|Home | About | Journals | Submit | Contact Us | Français|
Abbreviations: CI, confidence interval; FFQ, food frequency questionnaire.
Fruit and vegetable intake may protect against pancreatic cancer, since fruits and vegetables are rich in potentially cancer-preventive nutrients. Most case-control studies have found inverse associations between fruit and vegetable intake and pancreatic cancer risk, although bias due to reporting error cannot be ruled out. In most prospective studies, inverse associations have been weaker and imprecise because of small numbers of cases. The authors examined fruit and vegetable intake in relation to pancreatic cancer risk in a pooled analysis of 14 prospective studies from North America, Europe, and Australia (study periods between 1980 and 2005). Relative risks and 2-sided 95% confidence intervals were estimated separately for the 14 studies using the Cox proportional hazards model and were then pooled using a random-effects model. Of 862,584 men and women followed for 7−20 years, 2,212 developed pancreatic cancer. The pooled multivariate relative risks of pancreatic cancer per 100-g/day increase in intake were 1.01 (95% confidence interval (CI): 0.99, 1.03) for total fruits and vegetables, 1.01 (95% CI: 0.99, 1.03) for total fruits, and 1.02 (95% CI: 0.99, 1.06) for total vegetables. Associations were similar for men and women separately and across studies. These results suggest that fruit and vegetable intake during adulthood is not associated with a reduced pancreatic cancer risk.
Worldwide, pancreatic cancer represents one of the most rapidly fatal of all cancers, with a 1-year case-fatality greater than 97% (1). Most pancreatic cancers are diagnosed at advanced stages, when tumor resection is not possible. Furthermore, current chemotherapy regimens have been unsuccessful at improving survival, and treatment is thus focused on palliation (2). Until feasible screening methods become available, prevention offers the most promising approach to reducing the morbidity and mortality associated with pancreatic cancer.
Few risk factors for pancreatic cancer are known. Tobacco smoking and obesity are modifiable factors with convincing evidence of increasing pancreatic cancer risk (3, 4). Other factors that have been associated with increased risk include a family history of pancreatic cancer, chronic pancreatitis, type 2 diabetes, and some dietary factors, including alcohol consumption and a low intake of folate-containing foods (3–5).
Intake of fruits and vegetables may be protective for pancreatic cancer, since fruits and vegetables are rich in many potentially cancer-preventive agents (6). The risk of pancreatic cancer in relation to fruit and vegetable consumption has been examined in over 35 epidemiologic studies (4, 7–13), and most have found inverse associations with fruit and/or vegetable intake. However, associations have been stronger in case-control studies than in cohort studies, and most case-control studies have relied on proxy respondents for exposure information, with proportions of proxies exceeding 50% of the case series in many studies (14–21). Thus, bias due to reporting error is difficult to rule out. In 2007, an international panel reviewed the majority of studies published through 2006 and concluded that there was limited evidence suggesting that fruit consumption reduces pancreatic cancer risk (4). The panel was unable to make a conclusion for vegetable consumption, because the available data were inconsistent.
In order to gain a better understanding of the relation of fruit and vegetable consumption to pancreatic cancer risk, we analyzed intakes of total and specific fruits and vegetables in the Pooling Project of Prospective Studies of Diet and Cancer (the Pooling Project), an international consortium of prospective cohort studies. Only 4 (8, 13, 22; 2 studies were included in reference 8) of the 14 studies had previously published results on these relations. Using the primary data from these studies, we standardized the definitions of fruit and vegetable intake and covariate categories across studies and conducted multivariate analyses for the risk of pancreatic cancer overall and for particular population subgroups.
The Pooling Project has been described previously (23). Each of the 14 studies included (8, 13, 22, 24–32; 2 studies were included in reference 8, and 2 were included in reference 25) met the following predefined criteria: publication of results on the association between diet and cancer (at any site); diagnosis of at least 50 incident pancreatic cancer cases during follow-up; baseline assessment of usual diet; and conduct of a validation study of the dietary assessment method or a closely related instrument. Each study received approval from the institutional review board of the institution at which the study was conducted. The data sets taken from the 14 studies are listed in Table 1.
Incident cases of pancreatic cancer, defined as International Classification of Diseases, Ninth Revision, code 157.0 or International Classification of Diseases, Tenth Revision, code C25, were identified by self-report and confirmed through subsequent medical record review, linkage with a cancer registry, or both (5, 23). Mortality registries served as an additional source for ascertainment of incident cases for some studies. The proportion of cancer cases ascertained in each study has been estimated to exceed 90% for most of the studies (8, 23, 33, 34). Microscopic confirmation of the tumor was not available for most studies.
A self-administered food frequency questionnaire (FFQ) was used at baseline in each of the studies to assess usual consumption of specific food items (23). For most studies, the time frame for diet assessment was the past year (23). Food intake data were converted to grams consumed per day. We examined 3 main food groups: total fruits and vegetables, total fruits, and total vegetables. Food group intakes were calculated by summing the intakes of specific foods included in that group. Potatoes and mature beans were not classified as vegetables because of their high starch and protein content (35), respectively, compared with other vegetables. We examined fruits and vegetables grouped according to botanical taxonomy (36) to evaluate potentially rich sources of particular bioactive compounds. We also examined individual fruits and vegetables for which intake had been assessed in at least half of the studies.
We acknowledge that fruit and vegetable intake was measured with error in each study. Although validation studies were conducted for each of the FFQs or a similar FFQ, the validity of total fruit and total vegetable intake in particular was evaluated only in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (37), the Netherlands Cohort Study (38), the Cancer Prevention Study II Nutrition Cohort (39), the Health Professionals Follow-up Study (40), and the Swedish Mammography Cohort (41) (validity correlation coefficients for total fruits and total vegetables generally exceeded 0.35; correlations were higher for total fruits than for total vegetables in each of the studies). Thus, we were unable to correct our analyses for measurement error in fruit and vegetable intake.
The study-specific exclusion criteria were applied to the data, after which we excluded participants who had a prior cancer diagnosis (except nonmelanoma skin cancer) at baseline or who reported energy intakes beyond 3 standard deviations from the study-specific loge-transformed mean energy intake. For studies that enrolled both men and women, participants were separated into sex-specific cohorts. We first estimated study- and sex-specific relative risks and 95% confidence intervals using the Cox proportional hazards model (42, 43). Person-years of follow-up were calculated from the date of the baseline questionnaire to the date of pancreatic cancer diagnosis, death, loss to follow-up, or the end of the follow-up period. The Canadian National Breast Screening Study and the Netherlands Cohort Study were analyzed as case-cohort studies (44). Analyses were stratified according to age at baseline (in years) and the year in which the baseline questionnaire was returned. This is equivalent to a left-truncated survival analysis with age as the time scale, and it allows for the baseline incidence rates to vary jointly and arbitrarily by age at enrollment and calendar year. In multivariate analyses, we adjusted for smoking status, alcohol intake, history of diabetes, body mass index (weight (kg)/height (m)2), and total energy intake.
After estimating the study-specific relative risks, we calculated pooled relative risks by combining the study-specific loge relative risks, weighted by the inverse of their variance, using a random-effects model (45). The presence of heterogeneity between studies was evaluated using the Q statistic, which follows an approximate χ2 distribution (45, 46). Fruit and vegetable intakes were categorized according to study-specific quantiles and by common absolute cutpoints. Study-specific quantiles were chosen to maximize the contrast between the highest and lowest intakes while ensuring a sufficient number of participants in each category. Absolute cutpoints represented multiples of fruit/vegetable servings (about 100 g per serving, on average) (35). To calculate the P value for the test for trend across intake categories, we assigned participants the median value of their category, and this variable was entered as a continuous term into the regression model, the coefficient for which was evaluated via the Wald test. We also analyzed fruit and vegetable intakes as continuous variables. We first assessed whether the association was consistent with linearity by examining nonparametric regression curves using restricted cubic splines (47, 48). The fit of the model including the linear and cubic spline terms selected by means of a stepwise regression procedure was compared with the fit of the model including only the linear term, using the likelihood ratio test. For these analyses, data from all studies were combined into a single data set and stratified by study, age, and the year in which the questionnaire was returned. Persons reporting extreme fruit and/or vegetable intake (the top 1% of participants in each study) were excluded.
To assess the presence of heterogeneity by sex, age at diagnosis, follow-up period, and baseline smoking status, which were examined between studies or as a nominal variable, we used a mixed-effects meta-regression model and evaluated the statistical significance of the parameter estimate using a Wald test (23, 49). To evaluate modification of relative risks by baseline measures of alcohol consumption and body mass index, we calculated the pooled relative risks stratified according to levels of these risk factors and then assessed the statistical significance of the cross-product term between fruit/vegetable intake and the effect modifier using a Wald test. Participants with missing values for the modifying factor were excluded from these analyses. All statistical tests were 2-sided, and a P value less than 0.05 was considered statistically significant.
In the pooled cohort of 319,673 men and 542,911 women, 2,212 developed pancreatic cancer (1,057 men, 1,155 women) during a maximum follow-up period that ranged from 7 years to 20 years across the 14 studies (Table 1). Of the 1,669 pancreatic cancers for which histology information was available, 1,498 (90%) were adenocarcinomas. Total fruit and vegetable intake was lowest for the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (median, 225 g/day) and highest for the women in the Melbourne Collaborative Cohort Study (median, 618 g/day).
When fruit and vegetable intakes were modeled as categorical variables defined by common absolute intakes across studies, a statistically significant inverse trend was observed for total fruit intake in the model that was adjusted for age only, in all participants combined and in men (Table 2). In the multivariate analyses, intakes of total fruits and vegetables, total fruits, and total vegetables were not associated with pancreatic cancer risk overall, for men, or for women (Table 2). The strongest confounder affecting these analyses was smoking. When the data were analyzed according to study-specific quartiles, the pooled multivariate relative risks for the highest quartiles of intake versus the lowest were 1.02 (95% confidence interval (CI): 0.89, 1.16) for total fruits and vegetables, 0.96 (95% CI: 0.84, 1.09) for total fruits, and 1.06 (95% CI: 0.91, 1.22) for total vegetables (data not shown) among men and women combined.
The nonparametric regression analyses were consistent with a linear but null association between intakes of total fruits and vegetables, total fruits, and total vegetables and pancreatic cancer risk, for the whole study population and for men and women separately (for all food groups and populations, P for linearity > 0.05). When intakes were modeled as continuous variables, the results were consistent with the categorical and quartile analyses and indicated no association for total fruits and vegetables, total fruits, and total vegetables (Table 3). When total fruits and total vegetables were simultaneously included in the model, the observed relative risks (not shown) were almost the same as those presented in Table 3. The pooled multivariate relative risk for total fruits excluding fruit juices was 0.99 (95% CI: 0.95, 1.02) for each 100-g/day increase in intake, similar to the results for total fruit consumption.
There was no statistically significant heterogeneity between studies for any food group (Tables 2 and and3).3). Statistically significant associations were observed only in the Iowa Women's Health Study and for women in the Netherlands Cohort Study for total vegetable intake (Table 3); both of these associations were positive. In addition, the pooled relative risks were not modified by sex (Tables 2 and and3).3). Since body mass index (50) and diabetes (51) may lie on the causal pathway between fruit and vegetable consumption and pancreatic cancer risk, we removed these two covariates from the model; the relative risks (not shown) were almost identical to those presented in Tables 2 and and3.3. When we limited the study population to participants without diabetes at baseline (1,712 cases among 11 studies), we observed pooled multivariate relative risks (for each 100-g/day increase in intake) of 1.02 (95% CI: 1.00, 1.04) for total fruits and vegetables, 1.01 (95% CI: 0.99, 1.04) for total fruits, and 1.04 (95% CI: 1.00, 1.09; P = 0.06) for total vegetables. To determine whether preclinical disease might have influenced associations, we excluded pancreatic cancer cases that were diagnosed during the first 2 years of follow-up (2,024 cases were included in the analysis) and observed relative risks similar to those observed overall (not shown). To rule out heterogeneity by histologic type, we restricted the analysis to diagnoses of adenocarcinoma (1,498 cases), and the results were similar (not shown).
Associations did not greatly differ according to median age at diagnosis; the pooled multivariate relative risks ranged from 1.01 to 1.03 for total fruits, total vegetables, and total fruits and vegetables for participants diagnosed before age 69 years (n = 1,074 cases) and participants diagnosed at ages ≥69 years (n = 1,138 cases). However, relative risks differed according to period of follow-up for total vegetable intake (test of differences by period: P = 0.54 for total fruits and vegetables, P = 0.58 for total fruits, and P = 0.03 for total vegetables). The relative risk for each 100-g/day increase in total vegetable intake was 0.97 (95% CI: 0.91, 1.03) when the analyses were limited to the first 5 years of follow-up (767 cases) and 1.06 (95% CI: 1.01, 1.10) when the cases included were those that occurred during the follow-up period ≥5 years after baseline (1,445 cases).
When we examined associations according to baseline measurements of pancreatic cancer risk factors, we observed that the relative risks did not vary significantly by level of smoking or alcohol consumption (Table 4). However, relative risks for total vegetable consumption differed by body mass index (P for interaction = 0.03). In particular, increasing vegetable consumption was associated with an increased risk among normal-weight participants but not overweight participants.
When fruit and vegetable intake was grouped according to botanical taxonomy (Appendix Table 1), the pooled multivariate relative risks for each 100-g/day increment in intake were 1.02 (95% CI: 0.89, 1.17) for Cruciferae, 1.01 (95% CI: 0.88, 1.16) for Cucurbitaceae, 0.92 (95% CI: 0.77, 1.10) for Leguminosae, 1.00 (95% CI: 0.95, 1.06) for Rosaceae, 1.02 (95% CI: 0.98, 1.05) for Rutaceae, 1.04 (95% CI: 0.97, 1.12) for Solanaceae, and 0.97 (95% CI: 0.72, 1.30) for Umbelliferae. Marginally statistically significant between-studies heterogeneity was observed for Umbelliferae (study-specific relative risks ranged from 0.10 to 2.53; test for between-studies heterogeneity: P = 0.06). For green leafy vegetables, an increased risk was observed for each 100-g/day increment in intake (pooled multivariate relative risk = 1.17, 95% CI: 1.05, 1.30; P for between-studies heterogeneity = 0.86).
In the analyses of individual foods, statistically significant increased risks were observed for strawberries, fruit juice, brussels sprouts, green peppers, lettuce/salad, and tomatoes (Table 5). When we further adjusted strawberries and fruit juice for total fruits, to see whether the associations were independent of other fruit intake, the pooled multivariate relative risks for each 3-servings/week increment in intake were 1.12 (95% CI: 0.99, 1.27) for strawberries and 1.04 (95% CI: 1.00, 1.08) for fruit juice. Similarly, when each of the individual vegetables that were significantly associated with risk was further adjusted for total vegetables, the pooled multivariate relative risks for each 3-servings/week increment in intake were 1.20 (95% CI: 0.96, 1.49) for brussels sprouts, 1.10 (95% CI: 0.95, 1.27) for green peppers, 1.03 (95% CI: 0.99, 1.08) for lettuce/salad, and 1.02 (95% CI: 0.97, 1.08) for tomatoes. Relative risks for each of the individual foods did not appreciably differ according to period of follow-up or body mass index (not shown).
In this pooled analysis of data from 14 prospective cohort studies, increasing intake of total fruits and vegetables, total fruits, and total vegetables was not associated with pancreatic cancer risk overall. This lack of overall association was consistent regardless of whether intake was examined as a continuous measure or in categories based on absolute cutpoints or study-specific quartiles. There was no statistically significant evidence of heterogeneity between studies, and associations were similar by sex and age group. We observed an increased pancreatic cancer risk with increasing intake of strawberries and fruit juice, although none of the analyses for total fruit consumption showed a positive association. We also observed that intakes of green leafy vegetables, brussels sprouts, green peppers, lettuce/salad, and tomatoes were associated with increased risks. When we adjusted these specific food items for intake of total fruits or total vegetables, respectively, the point estimates were similar but no longer statistically significant. Additionally, we observed some suggestion that vegetable consumption increased risk for the later periods of follow-up (i.e., >5 years) and for participants who had normal weight at baseline.
The relation between fruit and vegetable intake and pancreatic cancer risk has been examined in 20 case-control studies (10, 12, 14–21, 52–61) and 16 cohort studies (8, 9, 11, 13, 22, 62–72), 4 of which were included in our pooled analysis (8, 13, 22; 2 studies were included in reference 8). Of the cohort studies not included in our analysis, 5 had brief food lists on their FFQs (63, 66, 67, 69, 70), 3 did not validate their FFQs (64, 65, 71), 2 were studies of mortality cohorts and thus were not included in the Pooling Project (68, 72), 1 had too few pancreatic cancer cases (62), and 2 joined the Pooling Project only after the pancreatic cancer database had been finalized (9, 11). The findings from most case-control studies have indicated inverse associations with both fruit intake and vegetable intake. However, many of these studies included proxy respondents, thus increasing the likelihood that information bias may have affected results. Of the case-control studies that did not include proxies (10, 12, 53, 56–59, 61), inverse associations with fruit and/or vegetable intake were reported by most, but not all (10, 57, 59), with risk reductions ranging from 40% to 46% (10, 12, 53, 56, 58) for the highest fruit intakes versus the lowest and from 33% to 53% for the highest vegetable intakes versus the lowest (12, 53, 56, 59, 61). However, bias due to differential diet recall or differential participation rates related to fruit and vegetable consumption between cases and controls is possible in these studies. The findings from prospective studies have been weaker. Of the 8 studies not included in our pooled analysis that evaluated total fruit intake (11, 62, 64, 65, 68, 70–72), associations were null in 3 (11, 62, 64), while risk reductions for the highest versus lowest intakes ranged from 11% to 49% in the others; findings were statistically significant in only 1 study (69). Similarly, associations with vegetable intake were null in 3 (11, 64, 66) of the 8 prospective studies that evaluated total vegetable intake, while risk reductions ranged from 14% to 36% in the other studies (9, 65, 68, 69, 72), only 1 of which had statistically significant findings (72).
In our pooled analysis, we observed no suggestion of a decreased pancreatic cancer risk with increasing fruit and vegetable intake. In fact, we observed positive associations for some individual fruit and vegetable items. These positive associations were not expected a priori and may have been due to chance. However, some research has suggested that exposure to pesticides, which may also be present on fruits and vegetables, may increase risk (73). Some experimental research has suggested that indole-3-carbinol, which is found in cruciferous vegetables such as brussels sprouts, may have some cancer-promoting effects, particularly when administered after carcinogen exposure (74, 75). However, we did not see a similar positive association with broccoli and cabbage, which are also sources of indole-3-carbinol. The modest positive association we observed with fruit juice may reflect its possible contribution to a higher glycemic load, which has been positively associated with diabetes (76, 77), a risk factor for pancreatic cancer. We also observed positive associations with total vegetable intake for certain population subgroups, namely those with a longer follow-up period (i.e., >5 years) and normal-weight participants. These associations were also unexpected.
We adjusted all of our analyses for smoking, body mass index, alcohol intake, and history of diabetes, and in multivariate analyses there was still no evidence of an inverse association. Nonetheless, we cannot rule out residual confounding from measurement error in the included covariates or uncontrolled confounding by an unknown or unmeasured factor. For instance, we did not have information on how fruits and vegetables were prepared and consumed; thus, confounding by foods consumed together with fruits and vegetables, such as cream sauces consumed with vegetables, may have been present. Furthermore, we cannot rule out misclassification of fruit and vegetable consumption from the use of dietary information collected only in adulthood. If fruit and vegetable intakes during childhood, adolescence, or early adulthood are more pertinent, our analysis of adult diet might not have captured the relevant exposure period. Another possible source of exposure misclassification might be the way fruit and vegetable intakes were modeled. However, regardless of whether fruit and vegetable intakes were modeled as continuous variables, study-specific quartiles, or categories defined by absolute intakes, our results were similar.
Greater misclassification in estimates of usual consumption may have occurred because we had only a single dietary assessment at baseline as opposed to diet information obtained from multiple assessments throughout follow-up. We could not correct for measurement error in intakes because the validity of total fruit and total vegetable intake was not evaluated in the majority of studies. However, to estimate what role measurement error may have played in our findings for total vegetable intake—the food group for which we observed some positive associations—we applied the method of regression calibration (78–80), using several assumptions to approximate the deattenuation factor. First, we assumed that the correlation between the FFQ data and data from multiple days of diet records in the validation studies for total vegetable intake would probably be similar to the correlation for folate intake, a nutrient that is concentrated in vegetables. For studies that had not assessed folate intake in their validation study, we imputed the sex-specific correlation from the combined data set of the studies with these data. Because of sample size limitations in the validation studies, we were only able to adjust for age, smoking, and total energy intake. However, in the uncorrected model, the age-, smoking-, and energy-adjusted relative risk was 1.02 (95% CI: 0.99, 1.06) for each 100-g/day increment in vegetable intake, which is the same as the multivariate uncorrected relative risk for total vegetables presented in Table 3. Second, we assumed that the ratio of the standard deviation of vegetable intake from the FFQ to that from the diet records in the validation studies of the 1980 Nurses' Health Study and the 1986 Health Professionals Follow-up Study could be applied to the other studies of women and men, respectively, in this pooled analysis. With this correlation and ratio, we constructed a likely deattenuation factor and performed the measurement error correction analysis. In this sensitivity analysis, we observed that the relative risk did not change greatly and was still not statistically significant (for a 100-g/day increment in vegetable intake, relative risk = 1.06, 95% CI: 0.91, 1.23).
This study had several strengths, including the fact that we prospectively examined 14 cohorts from North America, Europe, and Australia with a wide range of fruit and vegetable intakes. By conducting a pooled analysis, we were able to define fruit and vegetable intakes and other covariates in a standardized manner across studies, thus minimizing heterogeneity between studies due to differences in exposure and covariate definitions. This approach also limited publication bias by including not only results from previously published studies but also results from studies that had not been published. We examined several specific foods and food groups that have not been routinely reported on in all previous studies. By including over 2,000 cases in our analysis, we were able to evaluate whether associations were modified by other pancreatic cancer risk factors.
In summary, our results do not provide evidence that fruit and vegetable consumption during adulthood is associated with a decreased risk of pancreatic cancer overall. We observed some positive associations, which may have been due to chance but for which potential mechanisms might be explored in experimental research. Diets rich in fruits and vegetables remain important given the benefits that have been observed for other health outcomes, including cardiovascular disease (81), obesity avoidance (4), and some other cancers (7).
Author affiliations: University of Montreal Hospital Research Centre (CRCHUM), Montreal, Quebec, Canada (Anita Koushik); Department of Social and Preventive Medicine, Faculty of Medicine, University of Montreal, Montreal, Quebec, Canada (Anita Koushik); Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts (Donna Spiegelman, Walter C. Willett, Stephanie A. Smith-Warner); Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts (Donna Spiegelman); Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts (Donna Spiegelman, Walter C. Willett); Nutritional Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, Maryland (Demetrius Albanes, Arthur Schatzkin, Rachael Z. Stolzenberg-Solomon); Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, Minneapolis, Minnesota (Kristin E. Anderson, Kim Robien); Masonic Cancer Center, University of Minnesota, Minneapolis, Minnesota (Kristin E. Anderson, Kim Robien); Division of Cancer Etiology, Department of Population Science, Beckman Research Institute of City of Hope National Medical Center, Duarte, California (Leslie Bernstein); Department of Epidemiology, GROW School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht, the Netherlands (Piet A. van den Brandt); Department of Surgery and Centre for Clinical Research, Central Hospital, Västerås, Sweden (Leif Bergkvist); Cancer Epidemiology Centre, Cancer Council Victoria, Melbourne, Victoria, Australia (Dallas R. English, Graham G. Giles); Centre for Molecular, Environmental, Genetic, and Analytic Epidemiology, University of Melbourne, Melbourne, Victoria, Australia (Dallas R. English, Graham G. Giles); Department of Social and Preventive Medicine, School of Public Health and Health Professions, University at Buffalo, State University of New York, Buffalo, New York (Jo L. Freudenheim, Amy E. Millen); Department of Adult Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts (Charles S. Fuchs); Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York (Jeanine M. Genkinger); Department of Prevention and Health Care, TNO Quality of Life, Leiden, the Netherlands (R. Alexandra Goldbohm); Cancer Prevention Institute of California, Fremont, California (Pamela L. Horn-Ross); Department of Chronic Disease Prevention, National Institute for Health and Welfare, Helsinki, Finland (Satu Männistö); Epidemiology Research Program, American Cancer Society, Atlanta, Georgia (Marjorie L. McCullough); Division of Epidemiology, Dalla Lana School of Public Health, University of Toronto, Toronto, Ontario, Canada (Anthony B. Miller); Department of Epidemiology and Population Health, Albert Einstein College of Medicine, Yeshiva University, Bronx, New York (Thomas E. Rohan); Epidemiology and Biostatistics Program, Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, Maryland (Regina G. Zeigler); Division of Preventive Medicine, School of Medicine, University of Alabama at Birmingham, Birmingham, Alabama (James M. Shikany); Department of Nutrition, Harvard School of Public Health, Boston, Massachusetts (Walter C. Willett, Stephanie A. Smith-Warner); and Division of Nutritional Epidemiology, National Institute of Environmental Medicine, Karolinska Institute, Stockholm, Sweden (Alicja Wolk).
This work was supported by grant P01 CA55075 from the US National Cancer Institute. Dr. Anita Koushik currently holds a New Investigator Award from the Canadian Institutes of Health Research.
The authors are grateful to Shiaw-Shyuan Yaun and Ruifeng Li at the Harvard School of Public Health for their assistance with data management and statistical analyses.
Since July 2011, Dr. Amy E. Millen has been conducting work funded by the Mushroom Council.