Fruit and Vegetable Intake and Risk of Breast Cancer by Hormone Receptor Status

Abstract

Background

Estrogen receptor–negative (ER⁻) breast cancer has few known or modifiable risk factors. Because ER⁻ tumors account for only 15% to 20% of breast cancers, large pooled analyses are necessary to evaluate precisely the suspected inverse association between fruit and vegetable intake and risk of ER⁻ breast cancer.

Methods

Among 993 466 women followed for 11 to 20 years in 20 cohort studies, we documented 19 869 estrogen receptor positive (ER⁺) and 4821 ER⁻ breast cancers. We calculated study-specific multivariable relative risks (RRs) and 95% confidence intervals (CIs) using Cox proportional hazards regression analyses and then combined them using a random-effects model. All statistical tests were two-sided.

Results

Total fruit and vegetable intake was statistically significantly inversely associated with risk of ER⁻ breast cancer but not with risk of breast cancer overall or of ER⁺ tumors. The inverse association for ER⁻ tumors was observed primarily for vegetable consumption. The pooled relative risks comparing the highest vs lowest quintile of total vegetable consumption were 0.82 (95% CI = 0.74 to 0.90) for ER⁻ breast cancer and 1.04 (95% CI = 0.97 to 1.11) for ER⁺ breast cancer (P _{common-effects} by ER status < .001). Total fruit consumption was non-statistically significantly associated with risk of ER⁻ breast cancer (pooled multivariable RR comparing the highest vs lowest quintile = 0.94, 95% CI = 0.85 to 1.04).

Conclusions

We observed no association between total fruit and vegetable intake and risk of overall breast cancer. However, vegetable consumption was inversely associated with risk of ER⁻ breast cancer in our large pooled analyses.

Breast cancer is a heterogeneous disease. Expression of the estrogen receptor (ER) can be used to explain, in part, some of the differences in etiology, and clinical characteristics of breast cancer and survival rates among breast cancer patients (1–3). Breast tumors that express the ER (ER⁺ tumors) are more strongly associated with hormone-related factors than tumors that do not express the ER (ER⁻ tumors). Classic risk factors, such as late age at first birth and number of births, are more consistently associated with risk of ER⁺ breast cancer than with risk of ER⁻ breast cancer (1). In addition, few risk factors have been identified for ER⁻ breast cancer.

Fruit and vegetable intake has been hypothesized to reduce the risk of breast cancer, but the current evidence is inconclusive. A recent meta-analysis of 14 cohort studies reported a statistically significant 11% reduced risk of breast cancer overall comparing high vs low fruit and vegetable intake; results were not reported for breast cancer subtypes defined by receptor status (4). Three cohort studies (5–7) have recently investigated the association between fruit and vegetable intake and risk of breast cancer by receptor status. All three studies found that women who consumed higher levels of fruits and vegetables had a 32% to 50% lower risk of ER⁻ breast cancer compared with women who consumed low levels of fruits and vegetables. In another cohort study that evaluated only dietary patterns, women consuming a high fruit or salad pattern were observed to have a 45% lower risk of ER⁻ breast cancer (8). Because the number of studies examining ER⁻ breast cancer is limited, more evidence is needed to examine the role of diet in ER⁻ breast cancer.

One challenge in examining ER⁻ breast cancer in epidemiologic studies is that ER⁻ breast cancer accounts for only 15% to 20% of breast cancers (9). Thus, in most studies, the number of ER⁻ breast cancer case patients is relatively low, which affects the power to detect modest associations with certainty.

Therefore, with the advantage of a large number of case patients, we evaluated the relation between fruit and vegetable consumption and risk of breast cancer by ER status in a pooled analysis of 20 prospective cohorts. These analyses expand our prior analyses of fruit and vegetable consumption and breast cancer risk (10) by including 12 more prospective studies and increasing the follow-up time for most of the studies in the original analyses. In addition, only two of the 20 studies in this pooled analysis have previously examined associations specifically with fruit and vegetable consumption and risk of ER⁻ breast cancer (5, 6). In secondary analyses, we examined tumors classified by progesterone receptor (PR) status.

Methods

Study Population

The Pooling Project of Prospective Studies of Diet and Cancer (Pooling Project) is a long-standing international consortium of prospective cohort studies (11). The analyses conducted here included 20 studies (Table 1) that met all of the following criteria: 1) at least one publication on any diet and cancer association; 2) comprehensive assessment of usual dietary intake; 3) validation of the dietary assessment method or a closely related instrument; and 4) at least 25 incident breast cancer case patients of the specific hormone receptor subtype being evaluated in that analysis. Each study included in our analyses was reviewed and approved by the institutional review boards of the institution where the study was conducted.

Table 1.

Characteristics of the cohort studies included in the pooled analyses of fruit and vegetable intake and breast cancer risk by hormone receptor status*

Study (country)	Years of Follow-up	Baseline cohort size†	Age range, y	No. of case patients‡							Total Fruits, g/day		Total Vegetables, g/day
				Total	ER+	ER−	ER missing, %	PR+	PR−	PR missing, %	No. of questions	Median (10th–90th)	No. of questions	Median (10th–90th)
Beta-Carotene and Retinol Efficacy Trial (United States)	1985–2005	6000	49–71	367	193	31	39	163	48	43	15	208 (46–500)	14	157 (68–325)
Black Women’s Health Study (United States)	1995–2008	51 890	21–69	670	416	254	— §	326	331	— §	7	182 (34–511)	9	68 (20–196)
Breast Cancer Detection Demonstration Project Follow-up Study (United States)	1987–1999	42 061	40–93	1305	793	166	27	667	270	28	5	173 (33–388)	8	135 (51–288)
California Teachers Study (United States)	1995–2003	100 064	22–104	2696	1930	343	16	1544	625	20	11	180 (52–380)	13	150 (61–305)
Canadian National Breast Screening Study (Canada)	1980–2000	48 900	40–59	1240	367	125	60	309	140	64	6	291 (97–537)	10	220 (102–432)
Cancer Prevention Study II Nutrition Cohort (United States)	1992–2003	74 138	40–87	2999	1835	323	28	1483	561	32	7	195 (52–396)	8	147 (61–302)
CLUE II: Campaign Against Cancer and Heart Disease (United States)	1989–2007	8279	18–93	288	198	50	14	168	78	14	7	152 (25–390)	8	116 (40–255)
Iowa Women’s Health Study (United States)	1986–2004	34 580	52–71	1849	1329	238	15	1117	388	19	15	338 (130–625)	25	195 (91–383)
Japan Public Health Center-Based Prospective Study I (Japan)	1990–2004	21 609	40–59	289	111	69	38	87	82	42	4	118 (32–193)	3	157 (66–233)
Melbourne Collaborative Cohort Study (Australia)	1990–2006	22 456	31–75	799	493	171	17	420	240	17	19	392 (136–835)	13	217 (100–392)
National Institutes of Health–AARP Diet and Health Study (United States)	1995–2003	200 049	50–71	5972	2322	464	53	1917	786	55	13	289 (77–696)	13	162 (59–389)
Netherlands Cohort Study (Netherlands)	1986–1999	62 573	54–70	2013	700	183	56	361	199	72	12	207 (80–393)	16	162 (86–287)
New York University Women’s Health Study (United States)	1986–2003	13 257	31–70	919	392	121	44	296	204	46	11	290 (94–595)	11	200 (75–424)
Nurses’ Health Study (a) (United States)	1980–1986	88 605	34–67	1122	528	255	30	389	304	38	6	272 (73–560)	7	150 (68–292)
Nurses’ Health Study (b) (United States) ||	1986–2006	68 337	40–67	4462	3075	757	14	2475	1276	16	21	329 (115–642)	27	259 (129–470)
Nurses’ Health Study II (United States)	1991–2003	93 765	26–46	1331	846	303	14	765	369	15	15	223 (68–508)	25	206 (95–399)
Prospective Study on Hormones, Diet and Breast Cancer (Italy)	1987–2002	9 044	34–70	283	206	67	4	180	92	4	6	330 (174–541)	23	190 (94–348)
Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (United States)	1993–2007	28 276	52–74	1090	858	137	9	758	227	10	16	283 (97–616)	16	252 (120–493)
Swedish Mammography Cohort (Sweden)	1987–2005	60 464	38–76	2585	1603	381	24	1305	671	24	4	167 (46–374)	5	77 (29–158)
Women’s Health Study (United States)	1993–2004	38 349	38–89	1175	937	187	5	819	288	6	15	266 (86–539)	25	236 (112–452)
Women’s Lifestyle and Health Study (Sweden)	1991–2006	47 514	30–50	1072	737	196	13	613	309	14	5	136 (37–297)	8	61 (22–125)

* ER = estrogen receptor; PR = progesterone receptor

† Cohort size after applying study-specific exclusion criteria and then excluding women with log_e-transformed energy intake values beyond three standard deviations from the study-specific mean and previous cancer diagnosis (other than nonmelanoma skin cancer). The Netherlands Cohort Study was analyzed as a case–cohort study, so its baseline cohort size does not reflect the above exclusions. Total cohort size was 993 466.

‡ Total number of patients was 34 526 for total breast cancer, 19 869 for ER⁺, 4821 for ER⁻, 16 162 for PR⁺, and 7488 for PR⁻ breast cancer.

§ All case patients in these analyses from this study had information on estrogen and progesterone receptor status.

|| Nurses’ Health Study (b) was not included as part of total cohort size because the participants in this study were included in Nurses’ Health Study (a).

Assessment of Dietary and Nondietary Factors

Dietary intake was assessed by a validated food frequency questionnaire (FFQ) at baseline in each study. Each FFQ listed three to 15 fruit and six to 30 vegetable items. Due to the varying frequency responses and portion sizes in the diet assessment methods across studies, fruit and vegetable intake was expressed as grams consumed per day (10).

We analyzed intakes of total fruit and vegetables, total fruit, and total vegetables. To examine the potential beneficial or adverse effect of particular bioactive compounds (12), we also analyzed botanically defined fruit and vegetable subgroups (13) and specific fruits and vegetables that were assessed as separate items in more than half the cohorts. Because of their high protein or starch content, mature beans and potatoes were excluded from the total vegetable group (14) but were included in their relevant botanical group. Pickled fruit and vegetables were also excluded from the fruit and vegetable groups because they contain potentially carcinogenic nitrates and preservatives (15). In each study, nutrient intake estimates from the FFQ used in that study or a closely related instrument were compared with multiple 24-hour dietary recalls or dietary records. However, only five studies compared intakes between the FFQ and comparison methods for intakes of total fruit and vegetables (16), total vegetables (17–19), total fruit (17–19), or individual fruits or vegetables (20); correlations that compared intake estimates from the FFQs and comparison methods for these fruit and vegetable groups or items exceeded 0.35. Validity correlations for folate, vitamin C, and carotene intakes, which are closely related to fruit and vegetable intake (21), ranged 0.3 to 0.65 (11,17,20,22–27).

At baseline, each study collected information on age, height, and body weight. Most studies also measured reproductive factors, education, physical activity, smoking history, and family history of breast cancer.

Case Ascertainment

Incident invasive breast cancer case patients in each study were identified by follow-up questionnaires and subsequent review of medical records (28–30), linkage to cancer registries (31–41), or both (42–45). Some studies also used linkage to mortality registries to identify case patients (16,28,30,33,34,39,43,45,46). We considered case patients with borderline hormone receptor status as positive for that hormone receptor.

Statistical Analysis

After applying the exclusion criteria used by each study, we further excluded participants with a prior cancer diagnosis (except nonmelanoma skin cancer) at baseline, with energy intakes beyond 3 standard deviations from the study-specific log_e-transformed mean energy intake, and with missing data for total fruit or total vegetable intake.

Associations for total, ER⁻, ER⁺, PR⁻, and PR⁺ breast cancer were evaluated separately in each study by Cox proportional hazards regression analyses (47) using SAS PROC PHREG (48). The Netherlands Cohort Study was analyzed as a case–cohort study (49). The Nurses’ Health Study was analyzed as two separate cohorts (1980–1986 Nurses’ Health Study[a]; 1986–2006 Nurses’ Health Study[b]) to take advantage of the more detailed dietary assessment in 1986. Because blocks of person-time in different time periods are asymptotically uncorrelated according to survival data theory, pooling of estimates from these two time periods produces valid estimates (50). For all cohorts, person-years of follow-up were calculated from the date the baseline questionnaire was returned or completed to the date of diagnosis of incident breast cancer, death, loss to follow-up (if applicable), or end of follow-up, whichever came first. Breast cancer case patients without hormone receptor status information or with a different subtype than the one being evaluated were censored at their date of diagnosis. Age at baseline and year of baseline questionnaire return were used as stratification variables so we could account for age, calendar time, and time since entry into the study. In multivariable analyses, for studies with more than 200 case patients of the outcome evaluated, we included the confounding variables (see Table 2 for a list of the confounding variables and their categorizations) directly in the model. Otherwise, we used propensity scores to adjust for confounding (51–53). Missing indicator variables were created for missing responses for each measured confounding variable in a study, if applicable.

Table 2.

Pooled multivariable relative risks* (95% confidence interval) of breast cancer by estrogen receptor/progesterone receptor subtype for quintiles of fruit and vegetable intake

	Quintile of intake					P trend†	P between-studies heterogeneity for quintile 5‡	P common effects by receptor status for quintile 5§
	Q1	Q2	Q3	Q4	Q5
Total breast cancer
Total fruit and vegetables||
No. of case patients	6643	6915	6955	7052	6981
RR (95% CI)	1.00 (referent)	1.00 (0.96 to 1.04)	0.99 (0.94 to 1.03)	0.99 (0.94 to 1.04)	0.98 (0.93 to 1.02)	.30	.16
Total vegetables
No. of case patients	6555	6843	6943	7108	7086
RR (95% CI)	1.00 (referent)	1.00 (0.96 to 1.03)	1.01 (0.96 to 1.05)	1.00 (0.95 to 1.05)	0.99 (0.95 to 1.04)	.77	.21
Total fruits
No. of case patients	6688	6887	7143	6909	6910
RR (95% CI)	1.00 (referent)	1.00 (0.96 to 1.05)	1.01 (0.98 to 1.05)	0.98 (0.94 to 1.03)	0.99 (0.95 to 1.03)	.36	.20
By ER status
Total fruit and vegetables
ER− No. of case patients	961	965	987	950	958
RR (95% CI)	1.00 (referent)	0.95 (0.87 to 1.04)	0.96 (0.87 to 1.06)	0.90 (0.81 to 1.00)	0.90 (0.81 to 1.01)	.03	.25	.12
ER+ No. of case patients	3701	3996	3978	4142	4052
RR (95% CI)	1.00 (referent)	1.02 (0.97 to 1.08)	0.99 (0.94 to 1.05)	1.02 (0.97 to 1.07)	1.00 (0.94 to 1.07)	.91	.06
Total vegetables
ER− No. of case patients	995	968	965	975	917
RR (95% CI)	1.00 (referent)	0.92 (0.84 to 1.01)	0.89 (0.82 to 0.98)	0.90 (0.80 to 1.01)	0.82 (0.74 to 0.90)	<.001	.50	<.001
ER+ No. of case patients	3662	3852	4026	4103	4217
RR (95% CI)	1.00 (referent)	0.99 (0.94 to 1.05)	1.03 (0.97 to 1.09)	1.02 (0.95 to 1.08)	1.04 (0.97 to 1.11)	.06	.04
Total fruits
ER− No. of case patients	967	967	973	949	964
RR (95% CI)	1.00 (referent)	0.96 (0.85 to 1.09)	0.94 (0.86 to 1.03)	0.92 (0.81 to 1.05)	0.94 (0.85 to 1.04)	.13	.65	.36
ER+ No. of case patients	3770	3992	4136	3980	3987
RR (95% CI)	1.00 (referent)	1.02 (0.95 to 1.09)	1.03 (0.97 to 1.08)	0.99 (0.93 to 1.05)	0.99 (0.93 to 1.07)	.57	.02
By PR status
Total fruit and vegetables
PR− No. of case patients	1440	1494	1496	1529	1529
RR (95% CI)	1.00 (referent)	0.99 (0.91 to 1.07)	0.98 (0.89 to 1.07)	0.97 (0.88 to 1.07)	0.97 (0.87 to 1.09)	.60	.04	.79
PR+ No. of case patients	3028	3265	3225	3348	3296
RR (95% CI)	1.00 (referent)	1.02 (0.97 to 1.07)	0.98 (0.93 to 1.03)	1.00 (0.95 to 1.06)	0.99 (0.92 to 1.07)	.88	.08
Total vegetables
PR− No. of case patients	1472	1483	1494	1514	1521
RR (95% CI)	1.00	0.96 (0.89 to 1.03)	0.95 (0.86 to 1.03)	0.94 (0.85 to 1.04)	0.94 (0.84 to 1.03)	.30	.08	.12
PR+ No. of case patients	2979	3163	3298	3313	3403
RR (95% CI)	1.00 (referent)	1.01 (0.94 to 1.07)	1.03 (0.96 to 1.11)	1.01 (0.95 to 1.08)	1.02 (0.96 to 1.10)	.45	.15
Total fruits
PR− No. of case patients	1471	1475	1524	1492	1524
RR (95% CI)	1.00 (referent)	0.96 (0.88 to 1.06)	0.97 (0.90 to 1.04)	0.95 (0.87 to 1.04)	0.98 (0.91 to 1.06)	.49	.50	.62
PR+ No. of case patients	3066	3261	3344	3226	3262
RR (95% CI)	1.00 (referent)	1.02 (0.96 to 1.09)	1.02 (0.97 to 1.08)	0.99 (0.92 to 1.05)	1.01 (0.93 to 1.10)	.97	.004

* The relative risks were adjusted for ethnicity (White, African-American, Hispanic, Asian, others), family history of breast cancer (yes, no), personal history of benign breast disease (yes, no), alcohol consumption (nondrinkers, >0 to <5, 5 to <15, 15 to <30, ≥30g/day), smoking status (never, past, current), education (<high school, high school, >high school), physical activity (low, medium, high), age at menarche (<11, 11 to 12, 13 to 14, ≥15 years), body mass index (<23, 23 to <25, 25 to <30, ≥30kg/m²), height (<1.60, 1.60 to <1.65, 1.65 to <1.70, 1.70 to <1.75, ≥1.75 m), oral contraceptive use (never, ever), menopausal status (premenopausal women, never user of hormone replacement therapy among postmenopausal women, past user of hormone replacement therapy among postmenopausal women, and current user of hormone replacement therapy among postmenopausal women), energy intake (kcal/day, continuous), combination between parity (0, 1 to 2, ≥3) and age of first birth (≤25, >25 years). Age in years and year of questionnaire return were included as stratification variables. All statistical tests were two-sided. RR = relative risk; CI = confidence interval; ER = estrogen receptor; PR = progesterone receptor.

† P, test for trend was calculated using the Wald test statistic.

‡ P, test for between-studies heterogeneity for quintile 5 was calculated using the Q statistic.

§ P, test for common effects by receptor status for quintile 5 was calculated using a contrast test.

|| There were nine participants who developed breast cancer for whom data on total fruits were missing and 11 participants for whom data on total vegetables were missing who were included in the analyses of total fruit and vegetables.

We pooled the study-specific rate ratios using a random-effects model (54) weighted by the sum of the inverse of the variance and the estimated between-studies variance components. Between-studies heterogeneity was evaluated using the Q statistic (54,55). To test the assumption of proportional hazards, we added an interaction term between age and fruit and vegetable intake into the model and pooled the study-specific parameter estimates of the interaction term using the random-effects model (54). We observed no violation of the proportional hazards assumption.

We modeled fruit and vegetable intake using study-specific quintiles and categories defined by common absolute intake cut points across studies. To test for trend across categories of intake, the median value of the intake category for each participant was entered as a continuous term in the model. We also modeled fruit and vegetable intakes as continuous variables when nonparametric regression analyses showed that the associations were linear. To test for linearity, all studies were aggregated into a single dataset, participants within the top 1% of fruit and vegetable intake were excluded to prevent a spurious association due to the influence of extreme values, and participants were stratified by age at baseline, year of baseline questionnaire return, and study. We tested for linearity using the likelihood ratio test, comparing the model fit including the linear term and cubic spline terms selected by stepwise regression with that of a model including only the linear term (56,57). We examined whether the effects of fruit and vegetable intake varied by cancer subtype using a contrast test (58,59).

To investigate whether associations between fruit and vegetable consumption and breast cancer risk varied by factors that may influence estrogen levels or the bioactive functions of phytochemicals, we further evaluated the associations between total fruit and vegetable, total fruit, and total vegetable consumption and risk of ER⁻ and ER⁺ breast cancer by menopausal status at diagnosis using a previously created algorithm (60) (premenopausal, postmenopausal), postmenopausal hormone use (never, past, current user at baseline among postmenopausal women), alcohol consumption (nondrinker, <15g/day, ≥15g/day), body mass index (<25, ≥25kg/m²), multivitamin use (current user, nonuser), family history of breast cancer (yes, no), approximate median age at diagnosis (<64, ≥64 years), and smoking status (never, past, current smoker) using a mixed-effects meta-regression model (11,61). We also explored sources of heterogeneity by region, number of fruit and vegetable items on the FFQ, median follow-up time, median age at diagnosis, and the proportion of women who were postmenopausal at baseline. For all tests, two-sided 95% confidence intervals (CIs) were calculated, and P less than.05 was considered statistically significant.

Results

In the 20 prospective studies with maximum follow-up of 11 to 20 years, 34 526 incident invasive breast cancer case patients were identified among a total of 993 466 women. Receptor status information was available for 24 673 breast cancers, among which there were 19 869 ER⁺, 4821 ER⁻, 16 162 PR⁺, and 7488 PR⁻ breast cancer case patients (Table 1). Fruit and vegetable consumption varied substantially across studies, with the median intakes across studies ranging from 118 to 392g/day for total fruit and from 61 to 259g/day for total vegetables (Table 1).

In all analyses presented, associations were similar in age-adjusted and multivariable models; therefore, we only present the multivariable results (Table 2). For ER⁻ breast cancer, there was a statistically significant inverse trend for total fruit and vegetable consumption (P_trend = .03). When intakes of fruits and vegetables were examined separately, inverse associations were observed for both total fruit (pooled multivariable relative risk [RR] comparing the highest vs lowest quintile = 0.94, 95% CI = 0.85 to 1.04) and total vegetable intake, with a stronger association observed for total vegetable intake (pooled multivariable RR = 0.82, 95% CI = 0.74 to 0.90 comparing the highest vs lowest quintile; P_{heterogeneity} for quintile 5 = .50) (Figure 1). The difference in the associations for ER⁻ and ER⁺ breast cancer was statistically significant only for total vegetable intake (P_{common-effects} for highest quintile < .001).

Figure 1.

Study-specific and pooled multivariable-adjusted relative risks of estrogen receptor–negative (ER⁻) breast cancer and total vegetable consumption, quintile 5 vs quintile1. The squares and horizontal lines correspond to the study-specific relative risks and 95% confidence intervals, respectively, for the comparison of quintile 5 (Q5) to quintile 1 (Q1) of total vegetable consumption. The size of the squares reflects the study-specific weight (inverse of the variance). The diamond represents the pooled relative risk and 95% confidence interval. All statistical tests were two-sided. AARP = National Institutes of Health–AARP Diet and Health Study; CARET =Beta-Carotene and Retinol Efficacy Trial; BWHS = Black Women’s Health Study; BCDDP = Breast Cancer Detection Demonstration Project Follow-up Study; CTS = California Teachers Study, CLUE2 = Campaign Against Cancer and Heart Disease; CNBSS = Canadian National Breast Screening Study; CPS2 = Cancer Prevention Study II Nutrition Cohort; IWHS = Iowa Women’s Health Study; JPHCI = Japan Public Health Center-Based Prospective Study I; MCCS = Melbourne Collaborative Cohort Study; NHSa = Nurses’ Health Study(a); NHSb = Nurses’ Helath Study(b); NHS2 = Nurses’ Health Study II; NLCS = Netherlands Cohort Study; NYU = New York University Women’s Health Study; ORDET = Prospective Study on Hormones, Diet and Breast Cancer; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; SMC = Swedish Mammography Cohort; WHS =Women’s Health Study; WLHS = Women’s Lifestyle and Health Study.

Intakes of total fruit and vegetables, total fruit, and total vegetables were not statistically significantly associated with the risk of breast cancer overall or ER⁺ breast cancer, with pooled multivariable relative risks ranging from 0.98 to 1.04 (Table 2). For example, the pooled multivariable relative risk comparing the highest vs lowest quintile of total vegetable consumption in relation to risk of ER⁺ breast cancer was 1.04 (95% CI= 0.97 to 1.11) (Table 2). The results for total breast cancer did not change materially when we limited the analyses to the 24 675 case patients with nonmissing ER data (data not shown). In addition, when we analyzed the 9856 case patients with missing ER data, the pooled multivariable relative risks for total fruit and vegetables, total fruit, and total vegetables were comparable with those for ER⁺ breast cancer, ranging from 0.95 to 1.01 when comparing the highest vs lowest quintile of intake. There was a suggestion that the study-specific results were heterogeneous for intakes of these three fruit and vegetable groups and risk of ER⁺ breast cancer (for each group, P _{heterogeneity} for quintile 5 ≤.06) (Table 2). For example, the study-specific multivariable relative risks for quintile 5 compared with quintile 1 ranged from 0.55 to 1.39 for the association between total vegetable consumption and risk of ER⁺ breast cancer, with positive associations observed in nine studies and inverse associations observed in 12 studies. In exploratory analyses to identify possible reasons for the heterogeneity, differences in the number of fruit and vegetable questions on the FFQs, median age at diagnosis, proportion of women who were postmenopausal at baseline, and length of follow-up did not account for the heterogeneity observed. There was a suggestion that the associations for quintile 5 differed by region, with stronger inverse associations observed among studies from continents other than North America compared with studies conducted in North America (P _difference by region ≤.02; data not shown).

Intakes of total fruits and vegetables, total fruits, and total vegetables were not statistically significantly associated with risk of PR⁻ or PR⁺ breast cancers. For total vegetable consumption, the pooled multivariable relative risk for PR⁻ breast cancer was 0.94 (95% CI = 0.84 to 1.03) comparing the highest to the lowest quintile (Table 2).

We also evaluated associations with fruit and vegetable intake categorized using common absolute intake cut points across studies (Supplementary Table 1, available online). Associations comparing the highest vs lowest category were of similar magnitude to those observed in the quintile analyses. For example, for total vegetable intake, the pooled multivariable relative risks comparing ≥ 400g/day to 100 to <200g/day were 0.85 (95% CI = 0.75 to 0.97) for ER⁻ tumors and 1.03 (95% CI = 0.97 to 1.10) for ER⁺ tumors (Supplementary Table 1, available online). The age-standardized incidence rates of ER⁻ tumors were 32 per 100 000 person-years among those who consumed ≥400g/day of total vegetables and 36 per 100 000 person-years among those who consumed <100g/day of total vegetables. In these analyses, statistically significant differences in the risk of ER⁻ vs ER⁺ breast cancer were observed for both total fruit and vegetable intake and total vegetable intake (P _{common-effects} by ER status for highest category ≤.02) (Supplementary Table 1, available online). To compare more extreme intakes, we further modeled total fruit and vegetable, total fruit, and total vegetable consumption as deciles; the risk estimates for these three groups for both ER⁻ and ER⁺ breast cancer (data not shown) were consistent with those observed for the analyses of quintiles and categories based on common absolute intake cutpoints.

The nonparametric regression analyses indicated that all associations presented in Table 2 were linear (P _nonlinearity > .05). Therefore, we conducted analyses in which fruit and vegetable intakes were modeled as continuous variables. The pooled multivariable relative risks for ER⁻ breast cancer for a 300g/day increment (approximately three servings/day) in intake were 0.94 (95% CI = 0.91 to 0.98) for total fruits and vegetables, 0.88 (95% CI = 0.81 to 0.95) for total vegetables, and 0.96 (95% CI = 0.91 to 1.00) for total fruits (P _{heterogeneity} > .34 for each). These three groups were non-statistically significantly inversely associated with risk of PR⁻ breast cancer, but no associations or non-statistically significant positive associations were observed for the risk of ER⁺ and PR⁺ breast cancer (data not shown). Further adjustment for intake of beta-carotene, lutein, and fiber (which are concentrated in vegetables and could be driving the association for total vegetable consumption) (62) did not substantially change the inverse association between total vegetable consumption and ER⁻ breast cancer risk. Analyses that only included case patients diagnosed after 5 years of study enrollment yielded similar relative risks to those using all case patients (data not shown).

We further analyzed breast cancer subtypes classified by ER and PR status jointly (Table 3). For total vegetable intake, we observed a statistically significant 16% lower risk of ER⁻PR⁻ breast cancer (pooled multivariable RR comparing the highest vs lowest quintiles of total vegetables = 0.84, 95% CI = 0.75 to 0.93), but a non-statistically significant positive association for ER⁺PR⁺ breast cancer (pooled multivariable RR for same comparison = 1.04, 95% CI = 0.98 to 1.11) (Table 3). For total fruit and vegetable and total fruit consumption, we observed weaker non-statistically significant inverse associations with risk of ER⁻PR⁻ breast cancer than observed for total vegetable intake. The risk estimates for ER⁺PR⁺ breast cancer ranged from 1.00 to 1.04 comparing the highest vs lowest quintile for each of the three fruit and vegetable groups.

Table 3.

Pooled multivariable relative risks (95% confidence interval) of breast cancer by joint estrogen receptor/progesterone receptor subtype for quintiles of fruit and vegetable intake*,†

		Quintile of intake					P trend‡	P between-studies heterogeneity for quintile 5§	P common effects by receptor status for quintile 5||
		Q1	Q2	Q3	Q4	Q5
Total fruit and vegetables
ER−PR−	No. of case patients	773	793	793	771	774
	RR (95% CI)	1.00 (referent)	0.98 (0.89 to 1.08)	0.97 (0.87 to 1.09)	0.92 (0.82 to 1.03)	0.93 (0.80 to 1.08)	.29	.06	.14
ER−PR+	No. of case patients	124	102	105	98	113
	RR (95% CI)	1.00 (referent)	0.74 (0.57 to 0.97)	0.72 (0.53 to 0.97)	0.63 (0.48 to 0.84)	0.70 (0.51 to 0.96)	.21	.30
ER+PR−	No. of case patients	648	671	675	736	727
	RR (95% CI)	1.00 (referent)	0.97 (0.87 to 1.08)	0.95 (0.85 to 1.06)	1.03 (0.92 to 1.15)	1.02 (0.90 to 1.14)	.41	.69
ER+PR+	No. of case patients	2867	3115	3079	3198	3139
	RR (95% CI)	1.00 (referent)	1.03 (0.98 to 1.08)	0.99 (0.94 to 1.05)	1.01 (0.96 to 1.07)	1.00 (0.93 to 1.08)	.87	.12
Total vegetables
ER−PR−	No. of case patients	804	793	772	789	745
	RR (95% CI)	1.00 (referent)	0.94 (0.85 to 1.04)	0.89 (0.80 to 1.00)	0.91 (0.80 to 1.04)	0.84 (0.75 to 0.93)	.001	.53	<.001
ER−PR+	No. of case patients	120	108	105	106	103
	RR (95% CI)	1.00 (referent)	0.84 (0.61 to 1.15)	0.76 (0.58 to 1.00)	0.72 (0.52 to 1.00)	0.68 (0.51 to 0.90)	.04	.75
ER+PR−	No. of case patients	646	664	688	706	750
	RR (95% CI)	1.00 (referent)	0.96 (0.85 to 1.09)	0.97 (0.87 to 1.09)	0.97 (0.84 to 1.12)	1.04 (0.89 to 1.20)	.27	.10
ER+PR+	No. of case patients	2820	3008	3139	3166	3259
	RR (95% CI)	1.00 (referent)	1.01 (0.95 to 1.08)	1.03 (0.96 to 1.11)	1.02 (0.96 to 1.09)	1.04 (0.98 to 1.11)	.13	.17
Total fruits
ER−PR−	No. of case patients	778	781	778	785	782
	RR (95% CI)	1.00 (referent)	0.97 (0.84 to 1.11)	0.95 (0.85 to 1.05)	0.97 (0.85 to 1.11)	0.97 (0.86 to 1.09)	.75	.32	.83
ER−PR+	No. of case patients	114	112	111	92	113
	RR (95% CI)	1.00 (referent)	0.93 (0.71 to 1.22)	0.87 (0.66 to 1.14)	0.71 (0.53 to 0.97)	0.86 (0.58 to 1.27)	.41	.10
ER+PR−	No. of case patients	665	676	725	676	713
	RR (95% CI)	1.00 (referent)	0.98 (0.88 to 1.09)	1.00 (0.90 to 1.11)	0.92 (0.83 to 1.03)	0.99 (0.88 to 1.10)	.49	.83
ER+PR+	No. of case patients	2906	3107	3191	3096	3096
	RR (95% CI)	1.00 (referent)	1.03 (0.96 to 1.10)	1.03 (0.97 to 1.09)	1.00 (0.93 to 1.07)	1.01 (0.93 to 1.10)	.83	.01

* The relative risks were adjusted for ethnicity (White, African-American, Hispanic, Asian, others), family history of breast cancer (yes, no), personal history of benign breast disease (yes, no), alcohol consumption (non-drinkers, >0 to <5, 5 to <15, 15 to <30, ≥30 g/day), smoking status (never, past, current), education (<high school, high school, >high school), physical activity (low, medium, high), age at menarche (<11, 11 to 12, 13 to 14, ≥15 years), body mass index (<23, 23 to <25, 25 to <30, ≥30 kg/m²), height (<1.60, 1.60 to <1.65, 1.65 to <1.70, 1.70 to <1.75, ≥1.75 m), oral contraceptive use (never, ever), menopausal status (premenopausal women, never user of hormone replacement therapy among postmenopausal women, past user of hormone replacement therapy among postmenopausal women, and current user of hormone replacement therapy among postmenopausal women), energy intake (kcal/d, continuous), combination between parity (0,1 to 2, ≥3) and age of first birth (≤25, >25 years); age in years and year of questionnaire return were included as stratification variables. All statistical tests were two-sided. RR = relative risk; CI = confidence interval; ER = estrogen receptor; PR = progesterone receptor.

† There were nine participants who developed breast cancer for whom data on total fruits were missing and 11 participants for whom data on total vegetables were missing who were included in the analyses of total fruit and vegetables.

‡ P, test for trend was calculated using the Wald test statistic.

§ P, test for between-studies heterogeneity for quintile 5 was calculated using the Q statistic.

|| P, test for common effects by receptor status for quintile 5 was calculated using a contrast test.

When we examined fruit and vegetable intake grouped according to botanical taxonomy (13) (Supplementary Table 2, available online), only intake of the Rosaceae family (eg, apples, peaches) was statistically significantly inversely associated with the risk of ER⁻ breast cancer: the pooled multivariable relative risk for a 100g/day increment in intake (approximately 1 serving/day) was 0.91 (95% CIs = 0.88 to 0.95) (Supplementary Table 2, available online) for Rosaceae. Non-statistically significant inverse associations were observed for intakes of the Cruciferae (eg, brococoli, cabbage), Cucurbitaceae (eg, melon, squash), and Leguminosae (eg, beans, peas) families and risk of ER⁻ breast cancer. For these botanical groups, null associations with relative risks ranging from 0.99 to 1.02 were observed with risk of ER⁺ breast cancer. Intakes of the Rutaceae (eg, grapefruits, oranges) and Solanaceae (eg, potatoes, tomatoes) families were not associated with risk of ER⁻ or ER⁺ breast cancer.

In the analyses of intakes of specific fruits and vegetables (Table 4), we observed statistically significant inverse associations for intakes of apples/pears, peaches/nectarines/apricots, and strawberries, carrots, and lettuce/salad with risk of ER⁻ breast cancer but non-statistically significant associations with risk of ER⁺ breast cancer. For ER⁻ breast cancer, when we included intakes of strawberries, apples/pears, and peaches/nectarines/apricots or intakes of lettuce/salad and carrots simultaneously in the model, the association for each food did not change substantially and remained statistically significant except that the association for apple/pear intake was attenuated and became non-statistically significant (data not shown). Although some studies have reported potentially adverse effects of bioactive compounds present in fruits and vegetables (12,63–67), no statistically significant positive associations were observed for any of the specific fruit and vegetables or botanically defined fruit and vegetable subgroups examined in our analyses.

Table 4.

Pooled multivariable relative risks (95% confidence interval) of breast cancer by estrogen receptor subtype for specific fruits and vegetables*

Food item	No. of studies†	No. of case patients	Reference serving size	1 Serving (g/day)	RR (95% CI), unit: 1 serving/day	P between-studies heterogeneity‡	P common effects by receptor status§
Fruits
Apples, pears, applesauce||,¶
ER−	18	4657	1 or 1/2 cup	138	0.92 (0.85 to 0.99)	.43	.12
ER+	18	19 408			0.98 (0.94 to 1.02)	.20
Bananas||,¶,#,**,††
ER−	15	4103	1	114	0.98 (0.88 to 1.08)	.21	.89
ER+	15	16 505			0.97 (0.92 to 1.02)	.86
Cantaloupe||,¶,‡‡,§§,||||,¶¶,##
ER−	14	3505	1/4 melon	134	0.92 (0.74 to 1.15)	.79	.22
ER+	14	15 474			1.10 (0.93 to 1.29)	.05
Grapefruit||,‡‡,§§,¶¶,##,***
ER−	15	3635	1/2 fruit	120	1.06 (0.98 to 1.16)	.66	.33
ER+	15	15 942			1.00 (0.92 to 1.09)	.002
Oranges||,‡‡,§§,¶¶,##,***
ER−	15	3649	1	131	0.93 (0.83 to 1.04)	.35	.22
ER+	15	15 936			1.01 (0.95 to 1.06)	.89
Peaches, nectarines, apricots||,¶,#,**,††,§§,||||,##,†††
ER−	11	3113	1 or 1/2 cup	87	0.81 (0.70 to 0.94)	.30	.01
ER+	11	13 162			1.00 (0.95 to 1.05)	.47
Strawberries||,¶,#,**,††,‡‡,§§,||||,##,†††
ER−	11	2912	1/2 cup	75	0.56 (0.41 to 0.76)	.50	<.001
ER+	11	12 956			1.01 (0.91 to 1.13)	.60
Fruit juice
ER−	20	4777			1.01 (0.97 to 1.05)	.86
ER+	20	19 708	6 oz	190	1.01 (0.98 to 1.04)	.08	.98
Vegetables
Broccoli||,¶,§§,||||,##
ER−	15	3913	1/2 cup	78	0.93 (0.82 to 1.06)	.30	.07
ER+	15	16 449			1.06 (1.00 to 1.12)	.54
Cabbage||,¶,‡‡,¶¶
ER−	17	4271	1/2 cup	68	1.04 (0.87 to 1.24)	.60	.62
ER+	17	18 509			0.99 (0.91 to 1.08)	.91
Carrots||
ER−	19	4740	1/2 cup	57	0.92 (0.84 to 1.00)	.50	.02
ER+	19	19 678			1.03 (0.99 to 1.08)	.63
Lettuce, salad||,¶¶,##,‡‡‡
ER−	17	3828	1 cup	56	0.91 (0.84 to 0.98)	.19	.02
ER+	17	16 081			1.01 (0.97 to 1.05)	.11
Spinach||,¶¶,§§§
ER−	17	4344	1/2 cup	73	0.91 (0.80 to 1.04)	.70	.31
ER+	17	18 341			1.00 (0.90 to 1.11)	.04
Tomatoes||,‡‡‡
ER−	18	4273	1	122	0.92 (0.83 to 1.01)	.58	.10
ER+	18	17 332			1.01 (0.94 to 1.08)	.04
Yams||,¶,‡‡,§§,||||,##,§§§
ER−	13	3610	1/2 cup	128	1.13 (0.75 to 1.69)	.46	.45
ER+	13	15 532			0.94 (0.76 to 1.17)	.85
Potatoes||
ER−	19	4749	1	202	1.12 (0.99 to 1.26)	.66	.33
ER+	19	19 743			1.04 (0.96 to 1.13)	.15

* The relative risks were adjusted for ethnicity (White, African-American, Hispanic, Asian, others), family history of breast cancer (yes, no), personal history of benign breast disease (yes, no), alcohol consumption (non-drinkers, >0 to <5, 5 to <15, 15 to <30, ≥30g/day), smoking status (never, past, current), education (<high school, high school, >high school), physical activity (low, medium, high), age at menarche (<11, 11 to 12, 13 to 14, ≥15 years), body mass index (<23, 23 to <25, 25 to <30, ≥30kg/m²), height (<1.60, 1.60 to <1.65, 1.65 to <1.70, 1.70 to <1.75, ≥1.75 m), oral contraceptive use (never, ever), menopausal status (premenopausal women, never user of hormone replacement therapy among postmenopausal women, past user of hormone replacement therapy among postmenopausal women, and current user of hormone replacement therapy among postmenopausal women), energy intake (kcal/day, continuous), combination between parity (0, 1 to 2, ≥3) and age of first birth (≤25, >25 years). Age in years and year of questionnaire return were included as stratification variables. All statistical tests were two-sided. RR = relative risk; CI = confidence interval; ER = estrogen receptor.

† For the result of Cantaloupes, Grapefuits, Oranges, Strawberries, Cabbage and Lettuce, salad, the total number of studies that were not included and that were included in the analyses adds up to 21 because NHS (a) and NHS (b) were counted separately.

‡ P, test for between-studies heterogeneity was calculated using the Q statistic.

§ P, test for common effects by receptor status was calculated using a contrast test.

|| Japan Public Health Center-Based Prospective Study I and Prospective Study on Hormones, Diet and Breast Cancer were not included in this analysis because consumption of this item was not measured.

¶ Prospective Study on Hormones, Diet and Breast Cancer was not included in this analysis because consumption of this item was not measured.

# Breast Cancer Detection Demonstration Project Follow-up Study was not included in this analysis because consumption of this item was not measured.

** Cancer Prevention Study II Nutrition Cohort was not included in this analysis because consumption of this item was not measured.

†† CLUE II: Campaign Against Cancer and Heart Disease was not included in this analysis because consumption of this item was not measured.

‡‡ Canadian National Breast Screening Study was not included in this analysis because consumption of this item was not measured.

§§ Swedish Mammography Cohort was not included in this analysis because consumption of this item was not measured.

|||| Netherlands Cohort Study was not included in this analysis because consumption of this item was not measured.

¶¶ Nurses’ Health Study (a) was not included in this analysis because consumption of this item was not measured.

## Women’s Lifestyle and Health Study was not included in this analysis because consumption of this item was not measured.

*** New York University Women’s Health Study was not included in this analysis because consumption of this item was not measured.

††† Black Women’s Health Study was not included in this analysis because consumption of this item was not measured.

‡‡‡ National Institutes of Health–AARP Diet and Health Study was not included in this analysis because consumption of this item was not measured.

§§§ Melbourne Collaborative Cohort Study was not included in this analysis because consumption of this item was not measured.

We observed an inverse association between total fruit and vegetable, total fruit, and total vegetable intakes and risk of ER⁻ breast cancer in most of the population subgroups evaluated. These associations, as well as those for ER⁺ breast cancer, were not modified by several breast cancer risk factors (all P values, test for interaction ≥ 0.09) (Table 5).

Table 5.

Pooled multivariable relative risks (95% confidence interval) of breast cancer by estrogen receptor status for a 300g/day increment in fruit and vegetable intake by other risk factors*

Effect modifiers	No. of case patients	Total fruits and vegetables RR (95% CI)	P interaction†	Total vegetables RR (95% CI)	P interaction†	Total fruits RR (95% CI)	P interaction†
Subgroup results
Menopausal status
ER−
Premenopausal‡,§	956	1.00 (0.91 to 1.10)	.41	0.85 (0.66 to 1.04)	.35	1.07 (0.96 to 1.21)	0.09
Postmenopausal||	2809	0.96 (0.92 to 1.00)		0.91 (0.83 to 1.00)		0.97 (0.92 to 1.03)
ER+
Premenopausal‡,§	2276	0.98 (0.92 to 1.05)	.44	0.93 (0.83 to 1.05)	.08	1.01 (0.92 to 1.11)	.85
Postmenopausal	13 450	1.01 (0.98 to 1.03)		1.06 (1.02 to 1.10)		1.00 (0.96 to 1.03)
Postmenopausal hormone use¶
ER−
Never#	1696	0.94 (0.88 to 1.00)	.18	0.88 (0.77 to 1.00)	.35	0.93 (0.84 to 1.03)	.30
Ever**	1374	1.00 (0.94 to 1.06)		0.97 (0.85 to 1.10)		1.02 (0.95 to 1.10)
ER+
Never#	6128	1.01 (0.98 to 1.04)	.55	1.08 (1.01 to 1.16)	.14	0.98 (0.94 to 1.03)	.92
Ever††	7160	1.00 (0.97 to 1.03)		1.03 (0.98 to 1.09)		1.00 (0.95 to 1.05)
Oral contraceptive use ‡‡
ER−
Never	2029	0.93 (0.87 to 0.99)	.67	0.86 (0.74 to 0.99)	.49	0.94 (0.87 to 1.03)	.88
Ever	2337	0.95 (0.90 to 1.00)		0.91 (0.81 to 1.01)		0.96 (0.90 to 1.03)
ER+
Never	9658	0.99 (0.98 to 1.00)	.04	1.00 (0.9801.02)	.15	0.99 (0.97 to 1.01)	.08
Ever	8860	1.01 (1.00 to 1.01)		1.02 (1.00 to 1.04)		1.00 (0.99 to 1.02)
Multivitamin use§§
ER−
No||||	2131	0.97 (0.90 to 1.04)	.75	0.91 (0.79 to 1.04)	.50	0.98 (0.90 to 1.07)	.87
Yes¶¶	1702	0.94 (0.89 to 0.99)		0.84 (0.75 to 0.95)		0.96 (0.90 to 1.03)
ER+
No	8708	1.01 (1.00 to 1.01)	.22	1.01 (0.99 to 1.03)	.57	1.01 (1.00 to 1.02)	.15
Yes	7594	1.00 (0.99 to 1.01)		1.01 (0.99 to 1.03)		0.99 (0.98 to 1.01)
Alcohol consumption ##
ER−
Never	1518	0.92 (0.87 to 0.98)	.49	0.88 (0.75 to 1.02)	.18	0.94 (0.86 to 1.02)	.81
<1 drink/day***	2422	0.96 (0.91 to 1.01)		0.93 (0.84 to 1.04)		0.97 (0.91 to 1.03)
≥1drink/day†††	475	0.91 (0.80 to 1.02)		0.72 (0.56 to 0.93)		0.95 (0.83 to 1.10)
ER+
Never	6569	1.00 (0.99 to 1.01)	.94	1.01 (0.99 to 1.03)	.74	1.00 (0.98 to 1.02)	.96
<1 drink/day‡‡‡	10 256	1.00 (0.99 to 1.01)		1.00 (0.98 to 1.02)		1.00 (0.98 to 1.01)
≥1drink/day§§§	2474	0.99 (0.97 to 1.01)		1.02 (0.99 to 1.05)		0.99 (0.97 to 1.01)
Smoking status||||||
ER−
Never	2303	0.94 (0.90 to 0.99)	.71	0.89 (0.80 to 0.99)	.55	0.95 (0.90 to 1.02)	.86
Past¶¶¶	1277	0.97 (0.90 to 1.04)		0.94 (0.80 to 1.09)		0.98 (0.90 to 1.06)
Current###	701	0.93 (0.84 to 1.02)		0.82 (0.66 to 1.02)		0.96 (0.83 to 1.12)
ER+
Never	9228	1.00 (0.99 to 1.01)	.89	1.01 (1.00 to 1.03)	.18	1.00 (0.98 to 1.01)	.68
Past****	6069	1.00 (0.99 to 1.02)		1.02 (1.00 to 1.04)		1.00 (0.98 to 1.02)
Current****	2107	1.00 (0.98 to 1.01)		0.98 (0.95 to 1.01)		1.00 (0.98 to 1.03)
Body mass index
ER−
<25kg/m2	2572	0.94 (0.89 to 0.98)	.58	0.88 (0.79 to 0.99)	.97	0.95 (0.89 to 1.01)	.59
≥25kg/m2	2163	0.96 (0.89 to 1.04)		0.89 (0.77 to 1.02)		0.97 (0.89 to 1.06)
ER+
<25kg/m2	10 380	1.00 (0.99 to 1.00)	.24	1.01 (0.99 to 1.02)	.40	0.99 (0.98 to 1.00)	.32
≥25kg/m2	9289	1.00 (0.99 to 1.01)		1.01 (0.99 to 1.03)		1.00 (0.99 to 1.02)
Age at diagnosis
ER−
<64 y	2981	0.93 (0.88 to 0.98)	.22	0.87 (0.78 to 0.97)	.52	0.94 (0.89 to 1.00)	.23
≥64 y††††	1803	0.97 (0.92 to 1.03)		0.91 (0.78 to 1.06)		0.99 (0.93 to 1.06)
ER+
<64 y	9323	1.00 (0.99 to 1.01)	.86	1.00 (0.99 to 1.02)	.22	1.00 (0.99 to 1.01)	.30
≥64 y‡‡‡‡	10 080	1.00 (0.99 to 1.01)		1.02 (0.99 to 1.04)		1.00 (0.98 to 1.01)
Family history of breast cancer§§§§
ER−
No	3783	0.96 (0.93 to 1.00)	.23	0.92 (0.83 to 1.02)	.14	0.98 (0.93 to 1.03)	.92
Yes||||||||	583	0.90 (0.81 to 1.00)		0.75 (0.56 to 1.01)		0.97 (0.85 to 1.11)
ER+
No	15 759	1.00 (0.99 to 1.01)	.45	1.01 (0.99 to 1.02)	.99	1.00 (0.99 to 1.01)	.38
Yes¶¶¶¶	2732	1.00 (0.97 to 1.02)		1.01 (0.97 to 1.04)		0.99 (0.96 to 1.02)
Race ####
ER−
White*****	3875	0.94 (0.91 to 0.98)	.71	0.87 (0.80 to 0.95)	.20	0.96 (0.91 to 1.01)	.99
Black†††††	359	0.95 (0.84 to 1.07)		0.81 (0.58 to 1.14)		0.97 (0.84 to 1.11)
Other‡‡‡‡‡	276	1.01 (0.84 to 1.21)		1.11 (0.73 to 1.69)		0.97 (0.81 to 1.16)
ER+
White*****	17 439	1.01 (0.99 to 1.03)	.47	1.03 (0.98 to 1.07)	.73	1.01 (0.97 to 1.04)	.59
Black§§§§§	666	1.03 (0.88 to 1.20)		1.09 (0.88 to 1.35)		0.99 (0.83 to 1.18)
Other||||||||||	961	0.96 (0.89 to 1.04)		0.99 (0.86 to 1.15)		0.95 (0.86 to 1.05)

* The relative risks were adjusted for ethnicity (White, African-American, Hispanic, Asian, others), family history of breast cancer (yes, no), personal history of benign breast disease (yes, no), alcohol consumption (nondrinkers, >0 to <5, 5 to <15, 15 to <30, ≥30g/day), smoking status (never, past, current), education (<high school, high school, >high school), physical activity (low, medium, high), age at menarche (<11, 11 to 12, 13 to 14, ≥15 years), body mass index (<23, 23 to <25, 25 to <30, ≥30kg/m²), height (<1.60, 1.60 to <1.65, 1.65 to <1.70, 1.70 to <1.75, ≥1.75 m), oral contraceptive use (never, ever), menopausal status (premenopausal women, never user of hormone replacement therapy among postmenopausal women, past user of hormone replacement therapy among postmenopausal women, and current user of hormone replacement therapy among postmenopausal women), energy intake (kcal/day, continuous), combination between parity (0, 1 to 2, ≥3) and age of first birth (≤25, >25 years). Age in years and year of questionnaire return were included as stratification variables. The effect modifiers was not included as a covariate in the models evaluating that effect modifier. All statistical tests were two-sided. RR = relative risk; CI = confidence interval; ER = estrogen receptor.

† P, test for interaction was calculated by using a mixed-effects meta-regression.

‡ The Beta-Carotene and Retinol Efficacy Trial, Iowa Women’s Health Study, Netherlands Cohort Study, and Prostate, Long, Colorectal, and Ovarian Cancer Screening Trial were not included in this analyses because all case patients were postmenopausal women.

§ The National Institutes of Health–AARP Diet and Health Study, Breast Cancer Detection Demonstration Project Follow-up Study, CLUE II: Campaign Against Cancer and Heart Disease and Cancer Prevention Study II Nutrition Cohort were not included in this stratum because few case patients (n < 15) were premenopausal women.

|| The Nurses’ Health Study II and Women’s Lifestyle and Health Study were not included in this stratum because few case patients (n < 15) were postmenopausal women.

¶ The Beta-Carotene and Retinol Efficacy Trial, Japan Public Health Center-Based Prospective Study I, and New York University Women’s Health Study were not included in this analysis because this variable was not measured.

# The Nurses’ Health Study II was not included in this stratum because few case patients (n < 15) were never users of postmenopausal hormones.

** CLUE II: Campaign Against Cancer and Heart Disease, Nurses’ Health Study II, Prospective Study on Hormones, Diet and Breast Cancer, and Women’s Lifestyle and Health Study were not included in this stratum, because few case patients (n < 15) were ever users of postmenopausal hormones.

†† The Women’s Lifestyle and Health Study were not included in this stratum, because few case patients (n < 15) were ever users of postmenopausal hormones.

‡‡ The Beta-Carotene and Retinol Efficacy Trial, Japan Public Health Center-Based Prospective Study I, and New York University Women’s Health Study were not included in these analyses, because this variable was not measured.

§§ The Canadian National Breast Screening Study, Prospective Study on Hormones, Diet and Breast Cancer, Swedish Mammography Cohort, and Women’s Lifestyle and Health Study were not included in these analyses because data on multivitamin use were not available at baseline in these studies.

|||| The Beta-Carotene and Retinol Efficacy Trial and the CLUE II: Campaign Against Cancer and Heart Disease were not included in this stratum, because few case patients (n < 15) were nonusers of multivitamins.

¶¶ The Japan Public Health Center-Based Prospective Study I was not included in this stratum, because few case patients (n < 15) were using multivitamins.

## The Black Women’s Health Study and New York University Women’s Health Study were not included in these analyses because data on alcohol consumption were not available in our database or was not measured at baseline, respectively.

*** The Japan Public Health Center-Based Prospective Study I and the Beta-Carotene and Retinol Efficacy Trial were not included in this stratum, because few case patients (n < 15) drank alcohol.

††† The Beta-Carotene and Retinol Efficacy Trial, Japan Public Health Center-Based Prospective Study I, CLUE II: Campaign Against Cancer and Heart Disease, Swedish Mammography Cohort, and Women’s Lifestyle and Health Study were not included in this stratum, because few case patients (n < 15) drank at least one alcoholic beverage a day.

‡‡‡ The Japan Public Health Center-Based Prospective Study I was not included in this stratum, because few case patients (n < 15) drank alcohol.

§§§ The Japan Public Health Center-Base Prospective Study I and CLUE II: Campaign Against Cancer and Heart Disease were not included in this stratum because few case patients (n < 15) drank at least one alcoholic beverage a day.

|||||| The New York University Women’s Health Study and Swedish Mammography Cohort were not included in these analyses because smoking status was not measured at baseline.

¶¶¶ The Beta-Carotene and Retinol Efficacy Trial, CLUE II: Campaign Against Cancer and Heart Disease, Japan Public Health Center-Based Prospective Study I, and the Prospective Study on Hormones, Diet and Breast Cancer were not included in this stratum, because few case patients (n <15) were past smokers.

### CLUE II: Campaign Against Cancer and Heart Disease, Japan Public Health Center-Based Prospective Study I and Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial were not included in this stratum because few case patients (n < 15) were current smokers.

**** Japan Public Health Center-Based Prospective Study I was not included in this stratum because few case patients (n < 15) were past or current smokers.

†††† The Beta-Carotene and Retinol Efficacy Trial, Canadian National Breast Screening Study, Japan Public Health Center-Based Prospective Study I, Nurses’ Health Study (a), the Prospective Study on Hormones, Diet and Breast Cancer, and Women’s Lifestyle and Health Study were not included in this stratum, because few case patients (n < 15) were older than 64 years.

‡‡‡‡ The Nurses’ Health Study II, Women’s Lifestyle and Health Study, Japan Public Health Center-Based Prospective Study I, and The Canadian National Breast Screening Study were not included in this stratum, because few case patients (n < 15) were at least 64 years.

§§§§ The CLUE II: Campaign Against Cancer and Heart Disease and Melbourne Collaborative Cohort Study were not included in these analyses because this variable was not measured.

|||||||| The Beta-Carotene and Retinol Efficacy Trial, Japan Public Health Center-Based Prospective Study I, and the Prospective Study on Hormones, Diet and Breast Cancer were not included in this stratum because few case patients (n <15) had a family history of breast cancer.

¶¶¶¶ The Japan Public Health Center-Based Prospective Study I was not included in this stratum because few case patients (n < 15) had a family history of breast cancer.

#### The Canadian National Breast Screening Study was not included in these analyses because this variable was not measured.

***** The Black Women’s Health Study and The Japan Public Health Center-Based Prospective Study I were not included in this stratum because none of the participants were white.

††††† The Beta-Carotene and Retinol Efficacy Trial, The CLUE II: Campaign Against Cancer, The Japan Public Health Center-Based Prospective Study I, Melbourne Collaborative Cohort Study, Netherlands Cohort Study, Prospective Study on Hormones, Diet and Breast Cancer, Swedish Mammography Cohort, and Women’s Lifestyle and Health Study were not included in this stratum because few case patients (n < 15) were African-American.

‡‡‡‡‡ The Black Women’s Health Study, The Beta-Carotene and Retinol Efficacy Trial, The CLUE II: Campaign Against Cancer, Iowa Women’s Health Study, Melbourne Collaborative Cohort Study, Netherlands Cohort Study, Prospective Study on Hormones, Diet and Breast Cancer, Swedish Mammography Cohort, and Women’s Lifestyle and Health Study were not included in this stratum because few case patients (n < 15) were Hispanic, Asian, or other races.

§§§§§ The Breast Cancer Detection Demonstration Project Follow-up Study, The Beta-Carotene and Retinol Efficacy Trial, The CLUE II: Campaign Against Cancer, Cancer Prevention Study II Nutrition Cohort, California Teachers Study, Canadian National Breast Screening Study, Iowa Women’s Health Study, The Japan Public Health Center-Based Prospective Study I, Melbourne Collaborative Cohort Study, Nurses’ Health Study(a), Nurses’ Health Study II, Netherlands Cohort Study, Prospective Study on Hormones, Diet and Breast Cancer, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, Swedish Mammography Cohort, Women’s Health Study, and Women’s Lifestyle and Health Study were not included in this stratum because few case patients (n < 15) were African-American.

|||||||||| The Breast Cancer Detection Demonstration Project Follow-up Study, the Black Women’s Health Study, The Beta-Carotene and Retinol Efficacy Trial, The CLUE II: Campaign Against Cancer, Cancer Prevention Study II Nutrition Cohort, Iowa Women’s Health Study, Melbourne Collaborative Cohort Study, Nurses’ Health Study II, Netherlands Cohort Study, New York University Women’s Health Study, Prospective Study on Hormones, Diet and Breast Cancer, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, Swedish Mammography Cohort, Women’s Health Study, and Women’s Lifestyle and Health Study were not included in this stratum because few case patients (n < 15) were Hispanic, Asian, or other races.

Discussion

In this pooled analysis of 993 466 women from 20 prospective studies, we observed modest statistically significant inverse associations between total fruit and vegetable and total vegetable consumption and risk of ER⁻ breast cancer. For total fruit consumption, we found a weaker, non-statistically significant inverse association with risk of ER⁻ breast cancer. These results, which included only three previously published reports (5,7,68), were not statistically significantly heterogenous across studies and were not modified by other breast cancer risk factors, including menopausal status and age. Total fruit and vegetable, total fruit, and total vegetable intakes were observed to have non-statistically significant associations with risk of ER⁺ breast cancer.

We observed null associations between total fruit and vegetable, total vegetable, and total fruit intakes and risk of overall breast cancer. A recent meta-analysis of 14 cohort studies (4) found that high vs low fruit and vegetable intake was inversely associated with a statistically significant 11% lower risk of overall breast cancer. Based on more than 15 cohort and 40 case–control studies published through 2007, the World Cancer Research Fund/American Institute for Cancer Research concluded that the evidence for an association between fruit and vegetable consumption and breast cancer risk was limited (69). However, our analyses have demonstrated that examining associations with only total breast cancer may mask specific relationships for less common subtypes such as ER⁻ breast cancer.

We observed a statistically significant inverse association for total vegetable consumption with risk of ER⁻ breast cancer. To our knowledge, among the few cohort studies (5,7,68) that have previously examined associations between fruit and vegetable consumption and risk of breast cancer by hormone receptor status, only the Danish Diet, Cancer and Health Study (7) was not included in our analyses. That study reported that women with total fruit and vegetable intake greater than 570g/day had half the risk of ER⁻ breast cancer as women with intakes less than 255g/day (7), although the separate associations for fruit intake and vegetable intake were not statistically significant. No statistically significant associations were observed for risk of ER⁺ breast cancer, which is consistent with our findings (7). On the other hand, case–control studies that evaluated fruit and vegetable consumption in relation to breast cancer risk by hormone receptor status (70–72) have found non-statistically significant inverse associations for risk of ER⁻ breast cancer and statistically significant inverse associations between intakes of total fruit and vegetables (71), total fruit (70), and specific vegetables (72) and risk of ER⁺ breast cancer, with 31% to 35% reduced risks of ER⁺ breast cancer observed when comparing the highest vs lowest intakes. In contrast, in the Women’s Healthy Eating and Living randomized trial (73), the risk of breast cancer recurrence was not statistically significantly different between the intervention group, whose goal was to increase intake of fruit, vegetables, and fiber and reduce intake of total fat, and the control group, even when stratified by hormone receptor status of the initial tumor. However, the results of that study (73) might not be directly comparable with ours because it focused on breast cancer recurrence and not breast cancer incidence.

High vegetable consumption may be associated with risk of ER⁻ breast cancer but not with ER⁺ breast cancer, although the biological mechanism is not clear yet. There is substantial evidence that the etiology of breast cancer varies by ER status (1,3,74,75). The incidence curves increase continuously with age for ER⁺ breast tumors but plateau after menopause for ER⁻ breast cancer (2). Compared with ER⁺ tumors, ER⁻ breast cancers are more frequently diagnosed as larger and more rapidly proliferating tumors than ER⁺ tumors (76), have lower 5-year survival rates (77), and are more common in African American and Asian women (78). Most of all, ER⁻ tumors are less dependent on estrogen levels (1). The known risk factors for breast cancer overall, such as nulliparity, delayed childbearing, early menarche, and postmenopausal obesity, have been found to be more strongly associated with risk of ER⁺ than ER^- tumors (1,3). The beneficial effect of bioactive compounds in vegetables may be more detectable in preventing the less hormonally dependent ER⁻ tumors than ER⁺ breast cancer. Epidermal growth factor receptor tends to be overexpressed in ER⁻ breast tumors compared with ER⁺ breast tumors. This overexpression triggers nuclear factor-kappaB, which controls the transcription of DNA that is involved in immune responses (79). Furthermore, the cell cycle regulators that are overexpressed differ between ER⁻ and ER⁺ breast cancers. Cyclin E is overexpressed in ER⁻ breast cancer, but cyclin D is overexpressed in ER⁺ breast cancer (80,81). Phytochemicals found in vegetables have been suggested to reduce the level of epidermal growth factor receptor (82), nuclear factor-kappaB (83,84), and cyclin E (85), which may, in turn, reduce the risk of developing ER⁻ breast cancer.

We only observed a non-statistically significant inverse association between total fruit intake and risk of ER⁻ breast cancer. The reason why the association for vegetable consumption is stronger than that observed for fruit consumption is unclear. It may be that bioactive constituents that are more concentrated in commonly consumed vegetables than in fruits are more effective in preventing ER⁻ breast cancer. For example, intakes of alpha-carotene, beta-carotene, and lutein/zeaxanthin, which are especially abundant in vegetables (86), were each inversely associated with risk of ER⁻ breast cancer in a large pooled analysis that used the same cohorts as the current pooled analysis. However, only a non-statistically significant association was observed for beta-cryptoxanthin, which is abundant in fruits.

The main strength of this pooled analysis is that the large sample size enabled us to examine associations separately with relatively high precision for breast cancer subtypes defined by hormone receptor status and investigate further whether these associations were modified by menopausal status, age at diagnosis, or other accepted breast cancer risk factors. In addition, median fruit and vegetable intake varied threefold to fourfold across the 20 cohorts, which provided ample variation in the exposure, thereby minimizing the potential to miss an association due to homogenous dietary habits. With ample heterogeneity in the types of vegetables and fruits consumed across studies, we were able to analyze a variety of fruit and vegetable groups, ranging from comprehensive groupings to specific foods. Further, exposures and covariables were coded in the same manner, enabling more equal comparisons across studies than possible in a meta-analysis (11).

Limitations of our study include the between-studies variation in the dietary assessment methods, assessment of confounding factors, and measurement of hormone receptor status. To minimize the influence of the differences in dietary assessment methods, we categorized intake using both study-specific quintiles, which rank individuals according to their relative intake within a study, and absolute intake cut points, which take advantage of the variation in intake levels across studies. The interpretation of the results was the same in both analyses, adding confidence in these findings. Although the way confounders were measured varied across studies, the age-adjusted results were not materially different from the multivariable-adjusted results presented, suggesting that misclassification in the assessment of confounding variables may not have had a strong influence on the observed associations. The information on hormone receptor status was not available for 4% to 60% of case patients across studies. However, the distribution of the confounding variables and fruit and vegetable intake were similar between case patients with and without hormone receptor status information. Further, the results for total fruit and vegetable, total fruit, and total vegetable intakes and risk of overall breast cancer were similar when all case patients were analyzed or when we limited our case definition to those case patients with ER data.

Another limitation was having only a single measurement of fruit and vegetable consumption at baseline. Furthermore, breast cancer epidemiology suggests that the natural history of breast cancer has a long duration (87). Therefore, collecting dietary information only at baseline may have attenuated our observed results, particularly if diet earlier in life is most important. In addition, we cannot rule out misclassification in estimated fruit and vegetable intake, which may have attenuated the associations observed. Finally, because we evaluated associations with 15 specific fruit and vegetable items, differences observed between breast cancer subtypes for specific fruit and vegetable items could be because of chance, and these results should be interpreted with caution.

In conclusion, this large pooled analysis of prospectively collected data provides compelling support for an association between high vegetable and fruit consumption, especially vegetable consumption, and reduced risk of ER⁻ breast cancer. Our results support a beneficial effect of overall fruit and vegetable consumption rather than consumption of a few specific fruits and vegetables because associations were observed for several botanical families and several specific fruits and vegetables. In addition, when we controlled for several potential bioactive constituents concentrated in fruits and vegetables in the model, the inverse association for total vegetable consumption and risk of ER⁻ breast cancer remained. These findings support the value of examining etiologic factors in relation to breast cancer characterized by hormone receptor status in large pooled analyses because modest associations with less common breast cancer subtypes may have been missed in smaller studies. These analyses make an important contribution to our understanding of how fruit and vegetable consumption is associated with breast cancer etiology, particularly ER⁻ tumors, given the paucity of published prospective studies examining ER⁻ breast cancer (5,7,68), a subtype with a poorer prognosis than ER⁺ breast tumors, that occurs preferentially in younger women, and whose etiology is relatively unknown.

Funding

This work was supported by grants from National Institute of Health (CA55075 to WCW) and the Breast Cancer Research Foundation (to WCW) and the fellowship from Samsung Scholarship (to SJ).

All authors contributed to these pooled analyses and read and approved the final manuscript.

The sponsor had no role in the design, data collection, data analyses, interpretation of the results, preparation of the manuscript or the decision to submit the manuscript for publication.

We thank Shiaw-Shyuan Yaun, Tao Hou, and Ruifeng Li for their invaluable contributions for data management and statistical support.