Association of vegetable, fruit, and grain intakes with colorectal cancer: the Multiethnic Cohort Study^,

Abstract

Background

It is uncertain whether or not vegetables, fruit, or grains protect against colorectal cancer.

Objective

In a large prospective study, we investigated the association of vegetable, fruit, and grain intakes with colorectal cancer risk.

Design

Between 1993 and 1996, 85 903 men and 105 108 women completed a quantitative food-frequency questionnaire that included ≈180 foods and beverages in the Multiethnic Cohort Study. A diagnosis of colorectal cancer was made in 1138 men and 972 women after an average follow-up of 7.3 y. Cox proportional hazards models were used to calculate multivariate-adjusted relative risks and 95% CIs for colorectal cancer.

Results

In men, multivariate adjustment for energy intake, dietary, and nondietary variables resulted in relative risks in the highest quintile group of 0.74 (95% CI: 0.59, 0.93; P for trend = 0.02) for vegetables and fruit combined, 0.80 (95% CI: 0.64, 0.99; P for trend = 0.09) for fruit alone, and 0.85 (95% CI: 0.69, 1.05; P for trend = 0.05) for vegetables alone. When colon and rectal cases were separated among men, the inverse associations were stronger for colon than for rectal cancer. In women, none of the associations with vegetables, fruit, or vegetables and fruit combined were significant. Grain intake was not associated with colorectal cancer for either men or women.

Conclusion

The intake of vegetables and fruit was inversely related to colorectal cancer risk among men but not among women. The association appears stronger for colon than for rectal cancer.

INTRODUCTION

There has been much interest over the years in the possible protective effect of vegetables, fruit, and grains against colorectal cancer. These plant foods contain potential anticarcinogenic agents, such as antioxidants, flavonoids, isothiocyanates, phenols, and protease inhibitors (1). Nevertheless, the findings from epidemiologic cohort studies have produced inconsistent results.

The National Institutes of Health (NIH) and the AARP Diet and Health Study reported an inverse association of vegetable intake with colorectal cancer among men, but not among women; there was no association with fruit intake in either sex (2). The Cancer Prevention Study II by the American Cancer Society found that men with a very low intake of vegetables and women with a very low intake of fruit were at an increased risk of colon cancer (3). However, the Netherlands Cohort Study on Diet and Cancer reported an inverse association between colon cancer and the combined intake of vegetables and fruit among women but not among men (4).

Other investigations have not supported these findings. The Health Professionals Follow-Up Study and the Nurses’ Health Study found no association of vegetable or fruit consumption (even in subgroups of vegetables or fruit) with colorectal cancer among men and women (5). There was no relation between vegetable or fruit intake and colorectal cancer risk in the Women’s Health Study (6). The Japan Public Health Center study also reported no reduction in colorectal cancer risk with vegetable or fruit consumption (7).

In an earlier investigation, our results supported an inverse association of dietary fiber with colorectal cancer mainly in men (8). There was a suggestion that the association was primarily with vegetable and fruit fiber intakes. Because plant foods are rich sources of dietary fiber, and because of the equivocal results of past studies, we decided to conduct this investigation into the association of vegetables, fruit, grains, and certain subgroups of vegetables and fruit with colorectal cancer in our Multiethnic Cohort Study.

SUBJECTS AND METHODS

Study design and population

The Multiethnic Cohort Study in Hawaii and Los Angeles was designed to investigate the association of dietary, lifestyle, and genetic factors with the incidence of cancer and other chronic diseases. Its study design, questionnaire development, subject recruitment, and data collection were described elsewhere (9). Briefly, >215 000 men and women aged 45–75 y and living in Hawaii or in California (mainly in Los Angeles County) completed a 26-page self-administered mailed questionnaire between 1993 and 1996. The primary sampling frame for the study was the driver’s license files in both states, because they included the names of most adult residents, contained information on age, and encompassed all socioeconomic strata.

Study participants provided information on their diet, body weight and height, demographic factors, lifestyle practices (including smoking and physical activity), history of medical conditions, use of medications (including aspirin), use of dietary supplements, a family history of common cancers, and, for women, reproductive history and use of hormone replacement therapy. All questionnaire data were checked for consistency and legibility before scanning and stored in a secured database. The institutional review boards at the University of Hawaii and at the University of Southern California approved the study protocol.

Study exclusion criteria

For this analysis, we limited the study participants to the 5 major ethnic groups recruited into the study (African Americans, Japanese Americans, Latinos, Native Hawaiians, and whites). Latinos were defined as persons of Mexican or South or Central American ancestry, including immigrants from those countries. We excluded relatively small numbers of Chinese, Filipinos, and members of other ethnic groups (n = 13 994). In addition, we excluded individuals with implausible diets (n=8265) in each of the 5 remaining ethnic groups. Implausible diets were defined as those with energy intakes >3 Modified SDs from the mean, where the MSD was calculated after individuals in the top and bottom 10% tails of the log energy distribution by ethnicity were excluded and a truncated normal distribution was assumed. Similar exclusions were made for fat, carbohydrate, and protein intakes with the use of a range of mean ± 3.5 MSD to further identify individuals who failed to complete the quantitative food-frequency questionnaire (FFQ), but whose energy levels were reasonable. Subjects with a colorectal cancer diagnosis before baseline that was either self-reported in the FFQ or identified by registry linkages (n = 2561) were also excluded. Persons with other bowel diseases were not excluded from the cohort. As a result, 191 011 participants remained in the analysis.

Dietary assessment

Dietary intake was assessed at baseline by using a comprehensive questionnaire especially designed and validated for use in this multiethnic population. The development of the self-administered quantitative FFQ (QFFQ) was described elsewhere (9, 10). Briefly, the collection of 3-d measured dietary records from ≈60 men and women of each ethnic group served as the basis for the selection of food items for the QFFQ. The minimum set of food items contributing ≥85% of the intake of a specific list of nutrients for each ethnic group was selected and supplemented by the inclusion of food items that were common in the diet of a particular ethnic group, irrespective of their nutrient contributions (9). The QFFQ inquired about the usual frequency, based on 8 or 9 categories, and on the amount of food consumed, based on 3 portion sizes per food item. The portion size and gram weight of each portion size were derived from the 3-d measured dietary records. The amount listed for each portion size of each food was weighed and average weights were used to convert the portion sizes to grams. Photos showing examples of portion sizes were used to assist the responder and examples of portion sizes were described. As an example, the 3 choices of portion sizes for hamburgers (on a bun) were 1 regular size burger, 1 quarter-pound burger, or 1 large double burger. There were ≈180 foods and beverages included in the questionnaire.

For processing dietary intake data, we used a food-composition database that has been developed and maintained at the Cancer Research Center of Hawaii. The Cancer Research Center of Hawaii food-composition table includes a large recipe database and many unique foods consumed by the 5 multiethnic populations included in this analysis (9). For questionnaire items covering more than one food, nutrient profiles of the items were calculated by using a weighted average of the specific foods based on the frequency of use in the 24-h recalls obtained as part of a calibration study (10). Food intake measured by the QFFQ was linked to the Cancer Research Center of Hawaii food-composition table to convert daily grams to daily nutrients consumed from that food. Before food group intake was calculated, the food mixtures from the QFFQ were disaggregated to the ingredient level by using a customized recipe database. For example, the tomatoes on pizza were counted toward the vegetable group. Food group intake was calculated as grams per day of the basic food commodities. Food groups and foods included in this analysis were the following: total fruit and total vegetables combined; total fruit separately, citrus and yellow-orange fruit; total vegetables separately, light green, dark green, yellow-orange, and cruciferous vegetables, including broccoli; and grains. Fruit juices were included only in the total fruit categories, but not with citrus or yellow-orange fruit. Some vegetables were included in multiple categories. For example, broccoli is both a dark green and a cruciferous vegetable. Grains included rice, cereals, breads, and pasta—all in cooked form.

For validation and calibration purposes, a substudy was incorporated into the initial dietary assessment. Details about this calibration study were published previously (10). In total, 1606 study participants, who were randomly chosen out of subgroups defined by sex and ethnicity, completed 3 unannounced 24-h dietary recalls via telephone during a period of ≈3 mo and an additional QFFQ 3 mo afterward. Correlation coefficients between the dietary recall measurement and the QFFQ were 0.31 and 0.25 for vegetables, 0.58 and 0.52 for fruit, and 0.67 and 0.66 for grains among men and women, respectively. The instruments led to different levels of energy estimates, with correlations of 0.31 for men and 0.20 for women. Therefore, the correlations, based on density measurements (g ·1000 kcal⁻¹ · d⁻¹, adjusted for energy intake), were higher: 0.43 and 0.36 for vegetables, 0.60 and 0.60 for fruit, and 0.70 and 0.72 for grains among men and women, respectively.

Surveillance

Since 1993, the cohort has been under surveillance for colorectal cancer incidence and mortality by record linkages to the Hawaii Tumor Registry, the Cancer Surveillance Program for Los Angeles County, and the California State Cancer Registry in addition to the death certificate files in Hawaii and California and to the National Death Index. All 3 cancer registries are members of the NCI’s Surveillance, Epidemiology and End Results Program (11). The out-migration rate in the cohort has been low at 3.7% after 7 y of follow-up; therefore, few cases should be missed through passive follow-up. Case ascertainment was complete through 31 December 2001. Information was available on the histologic type of the tumor, its anatomical location, and the stage of the cancer. The identification of cases in this study was limited to patients with a diagnosis of invasive adenocarcinoma of the large bowel (n = 2110 cases). Colorectal cancer patients, who did not have adenocarcinoma of the large bowel (n = 111) or who had a diagnosis of carcinoma in situ (n = 183), were not included as cases. Colon cancer cases had an International Classification of Disease (ICD)-02 code of C18.0–C18.9 or C26.0. Rectal cancer cases had an ICD-02 code of C19.9 or C20.9. Stage of disease was based on the following categories: localized, regional, distant, and unstaged. Tumors with regional or distant spread of the disease were considered advanced tumors. In all, there were 1571 colon and 515 rectal cancer cases, whereas there were 24 cases of synchronous tumors at both sites.

Statistical analysis

We applied Cox proportional hazards models using age as the time metric to calculate relative risks. Person-times were calculated beginning at the date of cohort entry, defined as questionnaire completion, or, for the few individuals (n = 1113) who were slightly younger than 45 y of age when they completed the baseline questionnaire, as the date the participant turned 45 y of age. Person-times ended at the earliest of the following dates: date of colorectal cancer diagnosis, date of death, or 31 December 2001—the closure date of the study. Tests of proportional hazards assumptions, based on plotting the sum of scaled Schoenfeld residuals and parameter estimates against time, showed no violations for any analysis (12). All Cox models were stratified by follow-up time, categorized as ≤2, 2–5, and >5 y. Food groups were investigated in disease models in terms of quintiles. Four dummy variables were created to represent the sex-specific quintiles, which were based on the distribution of each exposure across the entire cohort of men or women. Trend variables assigned the sex- and ethnic-specific median values for the appropriate sex-specific quintiles were used in the respective models to test for dose-response. In a subsequent analysis, separate regression parameters were computed for colon cancer and rectal cancer, using competing risk techniques described in (12) and compared using a Wald test.

The following adjustment factors were used in the multivariate model: age at cohort entry, ethnicity (indicator variables), family history of colorectal cancer (indicator variable), history of colorectal polyp (indicator variable), pack-years of cigarette smoking, body mass index (weight in kg divided by the square of height in meters), physical activity (hours of vigorous activity in a week), aspirin use (indicator variable for at least twice a week for ≥1 mo), multivitamin use (indicator variable for at least once a week during the past year), use of hormone replacement therapy for women (indicator variable for current or past user of estrogen, with or without progesterone), energy intake (logarithmically transformed to reduce the correlation between the variability and the magnitude of values), alcohol, red meat, folate (food and supplements), vitamin D (food), and calcium (food and supplements). All adjustment variables were entered in the model as continuous measures, unless otherwise indicated. In a sensitivity analysis, the dietary adjustment variables were modeled by a variety of techniques, such as nutrient densities and by the method of residuals; however, the results for vegetables and fruit were unchanged. Anyone with missing data for any of the adjustment variables was excluded in the multivariate analysis. Multiple imputation was used to verify that the results were not biased by the pattern of missing data (data not shown).

To reduce measurement error in the dietary assessments, we analyzed food intakes in terms of densities, ie, adjusted for energy intake by creating ratios of amount per 1000 kcal per day. As noted above, in the validation study we found that energy-adjusted intake (by the methods of residuals or using densities) produced substantially higher correlation coefficients with the reference instrument than did crude intake (10). This phenomenon has also been reported in other studies (13). All analyses were performed by using SAS Statistical Software, version 9 (SAS Institute Inc, Cary, NC), and all statistical tests were 2-sided.

RESULTS

There were 85 903 men and 105 108 women included in this study. The mean duration of follow-up was 7.3 y, with a total of >1.4 million person-years of observation. During that time, 1138 men and 972 women were diagnosed with colorectal cancer. The mean age at diagnosis was 68.5 y for men and 68.0 y for women.

Baseline characteristics of colorectal cancer cases and the total study population by sex are shown in Table 1. The cases were older at the time of study entry, more likely to have a family history of colorectal cancer, more likely to be ever smokers, less likely to use multivitamins, and less physically active than were the study population (both men and women). In women, cases were also heavier and less likely to report the use of replacement hormones.

TABLE 1.

Baseline characteristics of the colorectal cancer cases and the study population by sex¹

	Men		Women
	Cases (n = 1138)	Population (n = 85 903)	Cases (n = 972)	Population (n = 105 108)
Age of cohort at entry (y)	64.7 ± 7.6	60.2 ± 8.9	64.1 ± 7.9	59.7 ± 8.8
Race-ethnicity (%)
African American	15.6 ± 1.1	13.9 ± 0.1	27.7 ± 1.4	19.8 ± 0.1
Japanese American	38.0 ± 1.4	29.5 ± 0.2	30.8 ± 1.5	27.2 ± 0.1
Latino	18.9 ± 1.2	24.4 ± 0.2	15.4 ± 1.2	21.5 ± 0.1
Native Hawaiian	7.1 ± 0.8	7.0 ± 0.1	6.2 ± 0.8	7.4 ± 0.1
White	20.4 ± 1.2	25.3 ± 0.2	20.0 ± 1.3	24.1 ± 0.1
Family history of colorectal cancer (%)	11.2 ± 0.9	7.2 ± 0.1	13.4 ± 1.1	8.5 ± 0.1
History of colorectal polyps (%)	5.9 ± 0.7	6.8 ± 0.1	5.3 ± 0.7	4.3 ± 0.1
Cigarette smoking status
Nonsmoker	23.4 ± 1.3	30.0 ± 0.2	52.7 ± 1.6	55.7 ± 0.1
Past smoker	58.2 ± 1.5	51.9 ± 0.2	33.9 ± 1.5	30.0 ± 0.1
Current smoker	18.4 ± 1.2	18.2 ± 0.1	13.4 ± 1.2	14.4 ± 0.1
Pack-years of cigarette smoking2	23.5 ± 17.9	20.6 ± 16.6	16.7 ± 14.8	15.4 ± 14.4
BMI (kg/m2)	26.0 ± 4.2	26.1 ± 4.2	26.5 ± 6.1	26.0 ± 5.7
Vigorous physical activity (h/wk)	3.0 ± 5.9	4.1 ± 7.1	0.9 ± 2.3	1.5 ± 3.8
Aspirin use (%)	40.0 ± 1.4	41.3 ± 0.2	38.9 ± 1.6	37.6 ± 0.2
Multivitamin use (%)	42.0 ± 1.5	47.7 ± 0.2	48.4 ± 1.6	54.0 ± 0.2
Replacement hormone use (%)			42.6 ± 1.6	46.6 ± 0.2
Energy intakes (kcal/d)	2329 ± 1044	2380 ± 1105	1854 ± 924	1947 ± 949
Alcohol (g/d)	17.3 ± 35.9	14.6 ± 32.6	5.6 ± 23.9	4.3 ± 14.9
Red meat (g · 1000 kcal−1 · d−1)	29.1 ± 15.7	29.5 ± 17.1	25.0 ± 16.4	24.4 ± 15.9
Folate intake (μg/d)	503 ± 330	548 ± 369	502 ± 357	528 ± 358
Vitamin D (IU · kcal −1 · d −1)	61.9 ± 37.5	62.9 ± 40.6	68.4 ± 44.0	69.6 ± 45.1
Calcium (mg/d)	887 ± 513	965 ± 564	954 ± 653	1059 ± 700

¹All values are $\stackrel{‒}{x}$ ± SD or percentage ± SE.

²For current and past smokers only.

The relative risk of colorectal cancer by intake of vegetables and fruit combined, vegetables (alone), fruit (alone), and grains is presented separately for men and women in Table 2. First, there was a strong inverse trend in colorectal cancer risk with adjustment for age and ethnicity among men for the intake of vegetables and fruit combined, vegetables alone, and fruits alone, but not for grains. After multivariate adjustment for energy intake and dietary and nondietary variables, the relative risks for the highest compared with the lowest quintiles were 0.74 for vegetables and fruit combined and 0.80 for fruit alone, whereas the inverse trend for vegetables alone was of borderline significance (P = 0.052).

TABLE 2.

Relative risk (95% CI) of colorectal cancer by quintile (Q) of vegetable, fruit, and grain intakes

Food group	No. of cases	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend1
Men
Vegetables and fruit
Median intake (g · 1000 kcal −1 · d−1)		134.7	205.3	267.7	343.2	483.2
Adjusted2	1138	1	0.79 (0.66, 0.94)	0.75 (0.63, 0.89)	0.71 (0.59, 0.85)	0.57 (0.47, 0.69)	<0.0001
Multivariate adjusted3	1023	1	0.88 (0.72, 1.06)	0.84 (0.69, 1.02)	0.84 (0.69, 1.03)	0.74 (0.59, 0.93)	0.018
Vegetables
Median intake (g · 1000 kcal−1 · d−1)		71.9	105.5	134.2	168.9	236.2
Adjusted2	1138	1	1.04 (0.88, 1.24)	0.83 (0.69, 0.99)	0.80 (0.67, 0.96)	0.76 (0.63, 0.91)	0.0002
Multivariate adjusted3	1023	1	1.07 (0.89, 1.28)	0.86 (0.71, 1.05)	0.91 (0.75, 1.11)	0.85 (0.69, 1.05)	0.052
Fruit
Median intake (g · 1000 kcal −1 · d−1)		30.1	73.6	118.9	179.9	295.9
Adjusted2	1138	1	0.79 (0.66, 0.95)	0.70 (0.58, 0.84)	0.66 (0.55, 0.79)	0.63 (0.53, 0.76)	<0.0001
Multivariate adjusted3	1023	1	0.86 (0.71, 1.05)	0.78 (0.64, 0.95)	0.79 (0.64, 0.97)	0.80 (0.64, 0.99)	0.089
Grains
Median intake (g · 1000 kcal −1 · d−1)		73.4	106.1	140.6	192.8	290.4
Adjusted2	1138	1	0.84 (0.69, 1.02)	0.97 (0.80, 1.17)	0.92 (0.75, 1.12)	0.85 (0.68, 1.05)	0.226
Multivariate adjusted3	1023	1	0.94 (0.76, 1.16)	1.13 (0.92, 1.39)	1.08 (0.86, 1.35)	0.98 (0.76, 1.26)	0.794
Women
Vegetables and fruit
Median intake (g · 1000 kcal −1 · d−1)		176.3	267.3	346.2	440.0	608.1
Adjusted2	972	1	0.97 (0.79, 1.19)	0.95 (0.77, 1.16)	0.85 (0.69, 1.05)	0.90 (0.73, 1.10)	0.191
Multivariate adjusted3	802	1	1.04 (0.82, 1.30)	1.04 (0.83, 1.31)	0.94 (0.74, 1.20)	1.04 (0.81, 1.33)	0.995
Vegetables
Median intake (g · 1000 kcal −1 · d−1)		85.5	125.8	160.5	202.7	286.5
Adjusted2	972	1	0.89 (0.73, 1.08)	0.81 (0.66, 0.99)	0.95 (0.78, 1.15)	0.85 (0.70, 1.04)	0.277
Multivariate adjusted3	802	1	0.87 (0.70, 1.09)	0.85 (0.68, 1.07)	0.94 (0.76, 1.17)	0.94 (0.75, 1.17)	0.920
Fruit
Median intake (g · 1000 kcal −1 · d−1)		47.5	108.5	168.5	243.4	381.5
Adjusted2	972	1	0.85 (0.69, 1.04)	0.94 (0.77, 1.15)	0.88 (0.72, 1.07)	0.79 (0.64, 0.97)	0.046
Multivariate adjusted3	802	1	0.85 (0.68, 1.08)	1.00 (0.80, 1.25)	0.97 (0.77, 1.22)	0.83 (0.65, 1.06)	0.272
Grains
Median intake (g · 1000 kcal −1 · d−1)		73.4	106.5	138.3	182.8	266.5
Adjusted2	972	1	1.12 (0.93, 1.36)	1.11 (0.91, 1.36)	0.84 (0.67, 1.05)	0.93 (0.74, 1.18)	0.221
Multivariate adjusted3	802	1	1.12 (0.90, 1.40)	1.19 (0.95, 1.49)	0.96 (0.75, 1.24)	1.05 (0.80, 1.38)	0.913

¹P value for Wald test of trend variables, assigned the median for the appropriate quintile.

²Estimated from Cox regression in which age is the time metric, adjusted for ethnicity and time since cohort entry as strata variables and age at cohort entry as an independent variable in the log linear model component.

³Estimated from Cox regression in which age is the time metric, adjusted for ethnicity and time since cohort entry as strata variables and age, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, aspirin use, multivitamin use, replacement hormone use (women), log energy intake, alcohol, red meat, folate, vitamin D, and calcium as independent variables in the log linear model component.

For women, there was an inverse trend with adjustment for age and ethnicity for fruit intake, but not for vegetables, vegetables and fruit combined, or grains. The highest quintile group had a relative risk of 0.79 for fruit. After multivariate adjustment, the inverse trend was no longer significant for fruit intake.

To reduce the chances that undiagnosed preclinical colorectal cancer could affect the results, we repeated the multivariate analysis for vegetables and fruit combined, excluding cases diagnosed within 2 y of their study enrollment. For the remaining subgroups, relative risks for the highest compared with the lowest quintiles were 0.72 (95% CI: 0.56, 0.93; P for trend = 0.03) for men and 1.08 (95% CI: 0.81, 1.46; P for trend = 0.84) for women (based on 763 and 601 cases, respectively). After reanalysis of the data including only advanced tumors with either regional or distant spread as endpoint events, the relative risks for the highest quintiles after multivariate adjustment were 0.77 (95% CI: 0.57, 1.05; P for trend = 0.08) for men and 1.18 (95% CI: 0.84, 1.66; P for trend=0.75) for women (based on 555 and 440 cases, respectively).

Next, the analysis in Table 2 for men was done separately for colon and rectal cancer cases. With multivariate adjustment, there was a significant trend between colon cancer and vegetables alone and vegetables and fruit combined, as shown in Table 3. Also, the highest quintile of fruit intake had a relative risk of 0.75 for colon cancer. However, none of the trends were significant for rectal cancer. The Wald test based on competing-risk analysis for a difference in the effect of fruit and vegetable intake on colon cancer risk and on rectal cancer was not significant (P=0.13). Similar analyses were done for women (Table 3). There were no significant trends in colon or rectal cancer risk with the intake of vegetables, fruit, vegetables and fruit combined, or grains.

TABLE 3.

Multivariate-adjusted relative risk (95% CI) of colon and rectal cancer by quintile of vegetable, fruit, and grain intakes¹

Cancer site and food group	No. of cases	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend2
Men
Colon
Vegetables and fruit	734	1	0.87 (0.69, 1.09)	0.88 (0.70, 1.11)	0.89 (0.70, 1.12)	0.72 (0.55, 0.94)	0.037
Vegetables	734	1	1.03 (0.83, 1.27)	0.77 (0.61, 0.98)	0.86 (0.68, 1.08)	0.80 (0.63, 1.03)	0.039
Fruit	734	1	0.81 (0.64, 1.02)	0.82 (0.65, 1.04)	0.82 (0.65, 1.05)	0.75 (0.58, 0.97)	0.108
Grains	734	1	0.90 (0.70, 1.15)	1.14 (0.89, 1.45)	1.08 (0.83, 1.41)	0.95 (0.70, 1.28)	0.714
Rectum
Vegetable and fruit	276	1	0.90 (0.63, 1.27)	0.71 (0.49, 1.05)	0.69 (0.46, 1.03)	0.72 (0.47, 1.11)	0.097
Vegetables	276	1	1.17 (0.81, 1.68)	1.05 (0.72, 1.53)	0.98 (0.66, 1.45)	0.97 (0.64, 1.46)	0.586
Fruit	276	1	0.95 (0.67, 1.33)	0.62 (0.44, 0.95)	0.67 (0.45, 0.99)	0.80 (0.53, 1.21)	0.219
Grains	276	1	1.05 (0.69, 1.60)	1.14 (0.75, 1.74)	1.06 (0.67, 1.67)	1.05 (0.64, 1.71)	0.978
Women
Colon
Vegetables and fruit	617	1	1.03 (0.79, 1.33)	1.08 (0.83, 1.40)	0.96 (0.73, 1.27)	1.03 (0.78, 1.38)	0.990
Vegetables	617	1	0.89 (0.69, 1.13)	0.81 (0.63, 1.04)	0.90 (0.70, 1.15)	0.90 (0.70, 1.17)	0.624
Fruit	617	1	0.96 (0.74, 1.25)	1.03 (0.79, 1.33)	1.03 (0.79, 1.35)	0.87 (0.65, 1.15)	0.355
Grains	617	1	1.06 (0.82, 1.36)	1.18 (0.92, 1.52)	0.86 (0.64, 1.15)	1.05 (0.78, 1.43)	0.986
Rectum
Vegetables and fruit	179	1	1.16 (0.72, 1.86)	0.93 (0.57, 1.54)	0.94 (0.57, 1.57)	1.10 (0.65, 1.85)	0.956
Vegetables	179	1	0.81 (0.49, 1.35)	1.06 (0.66, 1.69)	1.18 (0.74, 1.88)	1.09 (0.67, 1.77)	0.415
Fruit	179	1	0.60 (0.36, 0.99)	0.96 (0.61, 1.51)	0.79 (0.49, 1.28)	0.77 (0.46, 1.27)	0.609
Grains	179	1	1.45 (0.89, 2.37)	1.23 (0.73, 2.07)	1.47 (0.86, 2.50)	1.09 (0.60, 1.98)	0.887

¹Estimated from Cox regression in which age is the time metric, adjusted for ethnicity and time since cohort entry as strata variables and age, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, aspirin use, multivitamin use, replacement hormone use (women), log energy intake, alcohol, red meat, folate, vitamin D, and calcium as independent variables in the log linear model component. Intakes were quantified as g · 1000 kcal ⁻¹ · d⁻¹.

²P value for Wald test of trend variables, assigned the median for the appropriate quintile.

Because of the possible importance of the type of vegetables and fruit consumed, we determined the risk of colorectal cancer according to the consumption of 7 specific foods or food groups among men and women (Table 4). After multivariate adjustment, none of the trends was significant for either sex.

TABLE 4.

Multivariate-adjusted relative risk (95% CI) of colorectal cancer by intake of types of fruit and vegetables¹

Type of fruit and vegetables	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend2
Men
Yellow-orange fruit	1	0.97 (0.80, 1.18)	0.87 (0.71, 1.06)	0.89 (0.73, 1.09)	0.83 (0.68, 1.03)	0.081
Citrus fruit	1	0.91 (0.75, 1.10)	0.96 (0.80, 1.16)	0.72 (0.59, 0.89)	0.85 (0.70, 1.04)	0.082
Light-green vegetables	1	1.10 (0.90, 1.33)	1.01 (0.82, 1.23)	1.00 (0.82, 1.22)	0.93 (0.75, 1.15)	0.234
Dark-green vegetables	1	0.88 (0.73, 1.06)	0.81 (0.67, 0.99)	0.81 (0.67, 0.99)	0.88 (0.72, 1.07)	0.380
Yellow-orange	1	0.97 (0.80, 1.17)	0.94 (0.78, 1.15)	0.83 (0.68, 1.02)	0.90 (0.73, 1.11)	0.248
vegetables
Cruciferous vegetables	1	0.93 (0.76, 1.13)	0.88 (0.75, 1.15)	0.90 (0.73, 1.10)	0.87 (0.71, 1.08)	0.291
Broccoli	1	1.10 (0.91, 1.33)	0.95 (0.77, 1.16)	0.98 (0.80, 1.20)	0.94 (0.76, 1.15)	0.300
Women
Yellow-orange fruit	1	0.81 (0.65, 1.01)	0.99 (0.72, 1.12)	0.91 (0.73, 1.13)	0.83 (0.66, 1.05)	0.354
Citrus fruit	1	0.95 (0.75, 1.19)	0.90 (0.71, 1.13)	1.08 (0.86, 1.34)	1.04 (0.83, 1.30)	0.368
Light-green vegetables	1	0.99 (0.79, 1.24)	1.00 (0.80, 1.25)	0.79 (0.63, 1.01)	1.04 (0.83, 1.31)	0.911
Dark-green vegetables	1	0.85 (0.68, 1.05)	0.76 (0.61, 0.94)	0.73 (0.59, 0.92)	0.85 (0.68, 1.06)	0.344
Yellow-orange	1	0.74 (0.59, 0.93)	0.86 (0.69, 1.07)	0.84 (0.67, 1.05)	0.90 (0.72, 1.13)	0.891
vegetables
Cruciferous vegetables	1	0.91 (0.73, 1.14)	0.76 (0.60, 0.96)	0.79 (0.62, 0.99)	0.91 (0.73, 1.14)	0.787
Broccoli	1	0.93 (0.75, 1.16)	0.85 (0.68, 1.06)	0.72 (0.57, 0.91)	0.92 (0.75, 1.15)	0.652

¹Estimated from Cox regression in which age is the time metric, adjusted for ethnicity and time since cohort entry as strata variables and age, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, aspirin use, multivitamin use, replacement hormone use (women), log energy intake, alcohol, red meat, folate, vitamin D, and calcium as independent variables in the log linear model component. Intakes were quantified as g · 1000 kcal⁻¹ · d⁻¹.

²P value for Wald test of trend variables, assigned the median for the appropriate quintile.

To examine the stability of the overall findings for men and women, relative risks in the highest quintiles for the combined intake of vegetables and fruit were determined for each of the 10 sex-ethnic-specific groups after adjustment for energy intake and dietary and nondietary variables. The results are shown in Figure 1. The relative risk among all men combined was significant (Table 2), and the relative risks were <1 in each of the 5 ethnic groups. For women, the relative risks were <1 for Latinos and Native Hawaiians, but the CIs were wide.

FIGURE 1.

The relative risks and 95% CIs of the highest quintiles of the combined intake of vegetables and fruit compared with the lowest quintiles for colorectal cancer are represented as squares and horizontal lines by sex for the 5 ethnic groups. The relative risks were estimated from Cox regression, where age is the time metric, and adjusted for time since cohort entry as strata variabless and age, family history of colorectal cancer, history of colorectal polyp, pack-years of cigarette smoking, BMI, hours of vigorous activity, aspirin use, multivitamin use, replacement hormone use (women), log energy intake, alcohol, red meat, folate, vitamin D, and calcium as independent variables in the log linear model component.

The median intake of vegetables and fruit combined was higher for Latinos and Native Hawaiians (range of intake from 600 to 679 and 665 to 707 g/d, respectively, for men and women) and lower for African Americans, Japanese Americans, and whites (range of intake from 513 to 582 and 573 to 606 g/d, respectively, for men and women).

DISCUSSION

In our study, high consumers of vegetables had a reduced risk of colorectal cancer among men but not among women. Analysis by cancer site suggested that the reduced risk was stronger for colon than for rectal cancer. The NIH-AARP Study (2) and the Cancer Prevention Study II (3) also reported a low colon cancer risk with vegetable consumption among men. However, there was no association with vegetable intake among men in other cohort studies (5, 7, 14). The NIH-AARP Study found its inverse effect mainly with green leafy vegetables. The Netherlands Cohort Study reported that Brassica and cooked leafy vegetables, but not total vegetables, were inversely related to colorectal cancer in both sexes (4). A particular type of vegetable did not account for the association with total vegetables in our study. In support of our results in women, 4 cohort studies found no relation between vegetables and colorectal cancer risk (5–7, 15).

With regard to fruit intake, our finding of an inverse association with colorectal cancer among men was not supported by past cohort studies. There was no association between fruit intake and colorectal cancer among men in 6 studies (2–5, 7, 14). There was no clear indication that differences in types of consumed fruit in our study may have contributed to the varied findings. Among women, we found no association with fruit intake after adjustment for nondietary and dietary variables. Six other cohort studies (2, 5–7, 15, 16) also reported no relation between fruit and colorectal cancer among women. In contrast, the Swedish Mammography Screening Cohort Study (17) and the Cancer Prevention Study II (2) reported reduced risks of 32% and 26%, respectively, in the highest quantile of fruit intake after multivariate adjustment.

It is uncertain why the association with vegetables and fruit was weaker in women than in men in our cohort. Others have noted that sex differences in bile acid synthesis and composition may account for differences in colorectal cancer risk (18). Past studies have reported that estrogen replacement therapy among women protects against colorectal cancer (19, 20); we also found this inverse relation in the Multiethnic Cohort Study (data not shown). Nevertheless, when we separated hormone replacement therapy users from nonusers, relative risks in the highest quintile of total fruit and vegetable intake were 1.08 for users and 1.00 for nonusers, which suggested no effect. Some researchers have suggested that there are differences in risk factors between men and women with regard to colorectal neoplasia (21), but the issue still needs to be resolved. The consumption of grains, which included both refined and whole grains, was not related to colorectal cancer in our cohort. Previously, we found no association with fiber from grains (8) and therefore did not examine whole grains separately in the current analyses. Fiber from grains and whole grains was inversely associated with colorectal cancer in the NIH-AARP Study (22). Earlier, the Cancer Prevention II study also reported a low risk of fatal colon cancer among men and women with the intake of high-fiber grains (23). However, in a subsequent report, the researchers found no association with whole grain intake (3). A similar lack of association was reported by others with regard to the consumption of whole-grain cereals in men (14) or cereal fiber in women (17).

Variation in the results of different studies could be due to a number of factors mentioned by others. Measurement error is a concern (5, 15), especially with a single FFQ (24), which may not satisfactorily represent long-term intake (15). We attempted to minimize these potential limitations by rigorously designing our QFFQ, by including a comprehensive group of food items, and by using nutrient densities in the analyses, which resulted in better correlations between the questionnaire and more accurate measurements of diet (10, 13). We assumed that the relative ranking of participants with regard to intake of vegetables, fruit, and grains remained stable during surveillance. There is some assurance that this assumption may be appropriate (5, 25).

Another concern is that FFQs, out of practical necessity, may omit many foods (15). Vegetable and fruit intake was based on 9 items in the Japan Public Health study (7), 12 in the Swedish Mammography Screening Study (17), and 19 in the Breast Cancer Detection Demonstration Project (15). The more comprehensive the questionnaire, the more likely it will record intake of all vegetables and fruit. The Nurses’ Health Study had 15 fruits and 28 vegetables in its questionnaire (5). The Women’s Health Study used a 131-item questionnaire, which included ≥16 fruits and 34 vegetables (6). Our questionnaire recorded the intake of 15 fruits, 22 vegetables, and 15 items with fruit or vegetables in soups, mixed dishes, or desserts.

Past studies have noted that the amount of vegetables and fruit consumed may be insufficient to produce an effect, or the range of intake may be too narrow between the lowest and highest quintile groups. In the Breast Cancer Detection Demonstration Project, the authors stated that they and other investigators may not have observed an inverse association between vegetable intake and colorectal cancer because of the consumption of insufficient quantities to show a reduction in risk (15). In the Iowa Women’s Health Study, 29 vegetables and 15 fruits were in the questionnaire, but the researchers noted that the range of vegetable and fruit consumption was relatively narrow, which limited the ability to have contrasts between extremes of consumption (16).

Because of the comprehensiveness of our questionnaire, the recorded amounts were sizable. The median daily intake of vegetables and fruit was 293 and 259 g, respectively, for men and 286 and 298 g, respectively, for women. In addition, the range of intake of these foods was wide. There was a 3.4-fold difference in vegetable intake to a 9.8-fold difference in fruit intake between the lowest and highest quintile groups. In the Netherlands Cohort Study (4), the mean daily intake of vegetables and fruit was 187 and 154 g, respectively, for men and 191 and 196 g, respectively, for women. There was a 2.9-fold difference in vegetable intake to an 8.4-fold difference in fruit intake between the lowest and highest quintile groups.

Confounding by measured and unmeasured factors is another important concern in studies of diet and colorectal cancer (4, 5). Low levels of physical activity, cigarette use, and alcohol consumption have been linked to a low consumption of fruit and vegetables (26). These and unknown variables can have an impact on the relation of vegetables, fruit, and grains to colorectal cancer. Cohort studies, especially earlier ones, were not able to take into account many of these factors (4, 16, 23). We were able to control for a wide range of potentially confounding variables of colorectal cancer in the analyses, as was done in the Cancer Prevention Study II (3), the Women’s Health Study (6), and the Health Professionals’ Follow-Up Study and the Nurses’ Health Study (5).

In conclusion, we found an inverse relation of vegetable and fruit intake to colorectal cancer risk in men and not in women. The association appeared consistent across the different ethnic groups in the study and was stronger for colon cancer. Separation according to the different sources of vegetables and fruit did not identify specific food groups that accounted for the association. There was no association of grain intake with colorectal cancer for either men or women.