Association of Fruit, Vegetable, and Animal Food Intakes with Breast Cancer Risk Overall and by Molecular Subtype among Vietnamese Women

Breast cancer is the most common cancer and the leading cause of cancer-related deaths among women, with approximately 2.3 million incident cases and 685,00 deaths worldwide in 2020 (1). Breast cancer incidence rates in Asian countries, including Vietnam, where rates were historically low, have been rapidly increasing over the past decades (1). In Vietnam, breast cancer is the most common cancer and the fourth leading cause of cancer-related death among women, with an estimated 21,555 new cases and 9,345 deaths in 2020 (1). Decreases in birth rates, changes in lifestyles, such as an increase in body weight and physical inactivity and adoption of western dietary patterns, as well as improvements in breast cancer screening and awareness, have been suggested as major contributors to the increasing rates of breast cancer in Vietnam (1–4).

A plausible role of dietary factors in the etiology of breast cancer has been extensively investigated (5, 6). It has been suggested that vegetables and fruits, which contain high levels of nutrients and phytochemicals, including fiber, vitamins, minerals, flavonoids, and carotenoids, might act as potential antitumorigenic agents through their activities (7, 8). Conversely, red or processed meats and grilled or roasted meats, which may contain precursors of carcinogens, including heterocyclic aromatic amines, nitrosamines, and polycyclic aromatic hydrocarbons (PAH), have been linked to carcinogenesis (9).

A recent review based on 48 publications, including those from meta- and pooled analyses, reported that high intakes of total meat, red or processed meats, foods with a high glycemic index, and eggs were associated with higher risks of breast cancer, while high intakes of other foods, such as vegetables, citrus fruits, and mushrooms, were inversely associated with breast cancer risk (10). However, the World Cancer Research Fund International still considers the evidence that dietary intake affects breast cancer risk being inconclusive (11). In addition, data on the association between dietary factors with molecular subtypes of breast cancer is limited and inconsistent (12). Molecular subtypes of breast cancer have different therapeutic responses and clinical outcomes (13), and thus, may differ etiologically. However, this hypothesis has not been well investigated.

To our knowledge, no epidemiologic studies have been conducted among Vietnamese women to comprehensively evaluate associations of food groups with the risk of overall and molecular subtype of breast cancer. The current study was conducted to fill this gap.

Data of this study were from the Vietnamese Breast Cancer Study (VBCS). The VBCS recruited 501 newly diagnosed patients with breast cancer and 468 healthy controls aged 18 to 79 years in Hanoi, Vietnam, from July 2017 to June 2018. Patients with breast cancer (clinical diagnosis) were recruited from inpatient surgical units and chemotherapy inpatient and outpatient units of two major cancer hospitals, the Vietnam National Cancer Hospital (Hanoi, Vietnam) and Hanoi Oncology Hospital in North Vietnam (Hanoi, Vietnam). Individuals with a prior history of cancer were not eligible for the study. Controls were selected from healthy relatives, neighbors, or friends of patients with breast cancer in the communities where the cases resided (n = 64) or from healthy female caregivers of cancer patients other than breast cancer in these two hospitals (n = 402).

This study was originally intended to be an age frequency-matched case-control study. However, it adapted into an individually matched study design during implementation. Controls were individually matched to cases on age (±3 years), city, and province of residency. Women with the following conditions were not eligible to serve as controls: (i) have a history of cancer diagnosis; (ii) currently receiving treatment at a health facility or diagnosed with severe disease(s; e.g., myocardial infarction, stroke); (iii) determined by research staff to be physically or mentally incapable of completing the interview; or (iv) unable to provide consent. Of 501 cases and 468 controls enrolled in the study, there were 374 individually matched case-control pairs. All VBCS participants provided written informed consent. The response rates were 93.1% and 97.7%, respectively, for cases and controls. Approvals for human subject research were obtained from the Vietnam National Cancer Institute and Vanderbilt University Medical Center (Nashville, TN).

We excluded participants who were subsequently confirmed to have benign tumors based on pathologic reviews (n = 9) and who were diagnosed at stage 0 (n = 2). We also excluded 28 participants with unreasonably high caloric (>3,500 kcal/day; n = 21) or low caloric (<500 kcal/day; n = 7) intakes (14). Finally, 476 cases and 454 controls, of which there were 352 paired pairs, remained in this study.

Information on participants’ demographics characteristics, reproductive factors, menstrual history, family history of cancer, lifestyle factors, and dietary habits were collected through in-person interviews using a structured questionnaire via the Research Electronic Data Capture (REDCap) mobile application (15) at study enrollment by trained interviewers from the Vietnam National Cancer Institute. The median time interval from breast cancer diagnosis (i.e., surgical pathologic diagnosis) to the date of recruitment was one day [interquartile range (IQR): 17 days]. In addition, clinical features were abstracted by reviewing patients’ medical records, including the expression statuses of estrogen receptor (ER) and progesterone receptor (PR), HER2, and Ki-67. Breast cancer was classified into four major molecular subtypes: (i) luminal A (i.e., ER-positive, PR-positive, HER2-negative, and low Ki-67), (ii) luminal B (i.e., ER- and/or PR-positive, and either HER2-positive or HER2-negative with high Ki-67), (iii) HER2-enriched (i.e., HER2-positive, ER-negative, and PR-negative), and (iv) triple-negative/basal-like (i.e., ER-, PR-, and HER2-negative). Study data were managed using the REDCap data management platform hosted at Vanderbilt University (15).

The VBCS food frequency questionnaire (FFQ) was adapted from a validated, semiquantitative FFQ covering the previous 3-year period (16). The VBCS FFQ consisted of 68 food items/groups, including 9 fruits, 10 fresh vegetables, 4 soybean, and other legume products, 4 staple foods, 9 animal foods, 3 nuts, 3 sweet varieties, 6 bread and rice varieties, and 19 beverage items (i.e., tea, coffee, juice, and soft drinks). Information on frequency (four levels: daily, weekly, monthly, yearly) and the number of servings or kilograms per unit of time for each food/beverage consumed during the previous 5-year period was obtained through face-to-face interviews. Serving size was estimated by using a picture booklet developed in a previous study (17). In this study, fruits and vegetables were categorized into 11 food groups, including common tropical fruits, Asian tropical fruits, citrus fruits, melons, other fruits, dark green vegetables, cruciferous vegetables, fresh legumes (i.e., beans and peas), allium vegetables, tomatoes, melon vegetables, and other vegetables. Meat and fish were classified into four food groups: red meat and organ meats, poultry, freshwater fish, and marine fish. Total energy intake (kilocalories per day) was calculated by the sum of energy from individual food and beverage items/groups (without alcohol consumption) consumed per day (in kilocalories) and estimated based on the 2007 and 2017 Vietnamese food composition tables (18, 19).

Descriptive statistics on patients’ demographic characteristics were computed in percentages for categorical variables and median and IQR for continuous variables. Nonparametric tests (i.e., Mann–Whitney tests) were used to compare continuous variables between cases and controls and χ² tests were used to assess case-control differences in categorical variables. ORs and 95% confidence intervals (CI), derived from multivariable unconditional logistic regression analyses, were used to assess associations of food groups with breast cancer for overall and by molecular subtype (i.e., luminal A, luminal B, HER2-enriched and triple-negative/basal-like). OR was used as an approximation of relative risk. Tertile distributions of controls’ dietary intake of each food group were used to categorize intake variables, with the lowest tertile serving as the reference group. Intake of several food groups, including fruit intake, are not evenly distributed in our study population, so exact tertile categories could not be achieved. Results of two sets of analyses, i.e., minimal adjusted model (adjusted for age and energy intake only, both treated as continuous variables) and multivariable adjusted model, are presented. The latter additionally adjusted for education level (i.e., no formal education/primary school, middle school, high school, college or higher), average annual per capita income [continuous - million Vietnamese dong (VND)], study site (i.e., K3 Hospital, K1 Hospital, Hanoi Oncology Hospital; Hanoi, Vietnam), the recruitment month of year (i.e., January–April, May–August, September–December), family history of cancer first degree (yes vs. no), number of children level (i.e., 0–1, 2, 3, ≥4), menopausal status (postmenopausal vs. premenopausal), body mass index (BMI) level (i.e., <18.5 kg/m², 18.5–22.9 kg/m², 23.0–24.9 kg/m², > 25.0 kg/m²), regular exercise (yes vs. no), intensity of exercise [continuous: metabolic equivalent (MET)-hours/week], frequencies of grilled vegetables (continuous: times/month), frequencies of grilled meat/fish consumption (continuous: times/month), and total energy intake (continuous: kcal/day). The covariates were selected a priori based on their associations with breast cancer risk, according to the literature and our assessments of whether inclusion of covariates affected the associations under the study. Tests for trend were conducted using the median values of each tertile. Analyses were also carried out to evaluate the linear associations between breast cancer and intake of food groups with intake level treated as a continuous variable. Restricted cubic spline regression was applied to evaluate linearity of associations, with four knots specified at 5, 35, 65, and 95 percentiles. Because fruit intake was positively correlated to red and organ meat consumption, and both intakes were associated with breast cancer risk in our study, we further mutually adjusted for their intakes in the analyses to control for confounding effects by including both of them in the regression model during analysis. We also performed analyses stratified by menopausal status, average annual per capita income levels (tertile distributions), and recruitment time. Likelihood ratio tests were used to assess multiplicative interactions between these variables and food group intakes. To evaluate whether the observed risk estimates differed across molecular subtypes of breast cancer, multivariable multinomial logistic regression was performed. Likelihood ratio tests were used to evaluate heterogeneity across molecular subtypes. Sensitivity analyses were performed based on 352 matched pairs of cases and controls, using conditional logistic regression to assess associations of food groups with overall breast cancer risk. Stata 14.0 software package (StataCorp) was employed for analyses. A two-sided P value of less than 0.05 was considered statistically significant.

The mean age of study participants was 50.1 years for breast cancer cases and 49.9 years for controls. Compared with controls, women with breast cancer were similar regarding educational attainment, occupation, average annual per capita income, and residence but less likely to be married. No differences were observed between cases and controls regarding reproductive factors, menstrual history, family history of cancer, and lifestyle factors, except for the number of children and regular exercise. Cases had lower total number of children and fewer regular exercisers than controls (Table 1).

Compared with controls, breast cancer cases tended to have lower consumption of total fruit, types of fruit, total meat, red meat and organ meats, and freshwater fish, and higher intake of marine fish, but were similar regarding the intake of all vegetables and specific groups of vegetables. Intake of grilled vegetables was more frequently consumed, but grilled fish or meat was less frequently consumed among breast cancer cases than controls (Table 2).

Multivariable analyses showed that a high intake of fruit was significantly and inversely associated with the risk of breast cancer. The adjusted ORs and 95% CIs for tertiles 2 to 3 versus tertile 1 were 0.67 (0.47–0.95) and 0.41 (0.27–0.61), respectively (P_trend < 0.001). Analysis by fruit groups showed that high intakes of Asian tropical fruits (i.e., longan, dragon fruit, lychee, guava, persimmon, passion fruit, and rambutan), citrus fruits (i.e., grapefruit, lime, lemon, orange, and tangerines), and other fruits (e.g., berries, stone, apple, avocado, and others) were all inversely associated with the risk of breast cancer. The adjusted ORs and 95% CIs for tertiles 2 to 3 versus tertile 1 for Asian tropical fruit were: 0.63 (0.44–0.89), 0.37 (0.24–0.55), and P_trend < 0.001; for citrus fruits: 0.79 (0.56–1.12), 0.63 (0.43–0.93), and P_trend = 0.03; and for other fruits: 0.68 (0.47–0.98), 0.61 (0.41–0.90), and P_trend = 0.03. A test for nonlinearity was significant for all fruit consumption, indicating their associations with breast cancer did not follow a linear dose-response pattern. There were no significant associations between breast cancer risk and intake of total vegetables, and several major types of food groups (Table 3). Likewise, no significant associations were found for intakes of total meat, total fish, types of meat (i.e., red meat and organ meats, and poultry), and marine fish. The adjusted ORs and 95% CIs for tertiles 2 to 3 versus tertile 1 for the intakes of total vegetables were: 1.19 (0.83–1.71), 0.92 (0.63–1.35), and P_trend = 0.58; total meat: 0.74 (0.51–1.05), 0.72 (0.48–1.08), and P_trend = 0.09; and total fish: 0.97 (0.68–1.38), 0.75 (0.51–1.10), and P_trend = 0.13. We found that a high intake of freshwater fish was inversely associated with breast cancer risk with adjusted ORs and 95% CIs for tertiles 2 to 3 versus tertile 1 of 0.82 (0.59–1.14) and 0.63 (0.42–0.95; P_trend = 0.03). Testing for nonlinear association was significant for red and organ meat intake (Table 4). Sensitivity analyses based on 352 individually matched pairs did not materially change the above-reported associations (Supplementary Tables S1 and S2).

Stratified analyses by menopausal status, income, and recruitment month found that these factors did not modify the diet-breast cancer risk association, with the exception of total fruit and vegetable intake by recruitment month (Supplementary Tables S3–S5). There were indications of possible modification by recruitment month for intake of total fruit (P_interaction = 0.02) and total vegetables (P_interaction = 0.045) where the association patterns for participants recruited during May and August (seasonal harvest of many Asian tropical fruits or wet season in North Vietnam) differed from those recruited in other seasons. (Supplementary Table S5).

In our study, most breast cancer cases were luminal B (44.5%). The percentages of HER2-enriched, luminal A, and triple-negative/basal-like breast cancers were 18.3%, 15.5%, and 9.5%, respectively. The inverse association between breast cancer risk and fruit intake was observed for all molecular subtypes; but was stronger for triple-negative/basal-like (OR_T3vsT1 = 0.20; 95% CI, 0.06–0.64; P_trend = 0.006) than luminal A (OR_T3vsT1 = 0.40; 95% CI, 0.18–0.90; P_trend = 0.03), luminal B (OR_T3vsT1 = 0.43; 95% CI, 0.26–0.73; P_trend = 0.002), and HER2-enriched (OR_T3vsT1 = 0.42; 95% CI, 0.18–0.87; P_trend = 0.02) breast cancer and P_{heterogeneity} < 0.001. We found no other significant associations for vegetable, meat, and fish intakes and risk of breast cancer subtypes (Table 5). Analysis by food groups showed that high intakes of Asian tropical fruits were all inversely associated with the risk of luminal B, HER2-enriched, and triple-negative/basal-like subtypes (Supplementary Table S6). An inverse association between HER2-enriched subtype and consumption of red meat and organ meats was observed (OR_T3vsT1 = 0.40; 95% CI, 0.17–0.93; P_trend = 0.04 and P_{heterogeneity} = 0.50; Supplementary Table S7).

In this breast cancer case-control study from North Vietnam, we found high intakes of total fruit, and all subgroups of fruits were inversely associated with the risk of breast cancer. These inverse associations were stronger for triple-negative/basal-like breast cancer. Furthermore, we found a high intake of freshwater fish was inversely associated with overall breast cancer, while high consumption of red meat and organ meats was significantly associated with a lower risk of overall and HER2-enriched breast cancer. Total vegetables, total meat and total fish intakes were not significantly associated with overall and molecular subtypes of breast cancer.

Our results are consistent with findings from the majority of case-control studies conducted in the last two decades, in which a reduction of breast cancer risk was associated with high fruit consumption (5). A 2013 meta-analysis of 11 case-control and cohort studies among Chinese women concluded that individual fruit intake reduced breast cancer risk by 32% (20). A weaker reduction (8.0%) in the risk of breast cancer was found for high intake of fruit in a meta-analysis of 10 prospective studies among largely Western populations (21). A recent meta-analysis revealed a 3% reduced risk of breast cancer for each additional 100-g/day increase of fruit (in an analysis of 15 prospective studies with 7,071 breast cancer cases; ref. 22). Analyses of fruit groups showed that inverse associations were seen for specific types of fruit such as citrus fruits, blueberries, strawberries, and peaches/nectarines (12, 23–30). Consistent with these findings, we found that high intakes of most types of fruit were inversely associated with breast cancer risk. Many of the fruits commonly consumed by our participants, such as Asian tropical fruits and citrus fruits, are rich sources of fiber and vitamins, particularly vitamin C, which may possess anticancer effects. Fiber may affect steroid and estrogen metabolism by interacting with entero-hepatic circulation and may prevent carcinogenesis by improving insulin sensitivity and counteracting weight gain (31). Vitamin C, with its antioxidant properties, may trap free radicals and reactive oxygen molecules, regenerate the other antioxidant vitamins, and inhibit the formation of carcinogens that attacks DNA to cause mutagenic changes (32). In addition, lycopene (a carotenoid pigment, particularly common in red fruits), hesperetin, and naringenin (two flavonoids found in orange and grapefruit) were shown to prevent the development of breast cancer through antiproliferative and apoptosis-inducing activities in human breast cancer cell lines (33–35). In general, our results supported the role of fruit intake in relation to breast cancer risk among Vietnamese women. However, we don't have a ready explanation why the association was stronger for triple-negative breast cancer and call for more research to understand the underlying biological mechanism(s).

The association of vegetable intake with breast cancer risk in humans has been inconclusively reported in epidemiologic literature (11). A meta-analysis of nine prospective studies conducted largely in Western populations found no significant association for vegetable intake with breast cancer (21). However, a high intake of vegetables was associated with a 23% reduced breast cancer risk among Chinese women in a 2013 meta-analysis (20). Furthermore, a potential protective effect against breast cancer was suggested for high intakes of specific types of vegetables such as dark, leafy greens, cruciferous vegetables, allium vegetables, fresh legumes, carrots, and tomatoes (12, 23–29, 36). In our study, we found no significant associations between vegetable intake, overall or by subtype, and breast cancer risk. One possible explanation is the cooking methods as most vegetables are boiled or stir-fried before consumption in a typical Vietnamese meal, which may lead to destroying or changing some of the beneficial constituents of vegetables during cooking (5).

Many previous epidemiologic studies among several populations have suggested there were no associations between meat and fish consumption and breast cancer risk (37–42). However, the evidence of positive associations between breast cancer risk and high consumption of red or processed meats has been reported in some meta-analyses. A 2018 meta-analysis, including 18 prospective studies, reported a 6% to 9% higher breast cancer risk for high intakes of unprocessed and processed meats, respectively (43). We observed that a high intake of freshwater fish was inversely associated with the risk of breast cancer. This association might be related to n-3 polyunsaturated fatty acids (n-3 PUFA) such as eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) in fish, which has been found to have anti-inflammatory and antineoplastic effects (44). Most breast cancer cases in our study had luminal B (∼45%), which is a type of breast cancer with high proliferation markers such as the protein Ki-67. In an animal model, EPA and DHA supplementation decreased Ki-67 in benign and malignant mammary neoplasms (44, 45). A meta-analysis of 11 studies, including four cohort studies and seven case-control studies with 20,792 Asian breast cancer cases, showed that high consumption of dietary n-3 PUFA from fish was associated with a 20% reduced risk of breast cancer (46). However, the null association observed for marine fish in our study is inconsistent with the proposed connection.

Few studies have investigated the association between dietary intake and the risk of breast cancer molecular subtypes. A recent report from the Nurses' Health Study (NHS) and NHSII found significant heterogeneity in the associations between fruit and vegetable consumption and risk of breast cancer across molecular subtypes, with stronger inverse associations being observed for more aggressive forms of breast cancer, including HER2-enriched, and triple-negative/basal-like (12). Consistent with this finding, in our study, we found that inverse associations between breast cancer and high fruit intake were strongest for triple-negative/basal-like breast cancer, although inverse associations were also observed for other molecular subtypes.

Unexpectedly, we found that consumption of red meat and organ meats was inversely associated with overall breast cancer and HER2-enriched subtype. While a chance finding is a possible explanation, reverse causation is also a concern due to the delay in seeking medical care we observed in the study population. In the VBCS, we found that delays in diagnosis and treatment are common among Vietnamese patients with breast cancer. The median time interval from the first signs/noticeable breast cancer symptoms to diagnosis and initiation of treatment was 2.4 months (IQR: 1.1–7.1 months) for overall and 5.5 months (IQR: 2.5–9.3 months) for patients who postponed seeking medical care after first symptom recognition (47). It is noteworthy that most patients with breast cancer were diagnosed with symptoms, and approximately 23% of our study patients were diagnosed at stage III or IV. We speculate that physical and psychological stresses during the delay period may have influenced patients’ dietary intakes and their recall of dietary habits (47). It was previously suggested that during the period of symptom development and patients’ seeking medical care, decreased appetite and altered taste acuity (dysgeusia, hypogeusia, and ageusia) may influence patients’ habitual dietary intakes (48). In our study, we found that patients with breast cancer with a delay in seeking medical care, diagnosis, and treatment tended to have lower intake of meat and fish than the no-delay group (Supplementary Table S8). However, this cannot explain the inverse association, particularly the subtype-specific association observed in our study. Therefore, further study is warranted to elucidate this unexpected association.

Our study is the first to investigate associations of food groups with overall breast cancer and by molecular subtypes among Vietnamese women. The high participation rate and availability of breast cancer molecular subtypes are strengths of this study (∼8% of cases lacked information). In our study, we also used a validated FFQ and a structured questionnaire, administered by trained interviewers following a standardized protocol to minimize measurement errors. However, potential interviewer bias cannot be ruled out due to the interviewers' knowledge of the case and control status. Vietnamese people consume more fruits and vegetables during harvest or wet season (from May to August in North Vietnam; ref. 49). Because approximately 19% of cases and 40% of controls were interviewed from May to August, the seasonal intake variation could have influenced the associations under study. Although the influences of seasonal variations in fruit and vegetable intake on association assessment might be minimized by the use of an FFQ covering dietary habits during a 5-year period, influence of recent diet on long-term dietary assessment cannot be overstated. Indeed, we found that the association patterns for total fruit and vegetable intake differed by recruitment season, an effect modification likely to be caused by the influence of current diet on assessment of long-term dietary habits. These findings highlight the importance of closely matching cases and controls on interview season in future research of diet-health to minimize potential biases. Several other limitations should be considered when interpreting our findings. First, habitual long-term intake is difficult to capture and may be subject to differential recall in a retrospective case-control study setting. Recall bias and information bias are inevitable in our study. Second, selection bias is also likely due to the hospital-based design. Close to 85% of controls were selected from healthy women taking care of cancer patients other than breast cancer at these two hospitals. Therefore, they may not be a representative sample for the general Vietnamese population. Our findings may not be generalizable to breast cancer cases and controls living in other parts of Vietnam. Finally, some of the findings, such as red and organ meat intake with HER2-enriched subtype and overall breast cancer, may be due to chance findings resulting from multiple comparisons.

In conclusion, we found high intakes of freshwater fish and fruits were inversely associated with breast cancer risk; the latter was stronger for triple-negative breast cancer. The inverse association between red meat and organ meat intake and the HER2-enriched subtype was unexpected and worth further investigation.

O.T. Bui reports grants from NCI during the conduct of the study. M.J. Shrubsole reports grants from NCI during the conduct of the study. X.-O. Shu reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

S.M. Nguyen: Conceptualization, formal analysis, methodology, writing–original draft, project administration, acquisition of data. H.T.T. Tran: Supervision, project administration, acquisition of data. L.M. Nguyen: Project administration, acquisition of data. O.T. Bui: Project administration, acquisition of data. D.V. Hoang: Project administration, acquisition of data. M.J. Shrubsole: Project administration, acquisition of data. Q. Cai: Project administration, acquisition of data. F. Ye: Project administration, acquisition of data. W. Zheng: Project administration, acquisition of data. H.N. Luu: Project administration, acquisition of data. T.V. Tran: Supervision, project administration, acquisition of data. X.-O. Shu: Conceptualization, formal analysis, supervision, methodology, writing–original draft, project administration, acquisition of data.

This study was supported by NIH/NCI [grant nos. P20 CA210300 and OISE-19-66185-1; to principal investigators (PI) X.-O. Shu and T.V. Tran). S.M. Nguyen was supported by a Vanderbilt-Emory-Cornell-Duke Global Health Fellowship, funded by the NCI and the Fogarty International Center (FIC) of the NIH (D43 TW009337), and Ingram Cancer Professorship to X.-O. Shu. The authors would like to give special thanks to the participants and the research staff members of this project, without whom this study would not have been possible.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.