Adolescent and adult soy food intake and breast cancer risk: results from the Shanghai Women's Health Study^,

Abstract

Background: Soy food is a rich source of isoflavones—a class of phytoestrogens that has both antiestrogenic and anticarcinogenic properties.

Objective: The objective was to evaluate the association of adolescent and adult soy food intake with breast cancer risk in a cohort of 73,223 Chinese women who participated in the Shanghai Women's Health Study.

Design: A validated food-frequency questionnaire was used to assess usual dietary intake during adulthood and adolescence. After a mean follow-up of 7.4 y, 592 incident cases of breast cancer were identified for longitudinal analyses by using Cox regressions.

Results: Adult soy food consumption, measured either by soy protein or isoflavone intake, was inversely associated with the risk of premenopausal breast cancer, and the association was highly statistically significant (P for trend < 0.001). The multivariate-adjusted relative risks (RRs) for the upper intake quintile compared with the lowest quintile were 0.41 (95% CI: 0.25, 0.70) for soy protein intake and 0.44 (95% CI: 0.26, 0.73) for isoflavone intake. High intake of soy foods during adolescence was also associated with a reduced risk of premenopausal breast cancer (RR: 0.57; 95% CI: 0.34, 0.97). Women who consumed a high amount of soy foods consistently during adolescence and adulthood had a substantially reduced risk of breast cancer. No significant association with soy food consumption was found for postmenopausal breast cancer.

Conclusion: This large, population-based, prospective cohort study provides strong evidence of a protective effect of soy food intake against premenopausal breast cancer.

INTRODUCTION

Ecological and migration studies have suggested that the marked international variation in breast cancer incidence is largely due to environmental factors, such as reproductive patterns and dietary habits (1). The largely null association found for dietary fat intake and breast cancer risk (2) has shifted the research interest to other dietary components, such as soy foods and phytoestrogens (3). Isoflavones are a class of phytoestrogens abundant in soy foods that structurally and functionally resemble mammalian estrogens and compete with endogenous estrogens for estrogen receptors (4). Several experimental studies have shown that both soy and isoflavones may protect against hormone-related cancers (4, 5). In addition to their antiestrogenic effect, soy and isoflavones have been shown to have other cancer-inhibitory properties, including the inhibition of inflammation, angiogenesis, and cell proliferation and to stimulate the production of sex hormone–binding globulin (SHBG) (5).

A recent meta-analysis suggested an overall small, yet statistically significant, inverse association of soy intake with breast cancer risk (6). The association appeared to be somewhat stronger among premenopausal women than among postmenopausal women. However, substantial heterogeneity was found across the 18 studies included in the meta-analysis (6). Additional analyses of published studies, mostly case-control studies, showed that the inverse association between soy food intake and breast cancer risk was found in 6 of the 8 studies conducted in Asian women but in only 1 of the 11 studies conducted in white women (7). Soy food intake in most white populations, however, remains very low; thus, informative studies would have to be conducted in Chinese and in other Asian women who consume a much higher level of soy foods than their white counterparts. However, 7 of the 8 Asian studies included in the meta-analysis used a case-control study design, which inevitably suffer from selection and recall biases (7). The only cohort study included in the meta-analysis was conducted in Japan and reported a protective association with breast cancer risk (8). This finding, however, was contradictory to that reported very recently from another Japanese cohort study (9). Given the small sample size of these 2 cohort studies and their inconsistent findings, additional research is warranted to clarify the association. In 2001 we reported from a case-control study that soy food intake during adolescence was associated with a reduced risk of breast cancer (10). This inverse association was also reported in a subsequent case-control study conducted in Asian American women (11). However, to date, no prospective cohort study has validated these findings that may have significant implications in breast cancer prevention. In this study, we investigated the association of soy food intake with breast cancer risk using data from a prospective cohort study with a focus on evaluating the joint effect of soy food intake in adolescents and adults.

SUBJECTS AND METHODS

Participants

Details of the study design and cohort characteristics for the Shanghai Women's Health Study (SWHS) have been described elsewhere (12, 13). Briefly, between 1996 and 2000, 74,942 women aged 40–70 y were recruited from 80,891 eligible women who resided in 7 urban communities of Shanghai with a participation rate of 92.6%. Of the 5949 nonparticipants, 2407 refused to participate (3.0%), 2073 were absent during the study period (2.6%), and 1469 (1.8%) were excluded for other miscellaneous reasons. All participants were interviewed in person by trained interviewers using a structured questionnaire. The questionnaire included questions on sociodemographic factors, diet and lifestyle habits, menstrual and reproductive history, hormone use, and medical history. Anthropometric measurements, including current weight, height, and circumferences of the waist and hips, were also taken by using a standard protocol. The study was approved by the relevant institutional review boards for human research, and written informed consent was obtained from all participants.

Dietary assessment

A validated quantitative food-frequency questionnaire (FFQ) was used to assess usual dietary intake at the baseline survey and again at the first follow-up survey, which was conducted ≈2–3 y after the baseline survey. The second FFQ survey was completed for 92% of cohort members who were alive at the time of the survey. The FFQ covered virtually all soy foods consumed in urban Shanghai, including soy milk, tofu, soy products other than tofu, dried soybeans, soybean sprouts, and fresh soybeans (12, 13). During the in-person interview, each participant was first asked how often, on average, during the 12 mo preceding the interview she had consumed a specific food or food group (the possible responses were daily, weekly, monthly, yearly, or never), followed by a question on the amount consumed per unit of time. For seasonal foods such as fresh soybeans, participants were asked to describe their consumption during each season in which the food was available. Energy and nutrient intakes, including soy protein and isoflavone intakes, were calculated by multiplying the amount of the food consumed by the nutrient content per gram of the food, as obtained from the Chinese Food Composition Tables (14). The correlation coefficient was 0.40 for usual soy protein intake assessed at baseline compared with that assessed at the second FFQ survey. The FFQ was validated against the averages of multiple 24-h dietary recalls in a validation study including ≈200 women (13). The correlation coefficients between intakes derived from the FFQ and the averages of multiple 24-h recalls were 0.59–0.66 for macronutrients, 0.41–0.59 for micronutrients, 0.41 for soy food, and 0.41–0.66 for other major food groups (12). Dietary intake during adolescence (between the ages of 13 and 15 y) was ascertained by using a brief FFQ, including 19 raw food items or groups, and the following soy food items were included: 1) soy milk, 2) tofu and other soy products, 3) dry soybeans, and 4) fresh legumes, including soy beans. Information on the intake of these soy food items was used to derive the intakes of soy protein and soy isoflavones during adolescence. A similar questionnaire was used in a previous case-control study of breast cancer in Shanghai (10). The correlation coefficient was 0.25 (P < 0.001) between soy protein intake during adulthood and soy protein intake during adolescence.

Ascertainment of breast cancer cases

The cohort is followed by a combination of active biennial surveys and periodic linkage with a database maintained by the Shanghai Cancer Registry and the Shanghai death certificate registry. The response rates for the first (2000–2002), second (2002–2004), and third (2004–2007) in-person follow-up surveys were 99.8%, 98.7%, and 96.7%, respectively. Annual record linkage of cohort member information with the cancer and death certificate registries is conducted to ensure a timely and complete ascertainment of new cancer cases and all deaths in the study cohort. All possible matches are checked manually and verified through home visits. Copies of medical charts from hospitals where cancers were diagnosed are obtained to verify the diagnosis and to collect detailed information on the pathologic characteristics of tumors.

Statistical analysis

For this study, we excluded women with a history of cancer (n = 1576) at baseline, women who reported extreme total energy intake (<500 or ≥3500 kcal/d, n = 124), and women who were lost to follow-up (n = 8) shortly after recruitment, which resulted in a total of 73,225 women for the present study.

Two approaches were used to define usual soy food intake during adulthood. The first approach simply used the intake assessed at baseline, whereas the other used the average intake derived from 2 FFQs, administered at baseline and the first follow-up survey. The latter approach, also termed as the cumulative average method, has been reported to improve the assessment of usual dietary intake and is now commonly used in cohort studies that have data from repeat dietary assessments (15–18). Some women developed cancer, diabetes, myocardial infarction, or stroke during the period between the baseline and first follow-up surveys. It is possible that some of these women may have modified their dietary habits after disease diagnosis; therefore, for women who had any of these outcomes during this period, information from the baseline FFQ was used to define usual soy food intake.

Study participants were divided into quintiles according to their soy intake level. The lowest quintile served as the reference group. Relative risks (RRs) and 95% CIs associated with soy intake were estimated by using the Cox proportional hazards regression model (19). Entry time was defined as age at enrollment, and exit time was defined as age at cancer diagnosis or age at censoring due to death or loss to follow-up. The date of last follow-up was set as 31 December 2005 for study participants whose last in-person contact occurred before 31 December 2005, 6 mo before the most recent record linkage for the present version of the data set, to allow for delay in records processing. To control for confounding by age or birth cohort or any possible interaction between the 2, we used age as the time scale in the proportional hazards regression model and stratified the model on birth cohort (in 5-y intervals) (20). Covariates included in the model were education, physical activity, age at first live birth, body mass index, season of recruitment, family history of breast cancer, and total energy intake. The linear trend was evaluated by modeling categorical intake variables as ordinal variables in the model.

To evaluate any modifying effect of menopause on soy intake, the analyses were further stratified by menopausal status. Information on menopausal status was updated during the follow-up surveys and treated as a time-varying variable in the analyses. In addition, the combined effect of soy intake during adolescence and adulthood was evaluated. For this analysis, soy food consumption was measured by soy protein intake and categorized by tertile distribution to improve the stability of point estimates. The likelihood ratio test was used to evaluate potential multiplicative interactions of 2 study variables by comparing the models with and without the cross product terms of these variables. All statistical tests were based on 2-sided probability. Statistical analyses were carried out by using SAS version 9.1 (SAS Institute, Cary, NC).

RESULTS

Over a mean follow-up time of 7.4 y (540,156 person-years), 594 incident cases of breast cancer were identified. The mean (±SD) age at diagnosis was 52.1 ± 9.06 y. Significant differences in cases and noncases were found for education, usual occupation, and all established breast cancer risk factors evaluated in the study (Table 1). Participants in the higher quintile of soy protein intake were more likely to be older, to have higher body mass indexes and waist-to-hip ratios, and to have higher intakes of total energy, fruit, vegetables, and meat than did women with lower soy intakes (Table 1). No apparent association was found for soy food intake and menstrual and reproductive characteristics. Very few women in this cohort were regular alcohol drinkers (1.9%), cigarette smokers (2.4%), or hormone replacement therapy users (3.9%).

TABLE 1.

Soy food intake by quintile (Q) and by selected demographic characteristics and known risk factors for breast cancer in the Shanghai Women's Health Study¹

				Soy protein intake
	Breast cancer cases (n = 594)	Noncases (n = 72,631)	P value	Q1 (≤4.87)	Q2 (4.88–7.11)	Q3 (7.12–9.48)	Q4 (9.49–12.82)	Q5 (>12.82)	P value
Age at baseline (y)	52.1 ± 9.062	52.0 ± 9.06	0.29	51.5 ± 9.43	51.4 ± 9.03	51.8 ± 8.93	52.3 ± 8.91	53.1 ± 8.87	<0.001
Education (%)3
≤Elementary school	14.2	21.4		22.9	18.8	18.8	19.0	20.5
Middle school	32.4	37.2		19.5	20.3	20.2	20.2	19.8
High school	25.2	27.9		19.1	20.3	20.4	21.5	19.7
≥College	18.2	13.5	<0.001	18.7	20.5	20.5	20.0	20.3	<0.001
Income, whole family (yuan)
<10,000	13.6	16.1		21.2	18.7	18.4	19.5	22.2
10,000–19,999	39.1	38.2		19.4	20.0	20.1	20.1	20.4
20,000–29,999	27.9	28.1		20.3	20.2	20.3	20.3	18.9
≥30,000	19.4	17.5	0.33	19.7	21.0	20.8	19.8	18.8	<0.001
Occupation
Professional/technical	39.5	28.5		18.7	20.3	20.5	20.5	20.0
Clerical/service	20.9	20.8	<0.001	20.8	20.2	19.8	19.3	19.9
Manufacturing/construction	39.5	50.7		20.3	19.8	19.8	20.0	20.1	<0.023
Regular exercise (%)3	67.3	64.9	0.219	15.6	18.0	20.1	21.8	24.5	<0.001
(MET/wk)	12.3 ± 11.70	13.6 ± 14.68	0.209	12.7 ± 12.9	12.3 ± 13.7	13.2 ± 14.8	13.7 ± 14.5	15.4 ± 16.1	<0.001
Age at menarche (y)	14.8 ± 1.81	14.9 ±1.74	<0.01	15.0 ± 1.76	14.9 ± 1.73	14.9 ± 1.71	14.9 ± 1.75	14.9 ± 1.75	<0.001
Age at menopause	49.3 ± 4.52	48.6 ±.34	<0.001	48.4 ± 4.44	48.4 ± 4.32	48.6 ± 4.28	48.7 ± 4.29	48.7 ± 4.29	<0.001
Age at first live birth (y)	26.4 ± 4.08	25.6 ± 4.13	<0.001	25.6 ± 4.11	25.8 ± 4.09	25.7 ± 4.04	25.5 ± 4.16	25.2 ± 4.21	0.393
Nulliparous (%)3	4.2	3.3	0.205	20.1	19.5	18.3	20.2	22.0	0.131
Positive breast cancer family history (%)3	3.5	1.8	<0.01	17.4	19.1	20.2	22.2	21.1	0.006
BMI (kg/m2)	24.3 ± 3.40	24.0 ± 3.42	0.02	23.6 ± 3.46	23.8 ± 3.36	23.9 ± 3.37	24.2 ± 3.38	24.5 ± 3.47	<0.001
Waist-to-hip ratio	0.81 ± 0.05	0.81 ± 0.05	0.09	0.809 ± 0.054	0.810 ± 0.054	0.810 ± 0.053	0.812 ± 0.053	0.815 ± 0.054	<0.001
Total energy intake (kcal/d)	1656 ± 339	1642 ± 346	0.21	1449 ± 300.8	1555 ± 294.3	1630 ± 302.1	1710 ± 315.0	1866 ± 363.1	<0.001
Total fruit and vegetable intake (g/d)	526 ± 250	506 ± 238	0.10	403 ± 204.5	460 ± 208.8	498 ± 215.3	542 ± 225.8	628 ± 269.3	<0.001
Total meat intake (g/d)	90.7 ± 42.4	91.1 ± 45.4	0.69	75.1 ± 40.3	85.4 ± 40.5	91.3 ± 41.7	97.6 ± 44.7	105.8 ± 52.4	<0.001

¹MET, metabolic equivalent.

²Mean ± SD (all such values).

³Frequency.

The association of breast cancer with usual intake of soy protein or isoflavones assessed using either the baseline FFQ (baseline) or the average of the 2 FFQs administered at baseline and the first follow-up survey (average) is shown in Table 2. Overall, soy protein and isoflavone intakes were inversely associated with the risk of breast cancer, although the trend tests were not statistically significant. Analyses stratified by menopausal status showed an inverse dose-response association of soy protein or isoflavone intake with breast cancer risk among premenopausal women by using data from either the baseline FFQ or the average of the 2 FFQs, although the latter approach showed a stronger association. No apparent association was observed between soy food intake and postmenopausal breast cancer. Excluding women who reported a change in eating habits during the 5 y preceding study recruitment did not change the pattern of association (data not shown in table). Further adjustment for other dietary factors, such as the consumption of red meat, fruit, vegetables, and carotenoids, did not materially change the soy and breast cancer association (data not shown).

TABLE 2.

Soy food intake and breast cancer risk by quintile (Q) in the Shanghai Women's Health Study

	RR (95% CI)1
	Q1	Q2	Q3	Q4	Q5	P for trend
All women (594 breast cancer cases)
Soy protein (baseline)	1.00	0.97 (0.75, 1.25)	0.88 (0.68, 1.15)	1.05 (0.81, 1.35)	0.86 (0.65, 1.13)	0.499
Isoflavones (baseline)	1.00	0.92 (0.71, 1.19)	0.96 (0.75, 1.24)	1.00 (0.77, 1.29)	0.86 (0.65, 1.13)	0.514
Soy protein (average)	1.00	1.00 (0.78, 1.29)	0.86 (0.67, 1.12)	0.85 (0.65, 1.11)	0.89 (0.66, 1.15)	0.158
Isoflavones (average)	1.00	0.87 (0.67, 1.12)	1.02 (0.80, 1.30)	0.77 (0.58, 1.00)	0.81 (0.61, 1.07)	0.091
Premenopausal women (305 breast cancer cases)2
Soy protein (baseline)	1.00	0.97 (0.70, 1.35)	0.83 (0.58, 1.18)	0.93 (0.65, 1.31)	0.68 (0.45, 1.02)	0.093
Isoflavones (baseline)	1.00	0.85 (0.61, 1.18)	0.89 (0.64, 1.25)	0.75 (0.53, 1.08)	0.65 (0.43, 0.97)	0.034
Soy protein (average)	1.00	0.79 (0.54, 1.15)	0.59 (0.39, 0.91)	0.62 (0.40, 0.95)	0.41 (0.25, 0.70)	<0.001
Isoflavones (average)	1.00	0.66 (0.44, 0.98)	0.80 (0.55, 1.18)	0.48 (0.30, 0.77)	0.44 (0.26, 0.73)	<0.001
Postmenopausal women (289 breast cancer cases)3
Soy protein (baseline)	1.00	0.98 (0.66, 1.47)	0.98 (0.66, 1.46)	1.23 (0.84, 1.81)	1.08 (0.72, 1.61)	0.396
Isoflavones (baseline)	1.00	1.04 (0.69, 1.56)	1.11 (0.74, 1.66)	1.38 (0.94, 2.02)	1.16 (0.78, 1.74)	0.191
Soy protein (average)	1.00	1.21 (0.87, 1.69)	1.11 (0.80, 1.56)	1.07 (0.76, 1.50)	1.22 (0.87, 1.71)	0.504
Isoflavones (average)	1.00	1.05 (0.76, 1.47)	1.22 (0.88, 1.68)	1.00 (0.71, 1.40)	1.09 (0.78, 1.52)	0.800

¹Relative risks (RRs) and 95% CIs were derived from Cox regression analyses, compared with the lowest quintile, and adjusted for age, education, physical activity, age at first live birth, BMI, season of recruitment, family history of breast cancer, and total energy intake. Median values (and cutoffs) for the lowest to the highest quintile of soy protein intake (g/d) are as follows: 3.53 (≤4.87), 6.03 (4.88, 7.11), 8.23 (7.12, 9.48), 10.93 (9.49, 12.83), and 16.02 (≥12.84). The corresponding numbers for soy isoflavones (mg/d) are as follows: 11.23 (≤15.93), 20.06 (15.94, 23.88), 27.93 (23.89, 32.43), 37.57 (32.44, 44.23), and 54.97 (≥44.24).

²Mean age at cancer diagnosis was 47.9 y.

³Mean age at cancer diagnosis was 58.6 y.

A similar association pattern was observed between soy food intake during adolescence and breast cancer risk (Table 3). In particular, a 43% reduced risk (95% CI: 0.34, 0.97) of premenopausal breast cancer was found among those whose soy food intake during adolescence was in the highest intake group. However, the test of linear trend was only of borderline significance (P = 0.061). In postmenopausal women, a slight positive association was observed between breast cancer risk and adolescent soy food intake. When soy intake during adulthood and adolescence were both included in the same model, little change was seen in the risk estimates associated with adult soy intake, whereas the inverse association with adolescent soy intake was attenuated (data not shown). Further analysis of the combined effect of soy protein intake in adolescence and in adulthood on premenopausal breast cancer risk was carried out, and the results are presented in Table 4. High soy food intake during adolescence or adulthood alone was each related to a reduced risk of premenopausal breast cancer. Women in the highest tertile of soy protein intake in both adolescence and adulthood had the greatest decrease in relative risk compared with women in the lowest tertile of soy protein intake at both of these time points (RR: 0.41; 95% CI: 0.22, 0.75).

TABLE 3.

Association between soy food intake during adolescence and breast cancer risk by quintile (Q) in the Shanghai Women's Health Study

	RR (95% CI)1
Groups	Q1	Q2	Q3	Q4	Q5	P for trend
All women (594 breast cancer cases)
Soy protein	1.00	0.92 (0.71, 1.19)	1.07 (0.83, 1.38)	0.98 (0.76, 1.27)	0.97 (0.75, 1.27)	0.970
Isoflavones	1.00	1.07 (0.82, 1.39)	1.04 (0.80, 1.36)	1.12 (0.86, 1.45)	1.19 (0.91, 1.55)	0.196
Premenopausal (305 breast cancer cases)
Soy protein	1.00	0.78 (0.52, 1.18)	0.87 (0.58, 1.31)	0.77 (0.50, 1.18)	0.57 (0.34, 0.97)	0.061
Isoflavones	1.00	0.91 (0.59, 1.40)	0.76 (0.48, 1.18)	0.79 (0.51, 1.24)	0.89 (0.57, 1.40)	0.452
Postmenopausal (289 breast cancer cases)
Soy protein	1.00	1.01 (0.72, 1.41)	1.22 (0.88, 1.68)	1.13 (0.82, 1.56)	1.20 (0.87, 1.65)	0.200
Isoflavones	1.00	1.17 (0.83, 1.64)	1.23 (0.88, 1.72)	1.33 (0.96, 1.85)	1.38 (1.00, 1.91)	0.038

¹Relative risks (RRs) and 95% CIs were derived from Cox regression analyses, compared with the lowest quintile, and adjusted for age, education, physical activity, age at first live birth, BMI, season of recruitment, family history of breast cancer, total energy intake, and total fruit and vegetable intakes during adolescence. Median values (and cutoffs) for the lowest to the highest quintile of soy protein intake (g/d) are as follows: 1.73 (≤2.76), 3.76 (2.77, 4.70), 5.81 (4.71, 6.95), 8.86 (6.96, 11.32), and 15.16 (≥11.33). The corresponding numbers for soy isoflavones (mg/d) are as follows: 4.31 (≤7.34), 10.04 (7.35, 12.82), 15.99 (12.83, 19.43), 24.11 (19.44, 31.28), and 42.26 (≥31.28).

TABLE 4.

Joint effect of soy food intake during adulthood and adolescence on breast cancer risk among premenopausal women in the Shanghai Women's Health Study¹

	Soy protein intake in adulthood2
	<6.39		6.39–10.40		≥10.41
Soy protein intake in adolescence	n3	RR (95% CI)4	n3	RR (95% CI)4	n3	RR (95% CI)4
≤4.00	91	1.00 (reference)	56	0.57 (0.32, 0.99)	32	0.57 (0.30, 1.08)
4.01–8.04	76	0.96 (0.62, 1.49)	64	0.56 (0.33, 0.94)	70	0.59 (0.34, 1.03)
≥8.05	40	0.62 (0.33, 1.19)	74	0.67 (0.39, 1.15)	91	0.41 (0.22, 0.75)

¹Low, middle, and high intakes were defined after division by tertile of each intake of soy protein during adolescence and adulthood. P for interaction = 0.017.

²Median values for the lowest to the highest tertile of soy protein intake (g/d) are as follows: 4.46, 8.23, and 13.66 during adulthood and 2.43, 5.75, and 12.17 during adolescence.

³Number of breast cancer cases.

⁴Relative risks (RRs) and 95% CIs were derived from Cox regression analyses and adjusted for age, education, physical activity, age at first live birth, BMI, season of recruitment, family history of breast cancer, total energy intake, and total fruit and vegetable intakes during adolescence and at recruitment.

DISCUSSION

In this population-based cohort study, we found that high soy food intake was associated with a reduced risk of breast cancer among premenopausal women. Women who consumed a high level of soy food consistently during adolescence and adulthood had a substantially reduced risk of breast cancer. These results suggest that soy food intake may reduce the risk of breast cancer, and this inverse association may explain some of the low incidence of breast cancer observed in China and several other Asian countries.

Menopause is a milestone in a woman's life, resulting in dramatic changes in levels and production sites of estrogen. In premenopausal women, estrogen is predominantly produced in the ovaries, and the concentrations are high. After menopause, endogenous estrogen is primarily produced in adipose tissue (21–23), and concentrations of estrogen are substantially lower than in premenopausal women. Soy isoflavones, a group of phytoestrogens that are structurally similar to mammalian estrogen 17β-estradiol, have been shown to exert agonist or antagonist effects on various estrogen target tissues (3–5). Isoflavones, by competing with estrogen in binding estrogen receptors, may exert antiestrogenic effects in high estrogen environments, such as that in premenopausal women, reducing the risk of breast cancer. Several intervention studies have shown that a high dietary intake of soy products decreased serum estradiol concentrations in premenopausal women (24–26). Another study found that soy consumption increased menstrual cycle length and reduced the level of luteinizing hormone and follicle-stimulating hormone, a change that may reduce the risk of breast cancer (27). On the other hand, under a low estrogen environment, such as that among postmenopausal Chinese women who have a much lower estrogen concentration than their counterparts in Western society, soy phytoestrogens may exert their estrogenic effect, compromising other anticancer properties of soy foods.

Isoflavones have also been shown to have many other biological effects that potentially reduce breast cancer risk, including the inhibition of epidermal growth factor receptor tyrosine kinase activity, inhibition of topoisomerase II activity, arrest of cell cycle progression at the G2-M transition, induction of apoptosis, antioxidant properties, modification of eicosanoid metabolism, and inhibition of angiogenesis (5). Some studies have also shown increased ratios of urinary 2-hydroxy to 16α-hydroxy and 2-hydroxy to 4-hydroxy estrogens associated with soy intake (28, 29), which suggests a reduced formation of genotoxic and potentially carcinogenic estrogen metabolites.

Although cumulative evidence strongly supports cancer-inhibitory effects of soy and certain soy constituents, some studies also showed adverse effects of these groups of foods. In a mouse study, soy intake was shown to increase spontaneous mammary gland tumors (30). In another study, a 2-wk supplement of soy protein given to women with benign or malignant breast disease was found to have estrogenic effects in breast tissue (31). It is possible that the overall effect of soy foods and soy constituents on breast carcinogenesis may depend on other factors, such as an individual's endogenous estrogen concentration, as suggested in our study for an inverse association seen in premenopausal women but not in postmenopausal women. A recent meta-analysis of 18 epidemiologic studies (6) found an overall inverse association between soy intake and breast cancer risk. Among the 10 studies that performed analyses stratified by menopausal status, soy consumption was more strongly associated with breast cancer in premenopausal women than in postmenopausal women. A substantial heterogeneity among postmenopausal women, however, was suggested (6). A recent report from a cohort of 37,643 British women that investigated the association of a vegetarian diet and isoflavone intake with breast cancer risk found no evidence of an association between dietary isoflavone intake and breast cancer risk among either pre- or postmenopausal women (32). The most recent report from the Japan Collaborative Cohort study also showed no association between intake of tofu, boiled beans, or miso soup and the risk of breast cancer for women who were postmenopausal at baseline recruitment (9). This is contrary to an earlier report from Japan, which showed a protective effect of soy on breast cancer risk among postmenopausal women (8), and to a recent report from a nested case-control study, in which an inverse association between plasma isoflavone concentrations and breast cancer risk was observed (33). The sample size for both studies, however, was small (8, 33).

In vivo studies have consistently shown that prepubertal or pubertal exposure to genistein either via injections or feeding reduces the incidence and/or multiplicity of chemically induced mammary tumors in experimental animals (34). To evaluate the effect of prepubertal and pubertal soy exposure on breast cancer risk, we assessed usual soy food intake during the ages of 13 to 15 y, when most of the study participants started puberty. The mean age of menarche was 14.9 y. We previously reported that soy food intake during this period was inversely related to the risk of breast cancer among Chinese women (10). This association was subsequently replicated in a study of Asian American women (11). These results raise the question as to whether soy food intake during adolescence, a critical window for breast development, is the primary mechanism by which soy exerts its cancer-preventive effect on the breast. In our study, despite the fact that a high consumption of soy foods during both adolescence and adulthood was associated with the largest reduction in breast cancer risk; high consumption of soy food in adolescence or adulthood alone was each inversely associated with risk. Whereas laboratory studies have suggested that the mechanisms by which soy intake exerts its effect may differ during adolescence (eg, promotion of terminal bud differentiation) and adulthood (eg, competition with estrogen receptors, stimulation of SHBG syntheses, antioxidative, and antiangiogenesis), the reasons for the inverse association of adolescent soy intake with premenopausal but not with postmenopausal breast cancer are unclear, and further investigations are warranted.

As with any observational study, errors in assessing usual dietary intake were a potential concern in our study. This type of error, however, is likely to be nondifferential between cases and noncases in a prospective cohort study and thus does not explain the inverse association observed in our study. We carefully designed the questionnaire to include virtually all soy foods that are commonly consumed in Shanghai. The validation study, implemented as part of the SWHS, indicates that the FFQ used in the SWHS has good validity and reliability in assessing usual soy food intake in adulthood. In addition to the baseline survey, dietary information was also collected at the first follow-up survey implemented 2–3 y after the baseline survey, which helps to reduce random errors and improves dietary assessment.

Loss to follow-up can introduce selection biases. This, however, was not a major concern in the present study, because the response rates at follow-up are very high. High soy food intake could be related to certain lifestyles that may be associated with reduced risks of breast cancer. However, the inverse association between soy food intake and breast cancer risk remained after a wide range of sociodemographic, lifestyle, and known risk factors for breast cancer were controlled for. The statistical power in our study was low for some subgroup analyses. With an extended follow-up of this cohort and data from additional FFQ surveys, we should be able to re-evaluate the association of soy food intake with breast cancer risk in the future with a larger sample size and improved exposure data.

In summary, in a large prospective cohort study conducted among Chinese women, we found that high soy food intake was inversely associated with premenopausal breast cancer risk. These results are consistent with the cancer-inhibitory effects of soy isoflavones, which suggests that an increase in soy food intake can reduce the risk of breast cancer.