Dietary intake of garlic and other Allium vegetables and breast cancer risk in a prospective study of postmenopausal women
P Tsai, L Harnack, K Anderson, W Lohman, W Zheng
breast, cancer, diet, epidemiology, garlic
P Tsai, L Harnack, K Anderson, W Lohman, W Zheng. Dietary intake of garlic and other Allium vegetables and breast cancer risk in a prospective study of postmenopausal women. The Internet Journal of Epidemiology. 2007 Volume 6 Number 1.
Because garlic and its organosulfur compounds have been shown to inhibit the occurrence of mammary tumors in animal models, we sought to examine this association in the human population. During 15 years of follow-up, 34,388 postmenopausal women completing a food frequency questionnaire (FFQ) were followed for incident breast carcinoma. Supplemental data from a nested case-control study was analyzed to obtain consumption habits of other Allium vegetables not included in the FFQ. After 15 years, higher garlic intake was not found to be associated with lower breast cancer risk. A statistically significant inverse association was noted in the first 5 years (RR=0.71), but not in the second or third 5-year periods. In the nested case-control study, some inverse relationships were noted between the consumption of Allium vegetables and breast cancer, but overall do not show that a protective effect is afforded by more frequent intake of these herbs.
The Iowa Women's Health Study is supported by a grant from the National Cancer Institute (R01 CA39741).
Few human studies have been conducted to investigate the association of
Materials and Methods
The methods used in the Iowa Women's Health Study have been published elsewhere (14,15). To summarize, in January 1986, a questionnaire was sent to 99,826 randomly selected women, ages 55-69 years, whose names were included on the 1985 Iowa state drivers' license list. A total of 41,836 women (42.3%) completed the questionnaire and were followed for cancer incidence and mortality. The average age of respondents was 61.7 years, and 99% of respondents were Caucasian. The rates of breast cancer among respondents and nonrespondents were virtually identical after five years of follow-up (16). The cohort members were resurveyed by mailed questionnaire in 1987, 1989, 1992, and 1997, and the response rates were high: 91%, 89%, 83%, and 79% respectively. Women with a history of cancer at baseline other than skin cancer (n=3,830), those with a prior mastectomy or lumpectomy (n=1,884), those who were peri- or premenopausal (n=569), as well as those with 30 or more blank items on the food frequency questionnaire (FFQ) or with total energy intake values of <600 or ≥ 5,000 kcal per day (n=3,102) were excluded from all analyses. After all exclusions, a total of 34,388 women were eligible for follow-up. During analysis, those with missing covariate information were also excluded.
Identification of Cases of Breast Cancer
Vital status of the cohort members was determined through computer linkage of participant identifiers with Iowa death certificates; through follow-up questionnaires mailed in 1987, 1989, 1992, and 1997; and through the National Death Index for nonrespondents. Cancer incidence was ascertained through the State Health Registry of Iowa, a part of the National Cancer Institute's Surveillance, Epidemiology, and End Results Program (17). Through December 31, 2000, after 15 years of follow-up, 2031 women who met inclusion criteria were diagnosed with incident invasive or in situ breast carcinoma. Person-years were computed as the time from January 1986 to the first of a) breast cancer diagnosis, b) death (for residents of Iowa), c) midpoint of the interval between the date of last contact and the date of death (for residents outside of Iowa), d) December 31, 2000 (end of follow-up), e) emigration from Iowa (if date unknown), or f) midpoint of interval between the date of last contact and either the date of next follow-up or December 31, 2000 (if date of move was unknown). A diagnosis of breast cancer was treated as a censoring event.
Exposure was assessed at baseline by a self-administered questionnaire that solicited information on factors known or suspected to be relevant to breast cancer risk, such as family history of breast cancer, pregnancy history, menstrual history, and smoking history. Reported body weight and height were used to calculate body mass index (kg/m2). Diet was assessed using a semiquantitative food frequency questionnaire (FFQ) that was almost identical to the one used in the 1984 Nurses' Health Study (18,19). The usual intake frequency of specified portions of 127 food items, including alcoholic beverages (beer, red wine, white wine, and liquor), was ascertained for the year prior to baseline. One of the questions on the FFQ asked cohort members to provide their average use of garlic, fresh or powdered, using one clove or shake as the reference serving size. Frequency categories ranged from “never or less than once per month” to “6 or more per day.”
Food composition values used to generate nutrient intake estimates for the FFQ were obtained from the Harvard University Food Composition Database, which was derived from the US Department of Agriculture (20), and supplemented by manufacturer information and other published values. Nutrient intake was calculated by multiplying the frequency of consumption per day for each item by its nutrient content per serving and totaling the nutrient intake for all food items. In a validation study of 194 female nurses, a FFQ nearly identical to the one used in the IWHS was found to account for 93% of total caloric intake. Calorie-adjusted correlations between nutrient intake estimates from the FFQ and the criterion measure (multiple diet records) ranged from 0.28 for iron to 0.61 for total carbohydrate (21).
Nested Case-Control Study
Because no information was collected at the baseline survey on the usual intake of
In order to reduce potential effects of breast cancer diagnosis and pre-diagnostic disease symptoms on dietary intake, information was obtained for case subjects on usual dietary habits one year before cancer diagnosis. Because breast carcinoma cases were diagnosed during the period from 1992 through 1994, and dietary assessment was conducted during the period from 1995 through 1996, three reference years (1991, 1992, and 1993) were identified, corresponding to the years immediately before breast cancer diagnosis. Control subjects were randomly divided into three corresponding groups of approximately equal sample sizes to obtain their dietary habits during these three reference years. To help the women recall their eating habits that year, a list of national and international events that occurred during the reference year was provided with the questionnaire. Of those eligible for the study, 273 (60.3%) case subjects and 657 (75.0%) control subjects participated. The major reasons for nonparticipation were refusal (29.1% of cases, 18.7% of controls), inability to locate (4.9% of cases, 3.8% of controls), and death before contacting (5.7% of cases, 2.5% of controls).
Statistical Analysis: Cohort Study
The frequency of garlic consumption did not lend itself to the formation of quartiles, due to a large number of non-consumers/low consumers. Instead, consumption was categorized into four groups (<1 clove or shake/month, 1-3 cloves or shakes/month, 1 clove or shake/week, ≥2 cloves or shakes/week). Proportional hazards regression was used to derive relative risks (RRs) and 95% confidence intervals (CIs) adjusted for age and other potentially confounding variables (23), derived from a list that included well-confirmed risk factors for breast cancer (24). A linear trend test using the χ2 statistic was performed to examine trends in risk ratios across levels of garlic consumption (coded ordinally).
Statistical Analysis: Nested Case-Control Study
Although the questionnaire administered to case-control study participants requested the “usual serving size” for each
The distribution of demographic and risk factors from the baseline survey conducted in 1986 are shown for the eligible cohort population (left three columns) and the case-control population (right three columns) (Table 1). Among the cohort population, age, education, family history of breast cancer, history of benign breast disease, body mass index, age at first live birth, age at menopause, parity, and alcohol consumption showed at least a borderline significant association with risk for breast cancer (
Age-adjusted RR estimates for the entire 15 years of follow-up showed no independent association of garlic intake with breast cancer incidence (Table 2). Findings were not substantively changed after multivariate analysis was performed (
Intake frequencies of the six
After 15 years of follow-up, high frequency of garlic consumption (≥ 2 cloves or shakes/week) was not found to be significantly associated with lower breast cancer incidence. A statistically significant association was observed during the first 5 years of follow-up, though not in the second or third 5-year periods. A possible reason for this discrepancy might be that the first five years most accurately reflected the dietary patterns as indicated on the food frequency questionnaire because habits changed over time. These observations may suggest a need to reassess diet periodically to accurately measure consumption. Another possibility for the discrepancy in the results between periods might be that those not yet diagnosed with breast cancer at baseline, but who in actuality had advanced disease, might have had symptoms that influenced dietary habits and responses on the food frequency questionnaire; results from this effect would be most notable when analyzing the first 5-years, given the mortality rates associated with these individuals. To minimize this effect, our FFQ asked for usual intake in the year prior to baseline.
Although a few isolated statistically significant inverse associations between
A second potential explanation for the predominately null findings is low statistical power due to the limited number of case subjects (273) available for the analysis. A third possible explanation is that measurement errors in the assessment of dietary intake may have attenuated the risk estimates. The
Our findings contribute to a limited body of literature, consisting of only five epidemiological studies demonstrating conflicting results. Dorant et al. (11) recorded 469 incident breast cancer cases among their Dutch cohort of postmenopausal women during the first 3.3 years of follow-up; these cases were compared to 1713 cancer-free subcohort members who had previously been randomly chosen for follow-up. High intake of onions, leeks, and garlic supplements was not found to be significantly associated with breast cancer occurrence after controlling for dietary and nondietary risk factors (RRonions = 0.95, 95% CI = 0.61-1.47, RRleeks = 1.08, 95% CI = 0.79-1.48, and RRgarlic supplements = 0.87, 95% CI = 0.58-1.31). Dietary garlic (cloves or powder) consumption was not evaluated. Leek consumption was categorized as 0 times/month, ≤2 times/month, and >2 times/month. Onion consumption was categorized as 0 onions/day, ≤0.25 onions/day, 0.25-0.5 onions/day, and >0.5 onions/day. Garlic supplement use was dichotomized (yes/no). In Switzerland, Levi et al. (10) reported in a study of 107 cases and 318 hospital-based control cases an inverse association with frequency of onion intake, with an age-adjusted OR of 0.5 for high consumption (
More recently, Challier et al. (12) investigated the frequency of garlic and onion consumption in 345 breast cancer patients and 345 age and socio-economic status matched controls in France, reporting a highly significant inverse association between the combined consumption of garlic and onions and breast cancer (
Varying intake frequencies amongst the different study populations may account for the contradictory results. Unfortunately, none of the studies, including our own, quantitatively assessed the amount of
To our knowledge, our study is only the second prospective study to investigate the association between
The authors would like to thank Ching-Ping Hong for her immeasurable assistance with the Statistical Analysis Software (SAS) program. The authors had no conflicts of interest.
Phil B. Tsai, M.D., M.P.H. Email: firstname.lastname@example.org