Fung T, Hu FB, Fuchs C, Giovannucci E, Hunter DJ, Stampfer MJ, Colditz GA, Willett WC. Major Dietary Patterns and the Risk of Colorectal Cancer in Women. Arch Intern Med. 2003;163(3):309-314. doi:10.1001/archinte.163.3.309
Copyright 2003 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2003
Several foods and nutrients have been implicated in the development of colon and rectal cancers. In this study, we prospectively assessed the associations between major dietary patterns and the risks of these 2 cancers in women.
Using dietary information collected in 1984, 1986, 1990, and 1994 from 76 402 women aged 38 to 63 years without a history of cancer in 1984, we conducted factor analysis and identified 2 major dietary patterns: "prudent" and "Western." We calculated factor scores for each participant and examined prospectively the associations between dietary patterns and colon and rectal cancer risks.
The prudent pattern was characterized by higher intakes of fruits, vegetables, legumes, fish, poultry, and whole grains, while the Western pattern, by higher intakes of red and processed meats, sweets and desserts, french fries, and refined grains. During 12 years of follow-up, we identified 445 cases of colon cancer and 101 cases of rectal cancer. After adjusting for potential confounders, we observed a relative risk for colon cancer of 1.46 (95% confidence interval, 0.97-2.19) when comparing the highest with the lowest quintiles of the Western pattern (P value for trend across quintiles, .02). The prudent pattern had a nonsignificant inverse association with colon cancer (relative risk for fifth quintile compared with the first, 0.71; 95% confidence interval, 0.50-1.00; P for trend across quintiles, .31). We did not observe any significant association between dietary patterns and rectal cancer.
We found a significant positive association between the Western dietary pattern and the risk of colon cancer.
NUMEROUS STUDIES have assessed the associations between individual foods or nutrients and risk of colorectal cancer,1- 7 but results have not always been consistent. In particular, fruits and vegetables have had either null or inverse association,5,8,9 and recent data on fiber suggest no protective effect.1,10,11 Diet can be studied in terms of nutrients, foods, and food groups. In addition, pattern analysis may provide additional insight that takes into account the combined effects of foods. Specific dietary patterns have been associated with risk of colon cancer in case-control studies12,13 and with coronary heart disease in prospective studies.14,15 The utility of analyzing diet as a set of patterns provides an alternative to the use of highly correlated food groups and nutrients. In addition, dietary patterns have been associated with fasting insulin levels,16 which may be related to colon cancer risk.17 In the present article, factor analysis18 was used to identify major dietary patterns among a large cohort of women. In our analysis, we prospectively assessed the associations between major dietary patterns and risks of colon and rectal cancers in a large cohort of women.
The Nurses' Health Study (NHS) began in 1976 when 121 700 female nurses aged 30 to 55 years in 11 US states responded to a questionnaire regarding medical, lifestyle, and other health-related information.19 The NHS is approved by the institutional review board of the Brigham and Women's Hospital, Boston, Mass. Since 1976, questionnaires have been sent biennially to update this information. Follow-up was complete for greater than 95% of the potential person-time up to 1994. In 1980, the participants completed a 61-item food frequency questionnaire (FFQ). In 1984, the FFQ was expanded to 116 items. Similar FFQs were sent to the women in 1986, 1990, and 1994. We used the 1984 FFQ as baseline for this study because the expanded number of items was critical in characterizing dietary patterns.
For the present analysis, women were included if they completed the 1984 FFQ with fewer than 70 missing items and a total caloric range (as calculated from the FFQ) between 500 and 3500 kcal/d. We excluded women with a history of cancer. We thus included in this analysis 76 399 women with follow-up of up to 12 years, from 1984 to 1996.
Dietary intake information was collected by FFQs designed to assess average food intake over the previous year. A standard portion size was given for each food item. Cohort members were asked to choose from 9 possible frequency responses, ranging from "never" to "more than 6 times a day" for each food. Total energy intake was calculated by summing up energy intakes from all foods. For this analysis, we used information from the FFQs administered in 1984, 1986, 1990, and 1994. Foods from the FFQ were classified into 36 to 38 food groups based on nutrient profiles or culinary usage. This classification follows that of a similar study in men.14 Foods that did not fit into any of the groups or that may represent distinctive dietary behaviors were left as individual categories (eg, pizza, tea, and beer). Vitamin and mineral supplements were not included in the definition of the patterns, but they were adjusted in the analysis. Previous validation studies among members of the NHS cohort revealed good correlations between nutrients assessed by the FFQ and multiple weeks of food records completed over the previous year.20 For example, correlation coefficients between the 1986 FFQ and diet records obtained in 1986 were 0.68 for saturated fat, 0.76 for vitamin C, and 0.73 for dietary cholesterol. Mean correlation coefficient between frequencies of intake of 55 foods from 2 FFQs 12 months apart was 0.57.21
Our end points included colon and rectal cancers that occurred between the return of the 1984 questionnaire and June 1, 1996. When a participant reported a diagnosis in the biennial questionnaires, we sought permission to review medical records for confirmation. These records were reviewed by physicians who had no knowledge of the participants' risk factor status. The histological type, anatomic location, and stage of cancer were recorded. We did not include carcinomas in situ. Deaths were reported by family members, the postal service, or through searches in the National Death Index.
Dietary patterns were generated by factor analysis (principal components) based on predefined food groups using an orthogonal rotation procedure.22 This resulted in uncorrelated factors, which were easier to interpret. We determined the number of factors to retain by eigenvalue (>1), Scree test, and factor interpretability. The factor score for each pattern was calculated by summing intakes of food groups weighted by their factor loadings,18 and each woman received a factor score for each identified pattern. Good reproducibility of the patterns generated by this method has been demonstrated in a parallel cohort of men.23 The correlations between the scores of the 2 major patterns ("prudent" and "Western") generated from FFQ and diet records were 0.52 for the prudent pattern and 0.74 for the Western pattern. Factor analysis was conducted using SAS PROC FACTOR.24
Incidence rates were calculated using person-time.25 We used a pooled logistic regression to assess the associations between dietary pattern score and the risk of colon and rectal cancers.26,27 This method pools each 2-year period in the follow-up as independent blocks and is asymptomatically equivalent to the Cox proportional hazard model with low rates of disease during each time interval. To reduce random within-person variation and best represent long-term dietary intake, we calculated cumulative averages of dietary pattern scores from our repeated dietary measurements.28 For example, dietary intake in 1984 was used to predict colon cancer occurrence in 1984 to 1986, and the average of 1984 and 1986 intake was used to predict risk from 1986 to 1990 and so on. The regression analyses were adjusted for age (5 categories), family history of colorectal cancer (yes or no), body mass index (calculated as weight in kilograms divided by the square of the height in meters) (<22, 22-23.9, 23-24.9, 25-28.9, ≥29), aspirin use (yes or no), multivitamin use (yes or no), physical activity (5 categories), alcohol intake (0 g/d, 0.1-4.9 g/d, 5.0-14.9 g/d, 15.0-29.9 g/d, ≥30 g/d), and smoking (never, past, current smokers ≤24 cigarettes per day, current smokers >24 cigarettes per day, current smokers but unknown quantity), energy intake (quintiles), and failing to return a FFQ (indicator variables for each unreturned FFQ). The proportions with a missing FFQ in 1986, 1990, and 1994 were 17%, 16%, and 14%, respectively. In separate analyses, we examined associations between dietary patterns and colon and rectal cancers individually. We also conducted analyses jointly classifying women by major dietary patterns and family history of colorectal cancer, obesity status (body mass index >30), physical activity, multivitamin use, and smoking status.
Between 1984 and 1996, we documented 445 incident colon cancers and 101 cases of rectal cancer. We identified 2 major dietary patterns with the factor analysis procedure. The first pattern was characterized by high intakes of fruits, vegetables, whole grains, legumes, poultry, and fish (Table 1), and the second pattern was characterized by refined grains, processed and red meats, desserts, high-fat dairy products, and french fries. The characteristics of the patterns were stable across repeated FFQs. As with our previous studies, we labeled the first factor the prudent pattern and the second factor the Western pattern.23
As expected, a higher prudent pattern score was associated with healthy behaviors and a higher Western pattern score with less healthy behaviors (Table 2). Women with high prudent score tended to smoke less, exercise more, use multivitamins, and to consume more fiber, vegetables, fruits, and whole grains. On the other hand, those with a high Western score tended to smoke, use fewer multivitamins, and to consume more red and processed meats, refined grains, and to have a higher proportion of the energy intake from fat.
After adjusting for age, the prudent pattern had a nonsignificant inverse association (relative risk [RR] comparing top with bottom quintile, 0.83) with colon cancer, while the Western pattern showed a significant positive association (RR comparing fifth with first quintile, 1.39; 95% confidence interval [CI], 1.05-1.84; P value for trend, .01) (Table 3). The association with the prudent pattern was strengthened after adjusting for other risk factors and potential confounders, with an RR of 0.71 (95% CI, 0.50-1.00) comparing fifth with first quintile, though the test for trend was not significant. Similarly, the association for the Western pattern was also stronger after multivariate adjustment with an RR of 1.46 (95% CI, 0.97-2.19; P value for trend, .02) comparing extreme quintiles. The increase in risk was most apparent in the top quintile. Additional adjustment for total glycemic load of the diet29- 31 and cereal fiber intake did not alter the association.
We also assessed the associations of the dietary patterns with proximal and distal colon cancer separately. The RR for proximal colon cancer (194 cases) was 0.72 (95% CI, 0.42-1.25) comparing top with bottom quintile of the prudent pattern and 1.70 (95% CI, 0.91-3.17) for the Western pattern. For distal colon cancer (179 cases), the RRs comparing top with bottom quintile of the prudent pattern were 0.71 (95% CI, 0.42-1.21) and 1.57 (95% CI, 0.82-2.99) for the Western pattern. In an analysis stratified by known risk factors, the associations between the 2 dietary patterns and risk of colon cancer were generally consistent across different strata according to smoking status, body mass index, family history of colorectal cancer, physical activity, and vitamin use (data not shown).
No apparent association was observed between dietary pattern and rectal cancer risk, although the CIs were wide. After multivariate adjustment, the RR for women at the fifth quintile of the prudent pattern score was 1.17 (95% CI, 0.50-2.78) compared with women in the first quintile (P value for trend test, .87). The RR for the Western pattern comparing top with bottom quintile was 0.92 (95% CI, 0.37-2.28; P value for trend test, .60).
We observed a statistically significant positive association between the Western pattern and colon cancer and a suggestion of an inverse association with the prudent pattern. These results are in agreement with previous studies that used dietary patterns to predict colon cancer risk.12,13 In a case-control study in 1998, Slattery et al12 used factor analysis and identified several dietary patterns. The components of the Western and prudent diet in that study and their associations with colon cancer were similar to the Western and prudent patterns in our analysis. The Western pattern identified by Slattery et al was characterized by higher intakes of processed and red meats, eggs, fast food meats, butter, coffee, and potatoes. Women at the top quintile of this pattern had an odds ratio for colon cancer risk of 1.49 (95% CI, 1.05-2.12) compared with the bottom quintile. The prudent pattern identified was characterized by higher intakes of fruits, vegetables, legumes, and fish. Comparing top with bottom quintile, the odds ratio for colon cancer was 0.73 (95% CI, 0.55-0.97) among women. Similar dietary patterns were identified in a Swedish cohort of women.32 The
"healthy" pattern was characterized by higher intakes of fruits and vegetables, low-fat dairy products, whole grains, cereals, and eggs, and the "Western" pattern characterized by higher intakes of meats, refined grains, soda, and potatoes. However, neither pattern was significantly associated with colorectal cancer risk. In a case-control study by Randall et al,13 somewhat different dietary patterns, probably owing to different groupings of foods, were generated separately for men and women. In men, the "traditional" pattern (characterized by items such as beef, cakes, potatoes, and cabbage), "snack" pattern (characterized by desserts, sweets, and hamburger), and the "high-fat" pattern (characterized by items such as alcohol, eggs, and processed meats) were positively associated with colon cancer risk (odds ratios ranged from 1.26-1.32). In women, the "salad" pattern (characterized by salad vegetables),
"light" pattern (characterized by items such as fish, hard cheese, yogurt, and broccoli), and the "whole grain" pattern (characterized by whole grain products, nuts, and dried fruits) were inversely associated with colon cancer risk (odds ratios ranged from 0.67-0.80).
Several components of each pattern may contribute to our findings. Data from several populations have suggested a positive association between red meats (which loaded strongly in our Western pattern) and colon or rectal cancer.33- 36 A small cross-sectional study suggested that higher red meat intake may be associated with higher colorectal mucosa proliferation among those with a previous history of colon adenomas.37 Some, but not all, studies also pointed toward processed meat intake, another major component in the Western pattern, as a risk factor for colorectal cancer.33,35,36 In a recent meta-analysis, a 100-g/d increase in meat or red meat intake was associated with a 12% to 18% increased risk for colorectal cancer38 and a 25-g/d increase in processed meat intake was associated with a 49% increase in risk. Higher folate intake has been inversely associated with the risk of colon cancer.2 Therefore, the inverse, albeit nonsignificant association of the prudent diet with colon cancer may be partly attributed to the higher intake of folate among those with higher prudent pattern score. Folate is a methyl donor, and in rodent models, methyl-deficient diet may result in hypomethylation and therefore loss of regulation of proto-oncogenes.39 On the other hand, epidemiological data on the relationship between intake of fruits and vegetables, major components of the prudent diet, and colorectal cancer are less consistent. A pooled analysis of this cohort and the Health Professionals Follow-up Study showed no appreciable association with fruits and vegetables,5 whereas previous studies shown an inverse association with fruits or specific types of vegetables.6,8,40 Recently, evidence has pointed to the possible influence of insulin and insulinlike growth factor as mediators of colorectal cancer development.17,41 We have previously shown in men that the prudent pattern was inversely associated and the Western pattern positively associated with fasting insulin and C-peptide levels,16 which suggests the dietary patterns may act in part through insulin resistance.
Because this analysis used prospective data with a high rate of follow-up, recall and selection bias is highly unlikely. We have also controlled extensively for possible confounders. There is no standard method to identify dietary patterns. Instead of using predetermined patterns such as the Mediterranean diet, our purpose was to identify actual eating patterns in our cohort and examine their ability to predict disease risk. Although dietary patterns generated by factor analysis are not based on a priori concepts, the patterns that we generated in this manner were, in general, similar to those identified in other populations.12- 14 However, food consumption patterns can be population specific and may change over time owing to food availability and preferences. Therefore, repeated measurement of food intake in our study is an advantage. Nevertheless, because of the observational nature of our study, a causal relationship cannot be established based on our findings alone.
In conclusion, we found that a diet high in red and processed meats, refined grains, and other characteristics of the Western pattern was associated with a higher risk of colon cancer in women. On the other hand, a diet high in whole grains, fruits and vegetables, and other characteristics of the prudent pattern may be inversely associated with colon cancer. Our study provides further evidence that switching from a typical Western diet to a more prudent diet may reduce the risk of colon cancer.
Corresponding author: Teresa Fung, ScD, Programs in Nutrition, Simmons College, 300 The Fenway, Boston, MA 02115 (e-mail: email@example.com).
Accepted for publication June 13, 2002.
This work was supported by grant CA87969 from the National Institutes of Health, Bethesda, Md, and by the National Colorectal Cancer Research Alliance, Studio City, Calif.
The results of this study were presented at the Congress of Epidemiology (Society for Epidemiological Research), Toronto, Ontario, June 17, 2001.