was associated with luminal A tumors only (per 1-year increase HR=1.03 95%CI (1.02, 1.05), P_{heterogeneity}=0.04). Though heterogeneity was not observed, duration of lactation was inversely associated with risk of basal-like tumors only (7+ months vs. never HR=0.65 95%CI (0.49, 0.87), P_trend=0.02), P_{heterogeneity}=0.27). Years between menarche and first birth was strongly positively associated with luminal A and non-luminal subtypes (e.g. 22-year interval vs. nulliparous HR=1.80, 95%CI (1.08, 3.00) for basal-like tumors; P_{heterogeneity}=0.003), and evidence of effect modification by breastfeeding was observed. In summary, many reproductive risk factors for breast cancer appeared most strongly associated with the luminal A subtype. Our results support previous reports that lactation is protective against basal-like tumors, representing a potential modifiable risk factor for this aggressive subtype.

INTRODUCTION

Genomic analysis suggests breast cancer heterogeneity is more complex than previous categorizations based on estrogen and progesterone receptor (ER/PR) status. Currently, at least 4 distinct molecular subtypes have been identified^1–3 and can be reliably classified by immunohistochemistry of ER, PR and other markers.^4–8 Tumors that express hormone receptors (HR⁺) are classified as either luminal A or luminal B, with the latter characterized by higher expression of the proliferative marker Ki-67. HER2-enriched tumors do not express HR, but overexpress human epidermal growth factor receptor 2 (HER2). Among tumors lacking expression of all three markers (triple-negative cancers), basal-like tumors express the basal markers cytokeratins 5/6 (CK 5/6) and/or endothelial growth factor (EGFR), while tumors that express none of these markers are termed unclassified. While these subtypes vary in prognosis and response to treatment^9,10 they also may reflect distinct etiologic pathways given that subtype appears to be fixed from the time of tumor initiation.¹¹ Because luminal A tumors comprise about 60% of breast tumors, many known risk factors are likely to be associated with this phenotype; however, less is known about risk factors for HR-negative (HR⁻) subtypes.

Several reproductive risk factors are associated with breast cancer risk, possibly mediated by changes in sex hormones. Evidence suggests these risk factors may be more strongly associated with HR⁺ subtypes.^12–14 While some reproductive factors are associated with HR⁻ subtypes, these relationships do not always mirror those observed with HR⁺ disease. As recently reviewed,¹⁵ parity has been associated with decreased risk of ER⁺/PR⁺ breast cancers, while null or positive associations between parity and ER⁻/PR⁻ tumors have been reported in some studies. Conversely, while the association between breastfeeding and ER⁺/PR⁺ tumors has been inconsistent, inverse associations have been observed with ER⁻/PR⁻ and triple-negative tumors, suggesting lactation may mitigate the increased risk of these subtypes conferred by parity.^15–17 Few studies have examined reproductive risk factors in relation to the fully characterized intrinsic breast cancer subtypes, including our prior analysis in the Nurses' Health Study (NHS) with 2,022 cases.¹⁸ Results suggest parity may protect against luminal subtypes, while significant^18,19 or suggestive²⁰ positive associations were observed for basal-like tumors. Conversely, breastfeeding was inversely associated with basal-like tumors.^18,19 Associations with other risk factors were less consistent across studies. With the exception of the AMBER study of breast cancer in African American women, which included 678 triple-negative cases, each analysis to date included relatively few non-luminal cases, and immunohistological criteria used to define subtypes differed somewhat across studies.

We examined associations of reproductive risk factors with molecular subtypes of breast cancer defined by comprehensive immunohistochemical profiling in the NHS and NHSII. We added to our previous NHS analysis with an additional 8 years of follow-up, and included NHSII, for a total of nearly 4,000 cases with defined molecular subtypes.

METHODS

Study population

The Nurses' Health Studies are ongoing prospective cohorts of registered nurses in the United States. The NHS was established in 1976 among 121,700 women ages 30–55 years; the NHSII was established in 1989 among 116,430 women ages 25–42 years. Participants have reported reproductive history, lifestyle factors and disease diagnoses at baseline and biennial questionnaires. Including surveillance for deaths, follow-up rates in NHS and NHSII were over 95% through the 2006 and 2003 questionnaire cycles, respectively. This study was approved by the Institutional Review Board of Brigham and Women's Hospital in Boston, MA.

Outcome assessment

Incident breast cancer diagnoses were reported on biennial questionnaires or identified through death records. Permission to access medical records for breast cancer cases was sought from participants or next of kin; requests were made to pathology departments from treating hospitals to obtain formalin-fixed paraffin-embedded (FFPE) tissue samples. Detailed description of block collection has been published previously^21,22. Tissue blocks were obtained from approximately 70% of confirmed NHS cases and approximately 60% of NHSII cases reported through 2006.

Tissue microarrays and subtype classification

Tissue microarrays (TMAs) that included tumors from 4,308 NHS participants and 1,253 NHSII participants were constructed at the Dana Farber/Harvard Cancer Center Tissue Microarray Core Facility in Boston, MA, as previously detailed elsewhere.²² Briefly, three 0.6mm diameter cores were obtained from each participant tumor tissue sample, and inserted into TMA blocks that were subsequently cut into 5um paraffin sections prior to immunohistochemical staining. TMA sections were stained for a panel of immunohistochemical markers (ER, PR, HER2, CK 5/6, EGFR) to classify molecular subtypes.^{4–6,8,23,24} A subset of NHS cases (N=3,281, 76%) was additionally stained for the proliferative marker Ki-67; no NHSII cases had Ki-67 data. Only invasive tumors (N=4,501), as confirmed by study pathologists, were eligible for inclusion in this analysis.

Tissue microarrays and subtype classification

Positivity for each immunohistochemical marker was determined visually by a pathologist (LC) who reviewed each core under a microscope and assigned a score. The exception was Ki-67, which was evaluated automatically using an automated computational image analysis system (Definiens Tissue Studio software, Munich, Germany). Positivity for ER and PR was defined as any nuclear staining of these markers (≥1%), while HER2 over-expression was defined as strong membrane staining in >10% of cells in a given core. Cores were considered positive for CK 5/6 and EGFR if any cytoplasmic and/or membranous staining was present in any of a tumor's three cores. Lastly, we estimated the mean percentage of cells displaying nuclear staining Ki-67 by weighting each core by its total cell count; we considered cases positive if >14% of cells were positive for Ki-67.⁵

We classified tumors by molecular subtype using definitions that correlate with gene expression profile classifications.^4–8,23 For tumors missing Ki-67 expression data, we used histologic grade as a surrogate; NHS results for luminal tumors were similar when restricted to tumors with Ki-67 data. Luminal A tumors were ER-positive and/or PR-positive, HER2-negative, and Ki-67-negative (or histologic grade 1); luminal B tumors were either i) ER-positive and/or PR-positive and HER2-positive or ii) ER-positive and/or PR-positive, HER2-negative, and Ki-67-positive (or histologic grade 2/3); HER2-enriched tumors were ER-negative, PR-negative and HER2-positive; basal-like tumors were ER-negative, PR-negative, HER2-negative and CK 5/6-positive and/or EGFR-positive; unclassified tumors were ER-negative, PR-negative, HER2-negative, CK 5/6-negative and EGFR-negative. Tumors missing complete data on markers necessary to classify them into one of the five subtypes (N=368) were excluded from analysis.

Statistical analyses

Information on reproductive factors and other potential covariates was obtained from baseline and subsequent biennial questionnaires. No strong evidence of between-cohort heterogeneity in associations with reproductive factors was observed; therefore data from NHS and NHSII was pooled for all analyses. Multivariate-adjusted Cox proportional hazards models were used to calculate hazard ratios (HR) and 95% confidence intervals (CI) for each subtype and across all subtypes combined. Women contributed person-time from baseline questionnaire return until the first date of diagnosis of breast or other cancer (excluding non-melanoma skin cancer), death, or June 1, 2006. Models were stratified by cohort, age and calendar time, and updated through follow-up as available. Models included mutual adjustment for reproductive variables of interest, and the following known breast cancer risk factors: body mass index (BMI) at age 18, weight change since age 18, height, personal history of benign breast disease, family history of breast cancer, total physical activity (MET-hr/wk) and alcohol intake. We excluded women who were missing age at menarche, and person-time where parity and age at first birth were unknown. These criteria resulted in the exclusion of an additional 365 cases from the analyses. We included women with missing data on breastfeeding, age at menopause, or menopausal status by coding these exposures as missing; results were similar when restricted to women with complete data on these variables.

In secondary models, we examined the timing of reproductive events, and their interactions. These Cox models used calendar time as the metameter, and included terms for duration of pre- and postmenopause. We included a term for years between menarche and first birth to represent the one-time increase in breast cancer risk following a first pregnancy;²⁵(add Rosner et al., 1994) nulliparous women were assigned a value of zero. To capture the effect of timing and spacing of births on risk, we derived a birth index term, defined as:

$$\mathrm{Birth}\mathrm{index}=\sum _{i=1}^{s_t}(t^{\ast }-t_i)b_{it}$$

where t* = min(current age, age at menopause); b_it= 1 if parity ≥ i at age t, and 0 if nulliparous; and t_i = age at ith birth,²⁶ such that a higher birth index represents a higher number of births occurring at earlier ages. For example, a 50-year old woman with four births occurring at ages 20, 23, 26 and 29 would be assigned a value of 102, while a 50-year old woman with a single birth at age 35 would be assigned a value of 15. These models additionally excluded women with unknown menopausal status and incomplete reproductive histories. We examined whether the associations of menarche to first birth and birth index with risk varied by lactation status.

To evaluate whether associations differed by molecular subtype, we conducted competing risks analyses using the approach described by Lunn and McNeill.²⁷ P-values were two-sided and tests of significance were performed at the α=0.05 level. Analyses were conducted using SAS v. 9.3 (SAS Institute, Cary, NC).

RESULTS

Over follow-up, we identified 2,063 luminal A, 1,008 luminal B, 209 HER2-enriched, 378 basal-like and 110 unclassified tumors (Table 1). The frequency of subtypes appeared to differ somewhat across the 2,935 NHS and 832 NHSII tumors included in our analysis. Compared to tumors in the NHSII, we observed that NHS tumors included a higher proportion of luminal A (55.5% v. 52.3%), HER2-type (6.0% v. 4.1%), and unclassified subtypes (3.4% vs. 1.1%) and a lower proportion of luminal B (25.8% v. 29.9%) and basal-like subtypes (9.3% v. 12.6%). As in our previous study,¹⁸ basal-like tumors were diagnosed at the youngest mean age (54.1y) and were most frequently poorly differentiated (72.3%), while unclassified tumors were most likely to be metastatic at diagnosis (15.2%) (Table 1). Luminal A tumors were most likely to be small (0.1–1.0 cm: 26.9%) and have no nodal involvement at diagnosis (65.9%), while HER2-enriched tumors were most frequently stage III or IV (28.7%).

Table 1.

Tumor characteristics at diagnosis according to breast cancer phenotypes in the Nurses' Health Studies

	Luminal A	Luminal B	HER2-enriched	Basal-like	Unclassified
N (%)	2063 (54.8)	1008 (26.8)	209 (5.6)	378 (10.0)	110 (2.9)
Mean age at diagnosis, years	56.9	58.3	57.7	54.1	54.4
Median age at diagnosis, years	56.9	58.2	57.2	53.3	54.8
Tumor size, N (%)a,c
0.1 to 1.0 cm	535 (26.9)	223 (23.8)	34 (17.0)	47 (13.1)	20 (19.8)
1.1 to less than 2.0 cm	816 (41.1)	402 (41.1)	56 (28.0)	120 (33.4)	36 (35.6)
2.0 to less than 4 cm	449 (22.6)	269 (27.5)	85 (42.5)	165 (46.0)	30 (29.7)
>4+ cm	186 (9.4)	74 (7.6)	25 (12.5)	27 (7.5)	15 (14.9)
Missing	77	30	9	19	9
Nodal involvement, N (%)a,c
No nodes	1276 (65.9)	601 (64.3)	103 (51.8)	218 (61.4)	59 (59.6)
1–3 nodes	398 (20.6)	204 (21.8)	48 (24.1)	90 (25.4)	14 (14.1)
4–9 nodes	153 (7.9)	84 (9.0)	25 (12.6)	32 (9.0)	11 (11.1)
10+ nodes/metastatic	110 (5.7)	46 (4.9)	23 (11.6)	15 (4.2)	15 (15.2)
Missing	126	73	10	23	11
Stage, N (%)a,c
Stage I/II	1572 (83.9)	768 (83.9)	137 (71.4)	289 (85.0)	68 (72.3)
Stage III/IV	302 (16.1)	147 (16.1)	55 (28.7)	51 (15.0)	26 (27.7)
Missing	189	93	17	38	16
Grade, N (%)b,c
Well-differentiated	580 (28.2)	148 (14.8)	4 (2.0)	7 (1.9)	17 (15.5)
Intermediate differentiated	1328 (64.7)	522 (52.4)	96 (47.1)	97 (25.8)	43 (39.1)
Poorly differentiated	146 (7.1)	327 (32.8)	104 (51.0)	272 (72.3)	50 (45.5)
Missing	9	11	5	2	0
Histologic type, N (%)a,c
Invasive ductal	1637 (79.9)	863 (85.6)	203 (97.1)	351 (95.4)	95 (88.8)
Invasive lobular	291 (14.2)	83 (8.3)	2 (1.0)	4 (1.1)	7 (6.5)
Invasive ductal/lobular	109 (5.3)	42 (4.2)	0 (0.0)	2 (0.5)	2 (1.9)
Invasive NOS	13 (0.6)	9 (0.9)	2 (1.0)	11 (3.0)	3 (2.8)
Missing	13	11	2	10	3
Race, N (%)
White	1965 (95.3)	950 (94.3)	203 (97.1)	349 (92.3)	106 (96.4)
Black	23 (1.1)	7 (0.7)	1 (0.5)	13 (3.4)	1 (0.9)
Other	75 (3.6)	51 (5.1)	5 (2.4)	16 (4.2)	3 (2.7)

Percentages calculated among non-missing only, may not sum to 100 due to rounding

^aFrom hospital pathology reports;

^bFrom centralized pathology review;

^cProportions calculated among non-missing

Associations with some reproductive factors varied by breast cancer subtype (Table 2). For example, age at menopause was positively associated with risk of luminal A, luminal B and unclassified tumors, and suggestively associated with HER2-enriched tumors (per 1-year increase HRs=1.04–1.12), though no association with basal-like was observed (HR=1.01, P_{heterogeneity}=0.05). The inverse association with age at menarche was not heterogeneous across subtypes (>14 vs. <12y HRs: 0.65–0.85; P_{heterogeneity}=0.65), though a significant trend was only observed with luminal A (P_trend<0.0001). Risk of all but HER2-enriched tumors increased with current hormone therapy (HT) use, with stronger associations for longer durations of use, though no significant heterogeneity was observed (P_{heterogeneity}=0.33). For the luminal and basal-like subtypes, the greatest increases in risk were observed among current users of estrogen+progestin HT ≥5y, while risk of unclassified tumors was highest among current long-term users of estrogen-only HT.

Table 2.

Multivariate-adjusted¹ HRs (95% CI) for reproductive risk factors in relation to breast cancer subtypes in the Nurses' Health Studies

		Luminal A		Luminal B		HER2-enriched		Basal-like		Unclassified		All Subtypes
N, cases		2063		1008		209		378		110		3768
Age at menarche
<12 years	502	1 (ref.)	246	1 (ref.)	46	1 (ref.)	96	1 (ref.)	30	1 (ref.)	920	1 (ref.)
12 years	608	0.95 (0.84, 1.07)	257	0.84 (0.71, 1.00)	59	0.95 (0.65, 1.41)	93	0.79 (0.59, 1.05)	30	0.76 (0.46, 1.27)	1047	0.90 (0.82, 0.98)
13 years	593	0.80 (0.71, 0.90)	313	0.92 (0.78, 1.09)	60	0.83 (0.56, 1.23)	112	0.86 (0.65, 1.13)	31	0.65 (0.39, 1.09)	1109	0.84 (0.76, 0.91)
14 years	205	0.67 (0.56, 0.79)	126	0.94 (0.75, 1.17)	26	0.80 (0.49, 1.31)	50	0.94 (0.67, 1.34)	10	0.50 (0.24, 1.04)	417	0.77 (0.68,0.86)
> 14 years	155	0.73 (0.61, 0.88)	66	0.73 (0.55, 0.96)	18	0.85 (0.49, 1.49)	27	0.77 (0.50, 1.19)	9	0.65 (0.30, 1.40)	275	0.74 (0.65, 0.85)
P trend		<0.0001		0.13		0.33		0.21		0.06		<0.0001
P heterogeneity		0.65
Parity (ever/never)
Nulliparous	198	1 (ref.)	104	1 (ref.)	9	1 (ref.)	34	1 (ref.)	8	1 (ref.)	353	1 (ref.)
Parous	1865	0.89 (0.77, 1.04)	904	0.84 (0.68, 1.04)	200	2.01 (1.02, 3.97)	344	1.05 (0.73, 1.51)	102	0.96 (0.46, 2.00)	3415	0.92 (0.82, 1.03)
P heterogeneity		0.12
Parity (categorical) 2
1 child	202	1 (ref.)	105	1 (ref.)	23	1 (ref.)	34	1 (ref.)	12	1 (ref.)	376	1 (ref.)
2 children	603	0.97 (0.83, 1.15)	286	0.81 (0.64, 1.02)	62	0.88 (0.53, 1.46)	108	1.06 (0.71, 1.58)	30	0.77 (0.39, 1.56)	1089	0.92 (0.82, 1.04)
3 children	550	1.05 (0.88, 1.25)	263	0.86 (0.68, 1.11)	48	0.78 (0.46, 1.35)	105	1.29 (0.85, 1.96)	31	0.86 (0.42, 1.78)	997	0.99 (0.87, 1.13)
≥4 children	510	0.95 (0.79, 1.15)	250	0.83 (0.64, 1.08)	67	1.03 (0.59, 1.77)	97	1.32 (0.84, 2.05)	29	0.76 (0.36, 1.63)	953	0.95 (0.83, 1.08)
P trend		0.99		0.30		0.65		0.21		0.78		0.92
P heterogeneity		0.76
Age at first birth 2
Per 1-year increase		1.03 (1.02, 1.05)		1.01 (0.99, 1.03)		1.03 (0.99, 1.07)		1.01 (0.98, 1.05)		1.03 (0.97, 1.09)		1.03 (1.02, 1.03)
P heterogeneity		0.04
Breastfeeding 2
Never	607	1 (ref.)	267	1 (ref.)	62	1 (ref.)	121	1 (ref.)	29	1 (ref.)	1086	1 (ref.)
≤6 months or less	504	0.83 (0.73, 0.93)	265	1.00 (0.84, 1.19)	50	0.79 (0.54, 1.15)	88	0.76 (0.58, 1.01)	32	1.09 (0.66, 1.81)	939	0.86 (0.79, 0.94)
≥7 months	581	0.92 (0.81, 1.04)	285	0.97 (0.81, 1.16)	63	1.01 (0.69, 1.47)	97	0.65 (0.49, 0.87)	20	0.81 (0.45, 1.47)	1046	0.90 (0.82, 0.99)
P trend		0.99		0.81		0.10		0.02		0.47		0.58
P heterogeneity		0.27
Age at menopause
Per 1-year increase		1.04 (1.02, 1.06)		1.05 (1.03, 1.07)		1.05 (1.00, 1.10)		1.01 (0.98, 1.05)		1.12 (1.04, 1.21)		1.04 (1.03, 1.05)
P heterogeneity		0.05
Hormone therapy (HT)
Never used HT	1083	1 (ref.)	492	1 (ref.)	116	1 (ref.)	215	1 (ref.)	61	1 (ref.)	1967	1 (ref.)
Past HT use	237	0.90 (0.76, 1.05)	150	1.07 (0.86, 1.34)	36	0.94 (0.61, 1.44)	47	1.20 (0.83, 1.75)	14	1.42 (0.73, 2.75)	484	0.99 (0.88, 1.11)
Current E-only use, <5	98	1.10 (0.89, 1.37)	46	1.06 (0.77, 1.46)	5	0.40 (0.16, 1.00)	16	1.02 (0.59, 1.76)	4	0.84 (0.29, 2.40)	169	1.03 (0.87, 1.22)
Current E-only use, ≥5	182	1.35 (1.12, 1.62)	110	1.37 (1.07, 1.76)	14	0.71 (0.38, 1.32)	30	1.44 (0.92, 2.27)	12	3.16 (1.50, 6.67)	348	1.35 (1.18, 1.55)
Current E+P use, <5 yrs	144	1.67 (1.37, 2.02)	70	1.47 (1.11, 1.96)	11	0.83 (0.42, 1.62)	11	0.83 (0.43. 1.59)	5	1.33 (0.49, 3.63)	241	1.48 (1.27, 1.72)
Current E+P use, ≥5 yrs	96	2.09 (1.66, 2.65)	72	1.85 (1.39, 2.46)	11	1.48 (0.74, 2.95)	13	1.96 (1.05, 3.65)	2	1.93 (0.43, 8.75)	194	1.97 (1.67, 2.32)
Current other use	53	1.34 (1.01, 1.79)	18	0.86 (0.53, 1.40)	6	1.06 (0.45, 2.47)	10	1.76 (0.90, 3.42)	1	0.64 (0.09, 4.73)	88	1.21 (0.97, 1.51)
P heterogeneity		0.33

¹Mutually adjusted for all variables in table+ BMI at 18, weight change since 18, history of BBD, family history of breast cancer, total physical activity, alcohol intake, height, cohort

²Among parous women only

Among parous women, increasing parity was not associated with any subtype (P_{heterogeneity}=0.76), though HRs were non-significantly decreased for luminal B tumors (≥4 vs. 1 child HR=0.83 95%CI (0.64, 1.08), P_trend=0.30), and non-significantly increased for basal-like tumors (HR=1.32 95%CI (0.84, 2.05), P_trend=0.21) (Table 2). Age at first birth was only associated with luminal A tumors (per 1-year increase HR=1.03 95%CI (1.02, 1.05), P_{heterogeneity}=0.04), though HRs were similar, but not significant, for HER2-enriched and unclassified tumors (HRs=1.03). Breastfeeding appeared most strongly inversely associated with basal-like tumors (7+ months vs. never HR=0.65 95%CI (0.49, 0.87), P_trend=0.02), though heterogeneity across subtypes was not significant (P_{heterogeneity}=0.27).

In secondary models, time between menarche and first birth was associated with increased risk of all subtypes except luminal B (P_{heterogeneity}=0.003), and HRs appeared higher for non-luminal subtypes (Table 3). For example, comparing a uniparous woman with 22 years between menarche and first birth to a nulliparous woman, HR (95% CI)=1.80 (1.08, 3.00) for basal-like and 1.40 (1.13, 1.74) for luminal A tumors. Associations with birth index were not heterogeneous across subtypes (P_{heterogeneity}=0.50), though suggestive inverse associations with luminal subtypes were observed. When estimated among parous women only, the association between time between menarche and age at first birth appeared stronger for luminal A tumors (22-year increase HR (95% CI)=1.74 (1.22, 2.48)); however, results for other subtypes were similar. Significant or suggestive heterogeneity was observed for duration of premenopause (P_{heterogeneity} =0.06) and menopause (P_het=0.01), with associations strongest for luminal A (per 1-year premenopause HR (95% CI)=1.10 (1.09, 1.12); per 1-year menopause 1.05 (1.04, 1.06)).

Table 3.

Multivariate-adjusted¹ HRs (95% CI) for birth index and time between menarche and first birth in relation to breast cancer subtypes in the Nurses' Health Studies

	Luminal A	Luminal B	HER2-enriched	Basal-like	Unclassified	All Subtypes
N, cases	1572	771	163	284	78	2868
Time between menarche and first birth
22-year increase2	1.40 (1.13, 1.74)	0.82 (0.60, 1.12)	2.84 (1.38, 5.86)	1.80 (1.08 3.00)	2.57 (1.01, 6.57)	1.32 (1.12, 1.55)
P heterogeneity	0.003
Birth index
102-unit increase3	0.93 (0.77, 1.13)	0.79 (0.60, 1.03)	1.59 (0.86, 2.96)	1.50 (0.95, 2.39)	0.79 (0.32, 1.92)	0.95 (0.83, 1.10)
P heterogeneity	0.50
Duration of premenopause
1-year increase	1.10 (1.09, 1.12)	1.08 (1.06, 1.09)	1.08 (1.04, 1.13)	1.04 (1.01, 1.07)	1.09 (1.03, 1.15)	1.09 (1.08, 1.10)
P heterogeneity	0.06
Duration of menopause
1-year increase	1.05 (1.04, 1.06)	1.02 (1.01, 1.03)	1.04 (1.02, 1.07)	1.02 (0.99, 1.04)	0.97 (0.92, 1.02)	1.03 (1.03, 1.04)
P heterogeneity	0.01

¹Mutually adjusted for all variables in table + breastfeeding, HT use, BMI at 18, weight change since 18, history of BBD, family history of breast cancer, total physical activity, alcohol intake, height, cohort

²22 years: uniparous (age at menarche = 13 years, age first birth = 35 years) vs. nulliparous

³102 units: multiparous (4 births at ages 20, 23, 26 and 29 years) vs. nulliparous; age at menarche 13 years

Lactation modified the effects of time from menarche to first birth on risk of luminal A, HER2-enriched and basal-like tumors (Table 4). A longer time window was associated with these subtypes only among women who had never breastfed (i.e. luminal A tumors, P_difference=0.001 comparing ever vs. never breastfed). Lactation did not modify the effect of birth index on any subtype.

Table 4.

Multivariate-adjusted¹ HRs (95% CI) for birth index and time between menarche and first birth by history of breastfeeding (ever/never) in relation to breast cancer subtypes in the Nurses' Health Studies

	Luminal A	Luminal B	HER2-enriched	Basal	Unclassified	All Subtypes
N, cases	1419	684	143	254	61	2561
Time between menarche and first birth
Never breastfed
22-year increase2	1.66 (1.25, 2.22)	0.93 (0.59, 1.45)	3.51 (1.40, 8.82)	2.49 (1.27, 4.89)	1.83 (0.46, 7.34)	1.56 (1.26, 1.94)
P heterogeneity	0.07
Ever breastfed
22-year increase2	1.09 (0.85, 1.41)	0.80 (0.56, 1.16)	1.66 (0.67, 4.13)	1.22 (0.66, 2.25)	1.76 (0.51, 6.10)	1.04 (0.86, 1.26)
P heterogeneity	0.59
Pdifference (never v. ever breastfed)	0.001	0.46	0.04	0.03	0.97	<0.0001
Birth index
Never breastfed
102-unit increase3	0.83 (0.63, 1.10)	0.74 (0.50, 1.10)	0.96 (0.38, 2.45)	1.67 (0.90, 3.12)	0.61 (0.15, 2.42)	0.85 (0.69, 1.04)
P heterogeneity	0.24
Ever breastfed
102-unit increase3	0.86 (0.69, 1.08)	0.80 (0.59, 1.10)	1.45 (0.70, 2.99)	1.24 (0.71, 2.17)	0.40 (0.11, 1.41)	0.88 (0.75, 1.04)
P heterogeneity	0.35
Pdifference (never v. ever breastfed)	0.22	0.54	0.12	0.82	0.57	0.10

¹Mutually adjusted for all variables in table + HT use, BMI at 18, weight change since 18, history of BBD, family history of breast cancer, total physical activity, alcohol intake, height, cohort

²22 years: uniparous (age at menarche = 13 years, age first birth = 35 years) vs. nulliparous

³102 units: multiparous (4 births at ages 20, 23, 26 and 29 years) vs. nulliparous; age at menarche 13 years

Results restricted to postmenopausal cases were generally similar to overall results. Though we had small case numbers of non-luminal subtypes, we observed a significant inverse trend between parity and premenopausal luminal A tumors (≥4 vs. 1 child HR=0.69, 95% CI (0.48, 0.99), P_trend=0.04 for premenopausal vs. HR=1.02, 95% CI (0.80, 1.29), P_trend=0.47 for postmenopausal). Additionally, the association with birth index appeared stronger for premenopausal luminal A tumors. For example, HRs comparing a woman with four births at ages 20, 23, 26 and 29 years versus a nulliparous woman were 0.64, (0.44, 0.94) for premenopausal cases, compared to 1.06 (0.84, 1.34) for postmenopausal cases. Results were similar when restricted analyses to NHS; small case numbers precluded analysis within NHSII.

Lastly, we examined heterogeneity between the luminal (luminal A and luminal B tumors combined) and basal-like subtypes. For several risk factors, inferences from tests of heterogeneity were similar to those from main analyses (i.e. P_het for age at first birth=0.03; P_het for increasing parity=0.39). However, no heterogeneity between luminal and basal-like tumors was observed for time between menarche and age at first birth (P_het=0.80) or duration of menopause (P_het=0.53); for each exposure, significant heterogeneity was observed between associations with luminal A and luminal B subtypes (P_het=0.001 for time between menarche and age at first birth, P_het=0.005 for duration of menopause). In contrast, significant heterogeneity was observed between luminal and basal-like tumors for associations with age at menopause (P_het=0.004) and duration of premenopause (P_het=0.01).

DISCUSSION

In two large, prospective cohorts, we generally did not observe statistically significant heterogeneity across breast cancer subtypes for most reproductive risk factors, though many appeared most strongly associated with luminal A tumors. In particular, exposures hypothesized to affect lifetime exposure to sex hormones, including ages at menarche and menopause, and current HT use, were associated with this subtype. We also observed significant associations with some non-luminal subtypes, including an inverse association between breastfeeding duration and basal-like tumors. Further, our results suggest the relative timing of reproductive events, specifically time between menarche and first birth, may differentially affect risk of breast cancer subtypes, and that these associations may be modified by breastfeeding.

Classification of breast cancer by molecular subtype has clinical utility, as prognosis and response to treatment differ by subtypes.^9,10 However, whether these subtypes have relevance to targeted prevention efforts is not yet known, as it is unclear whether these phenotypes reflect unique etiologies. Though it is not currently possible to identify the specific cell(s) of origin of breast cancer,²⁸ epidemiologic evidence examining demography and risk factors may provide etiologic insights. In addition to our prior analysis in the NHS,¹⁸ the associations of reproductive risk factors with breast cancer subtypes have previously been investigated in two population-based case-control studies.^19,20 Additionally, as recently reviewed,¹⁵ over 30 studies have examined reproductive risk factors in relation to subtypes defined by ER, PR and/or HER2 status, including a large pooled analysis,¹³ which included a subset of cases with data on basal-like markers. Some evidence of heterogeneity for reproductive factors has been observed, with most studies indicating the protective effect of parity may be limited to HR⁺ subtypes, while breastfeeding is more strongly associated with risk of HR⁻ subtypes.

Circulating levels of sex hormones are associated with breast cancer risk,^29,30 and many, though not all, breast cancer risk factors have been associated with levels of these hormones.³¹ Menarche and menopause mark the onset and cessation of menstrual cycling, respectively, and represent lifetime exposure to ovarian hormones. In a recent meta-analysis,³² breast cancer risk increased by 5% for every year younger at menarche and 2.9% for every year older at menopause; while associations with age at menopause were stronger for HR⁺ tumors, no heterogeneity by HR status was observed for age at menarche. The difference in magnitude of associations with ages at menarche and menopause, in addition to the association between menarche and HR⁻ tumors, suggests that age at menarche may affect risk through mechanisms other than increasing exposure to hormones. Rapid expansion of the breast epithelium during puberty, specifically before first birth, may create a population of rapidly dividing cells that are susceptible to damage from environmental carcinogens.^25,33 We observed that older age at menarche was most strongly inversely associated with luminal A tumors, but non-significant inverse associations were observed for all other subtypes, consistent with other studies reporting associations with basal-like tumors.^19,20 Similar to the meta-analysis, we observed a stronger association with age at menopause for luminal subtypes. Further, we observed that duration of premenopause appeared most strongly associated with luminal A tumors, though tests of heterogeneity were not significant. Interestingly, HT use, which also increases lifetime exposure to hormones, was associated with some HR⁻ subtypes. Though some observational studies, including the NHS,³⁴ have suggested that current HT use is more strongly associated with HR⁺ tumors, results from the randomized arm of the Women's Health Initiative showed increased risks of HR⁻ subtypes with use of estrogen+progestin HT.³⁵ Thus, it appears some hormonally-related factors may be associated with HR⁻, as well as HR⁺, subtypes; however, it is unclear whether this reflects shared etiologic pathways, or distinct mechanisms.

We did not observe significant associations with parity for any subtype, but timing of reproductive events may be relevant, and vary across subtypes. Parity is thought to reduce lifetime breast cancer risk, possibly by inducing mammary cell differentiation.³⁶ However, an increased risk following first birth has been observed, particularly at older ages,^37,38 and the net effect of pregnancy may be detrimental when first birth occurs after age 35.^26,39,40 To examine the dual effects of pregnancy, we examined time from menarche to first birth to capture the initial increase in risk, and a birth index term representing the protective effect of subsequent pregnancies. Time between menarche and first birth appeared more strongly associated with non-luminal subtypes (P_{heterogeneity}=0.003), though other studies have observed stronger associations with this interval for HR⁺ tumors,^41–44 and one reported an inverse association with triple-negative tumors.⁴⁵ Although no significant heterogeneity was observed, birth index appeared somewhat more protective against luminal subtypes. In previous analyses in the NHS, birth index was inversely associated with ER⁺, but not ER⁻, tumors,^46,47 though no association between spacing of births and risk of any subtype was observed in a case-control study.¹⁹ Taken together, our findings suggest a one-time increase in risk of non-luminal subtypes following first birth, particularly those occurring at late ages, that is not offset by subsequent births, though more work is needed to confirm these associations.

Though tests of heterogeneity were not significant, our finding that breastfeeding was most strongly protective against basal-like tumors is consistent with previous studies,^18–20 including several in which the triple-negative phenotype was examined.^{15–17,45,48} Recently, among parous African-American women,¹⁶ the risks of ER⁻ and triple-negative cancers were approximately 20% lower comparing ever vs. never breastfed, and breastfeeding appeared to reduce the increased risk of ER⁻ tumors associated with increasing parity. Similarly, we observed that the increased risk of basal-like tumors associated with older age at first birth was limited to women who had not breastfed. Breastfeeding also appeared to mitigate the one-time increase in risk of luminal A tumors following a late first pregnancy. Recently, Ambrosone et al.⁴⁴ also noted that the risk of both ER+ and ER- breast cancer associated with a long interval between menarche and age at first birth was lower among women who had breastfed compared to those who had not. Hypothesized mechanisms through which breastfeeding may affect breast cancer risk are varied, and include: long-term changes in hormone levels; excretion of estrogens and other carcinogens; promotion of mammary tissue differentiation; and reduced number of lifetime menstrual cycles through delayed resumption of menstruation.⁴⁹ Therefore, lactation may affect risk of HR⁻ subtypes through non-hormonal mechanisms, though it is not clear why the effect may be more pronounced relative to HR⁺ subtypes.

Although our study benefits from a large sample size, comprehensive immunohistochemical data and detailed, prospectively collected data on many reproductive events and covariates, several limitations must be acknowledged. We were unable to obtain tissue samples on a large proportion of reported breast cancer cases. Given the destruction of eligible blocks after a fixed amount of time, per hospital policy, we tended to receive blocks from cases diagnosed more recently, and therefore at older ages. However, after accounting for these factors, distributions of reproductive factors were similar among NHS participants with vs. without tumor blocks.²² Among NHSII women eligible for inclusion in the TMA, similar age-adjusted distributions of reproductive factors were also observed for women with FFPE tissue compared to those without tissue blocks. Although our study comprises a large number of tumors, the relatively low proportion of non-luminal tumors may have limited our ability to detect significant heterogeneity across subtypes. Although Ki-67 was available on a subset of participants, histologic grade is highly correlated with Ki-67 expression,⁵⁰ and results were similar when restricted to cases with Ki-67.

In summary, we observed the associations of some reproductive factors with breast cancer risk may vary across intrinsic subtypes. Many factors were associated with luminal A tumors, which was unsurprising, given that this subtype comprises the majority of breast cancers, and may be driving associations between identified risk factors and overall breast cancer risk. Our results also add mounting evidence suggesting that breastfeeding may reduce risk of basal-like tumors. Given the relatively poor prognosis and limited treatment options for this subtype, further work is warranted to translate these findings into preventive strategies.

Novelty and impact.

Fewer risk factors have been identified for hormone receptor-negative breast cancer subtypes than for hormone receptor-positive disease. We prospectively examined associations between reproductive factors and breast cancer subtypes defined using comprehensive immunohistochemical profiling. Most factors were most strongly associated with luminal A tumors; however, an inverse association between breastfeeding and the basal-like phenotype supports previous findings that lactation represents a modifiable risk factor for this subtype, which is associated with poorer clinical outcomes.

Abbreviations

ER

Estrogen receptor

PR

progesterone receptor

HER2

human epidermal growth factor receptor 2

CK5/6

cytokeratins 5/6

EGFR

endothelial growth factor

NHSI, NHSII

Nurses' Health Studies

HR

hormone receptor

FFPE

formalin-fixed paraffin-embedded

TMA

tissue microarray

HR

hazard ratio

CI

confidence intervals

HT

hormone therapy