Dietary Patterns, Food Groups, and Rectal Cancer Risk in Whites and African-Americans

Colorectal cancer is the third most common cancer in the United States among men and women (1). Incidence and mortality rates are highest among African-Americans compared with other U.S. race/ethnic groups. Although some of this disparity can be attributed to access to care and socioeconomic differences (2), other reasons remain largely unknown. It is generally accepted that diet plays an etiologic role in colorectal cancer development; however, studies examining associations of specific foods and nutrients with colorectal cancer risk have been inconsistent. Moreover, most studies have focused on colon cancer only or the combination of colon and rectal cancer, whereas less attention has been given to the risk for rectal cancer specifically.

Most diet and cancer studies examine associations of individual nutrients with disease risk. Examining individual nutrients in relationship to cancer risk is beneficial for gaining insight into possible mechanisms of dietary components. This individual nutrient approach, however, is not adequate for considering the synergistic effect of highly correlated nutrients and other compounds found in foods (3). Other studies have focused on food groups, which take into account the way the foods are typically consumed. Nonetheless, it has been suggested that dietary patterns represent a more logical approach because the analysis of dietary patterns takes into consideration the synergistic effect of foods and nutrients, neither of which is consumed in isolation. Dietary patterns include numerous dietary exposures and are often associated with other health behaviors such as physical activity, smoking, and cancer screening (4). A common approach to identifying dietary patterns is factor analysis, which reduces a large number of variables into a small number of factors based on their degree of correlation (5). These factors then represent dietary patterns in the study population and are used as predictors in subsequent analyses of risk. Comparisons between the food/nutrient and dietary pattern approaches among previous studies have been difficult because of differences in study design, study populations, and statistical methods.

As noted above, few studies have examined associations of diet with rectal cancer risk separately because most have combined rectal and colon cancers. However, true mechanisms underlying the etiology of colon and rectal tumors may be different (6, 7). The objective of this work is to examine associations of food groups and dietary patterns (based on factor analysis) with risk for rectal cancer in a population-based case-control study on non-Hispanic Whites (Whites) and African-Americans in North Carolina. To our knowledge, this is the first population-based study to examine these relationships in a racially diverse U.S. population.

The North Carolina Colon Cancer Study—Phase II is a population-based case-control study in a 33-county area in central North Carolina. These counties include rural, suburban, and urban areas and are socioeconomically diverse. Participants were selected using a randomized recruitment strategy that oversampled African-Americans and involved matching on 5-y age, sex, and race. This study was approved by the institutional review board of the University of North Carolina.

Rectal cancer cases were identified by the North Carolina Central Cancer Registry rapid ascertainment system and included those with cancers of the rectum, sigmoid, and rectosigmoid junction (International Classification of Diseases 154). Eligibility criteria for cases included ages 40 to 79 y at time of diagnosis, diagnosed with a primary adenocarcinoma between May 2001 and September 2006, have a North Carolina driver's license or identification (because controls younger than 65 y were selected from Department of Motor Vehicle rosters), and able to give informed consent and complete the interview. All diagnoses were confirmed by the study pathologist through review of pathology slides and reports. Cases with a noninvasive carcinoma or a previous diagnosis of colorectal cancer were excluded. After notification of the primary physician, eligible cases were sent a letter describing the study and a race-matched enrollment specialist contacted them to explain the study and obtain their consent to participate. Interviews were scheduled for consenting cases. There were a total of 1,831 cases sampled, 1,417 of whom were eligible to participate. Of the eligible cases, 118 (8%) were unable to be contacted, 242 (17%) refused, and 1,057 (75%) were interviewed. The response rate (number of persons interviewed divided by the total number of eligible persons) was 76% and 70% for Whites and African-Americans, respectively.

Using lists provided by the agencies, controls were randomly identified from the North Carolina Department of Motor Vehicles (for those younger than 65 y) and from the Center for Medicaid and Medicare Services (formerly known as the Health Care Financing Administration) for those 65 y and older. Eligible controls were 40 to 79 y old at the time of selection, resided in the 33-county study area, and had no previous diagnosis of colorectal cancer. Similar to cases, potential controls were sent an introductory letter and contacted by a race-matched enrollment specialist, and in-person interviews were scheduled for controls who agreed to participate. Among eligible controls (1,827 of 2,345 sampled), 325 (18%) could not be contacted, 483 (26%) refused, and 1,019 (56%) were interviewed. The response rates were 58% and 46% for White and African-American controls, respectively.

All data were collected by trained nurse interviewers in the home of participants or other convenient location.

Usual dietary intake was assessed using the Diet History Questionnaire developed and tested for validity by the National Cancer Institute (NCI) (8-10). This instrument was validated in study samples that were racially diverse, with African-Americans representing 10% to 14% of these study samples (9, 10). The Diet History Questionnaire consists of 124 separate food items and assesses the frequency of consumption and portion size consumed for each food item. Participants were asked to estimate their food and beverage intake in the past 12 mo. The 12-mo period was chosen to take into account seasonal variation in food consumption. Cases were asked to estimate their usual frequency and portion size over the 12-mo period before diagnosis, and controls were asked to estimate consumption during the 1 y before interview. Daily intakes of nutrients and total energy were calculated with software provided by the NCI and developed for the survey instrument. Nutrient intakes were determined using the frequency of consumption, reported portion size, and nutrient content.

For the food group analysis, we examined the following U.S. Department of Agriculture pyramid food groups (11): total grains, whole grains, nonwhole grains (e.g., white bread, pasta, cereal), total vegetables, dark-green vegetables, deep-yellow vegetables, dry beans and peas, white potatoes, starchy vegetables, tomatoes, other vegetables (e.g., cabbage, cauliflower, brussels sprouts, onions), total fruits, citrus fruits (including melons and berries), other fruits, total dairy, milk, yogurt, cheese, total meat, beef/pork/lamb (that is, red meat), processed meats, organ meats, fish and other seafood, poultry, eggs (that is, eggs, egg whites, egg substitutes), soy products, nuts (e.g. peanuts, walnuts, seeds), added sugar (sugars added during processing, cooking, or at the table), and discretionary fat (that is, excess fat in foods and fat added to foods). Average weekly intakes were calculated for each food group. There was a large proportion of nonconsumers for the yogurt, organ meat, and soy food groups (58%, 49%, and 76%, respectively). For this reason, we dichotomized (consumers versus nonconsumers) these foods in the food group analysis, combined the yogurt group with the milk food group, and excluded the organ meat and soy food groups in the factor analysis.

The participant questionnaire queried age at diagnosis, sex, race, education, annual income, use of nonsteroidal anti-inflammatory drugs, smoking history, and first-degree family history of colorectal cancer.

The analyses were restricted to participants who completed all components of the study (n = 1,987). Participants with unreliable reported energy intakes (<800 and >5,000 kcal/d for men and <600 and >4,000 kcal/d for women) were also excluded (n = 83; 50 men, 33 women) because they were considered implausible based on daily energy requirements (12). Thus, the analytic sample for this report included 1,520 Whites (720 cases, 800 controls) and 384 African-Americans (225 cases, 159 controls).

Descriptive statistics (means, SDs, and frequencies) were computed for all study variables by case-control status and race to describe the demographic and dietary characteristics of the study population. Results were stratified by race because tests for interaction indicated the presence of effect modification by race for some of the demographic and dietary variables. Each food group was categorized into quartiles based on the distribution among race-specific controls. Odds ratios and 95% confidence intervals (95% CI) were calculated using unconditional logistic regression models to determine the association between the food groups and rectal cancer risk. These food group models included an offset term to account for the randomized recruitment and to allow us to obtain unbiased odds ratios, as well as the following covariates: age (continuous), sex, socioeconomic status [represented by education (less than or equal to high school, some college, college graduate/advanced degree) and income (categorized)], body mass index (BMI) 1 y ago (that is, in the year before interview for controls and diagnosis for cases; normal, overweight, obese), physical activity (continuous), family history (yes, no), nonsteroidal anti-inflammatory drug use (yes, no), and total energy intake (continuous).

Dietary patterns were identified separately among White and African-American controls using 21 predefined food groups in a principal components factor analysis. This analysis was conducted using the PROC FACTOR procedure in SAS. To determine the number of factors to retain, we considered eigenvalues >1, the scree plot, and the interpretability of the factors. Extraction of these factors was followed by orthogonal rotation (the varimax rotation option in SAS) to obtain uncorrelated factors and enhance interpretability. For each dietary pattern (factor), a factor score was calculated for cases and controls by summing intakes of the food items weighted by their factor loadings. Pearson and Spearman r's were used to examine the correlation of factor scores for each dietary pattern with other participant characteristics and dietary variables. Partial Pearson r's adjusted for energy were obtained for the dietary variables. Factor scores were categorized into quartiles based on the distribution in the control population for African-Americans and Whites separately. To determine the relationship between these dietary patterns and rectal cancer, we used unconditional logistic regression models to obtain odds ratios and 95% CIs. Test for trend was conducted by incorporating a variable for the median values of factor scores among race-specific controls observed for each food group quartile into a logistic regression model. The trend test was weighted by the inverse of the variance for the quartiles. All logistic regression models were adjusted for the same covariates as in the food group models.

All analyses were done using SAS 9.1 software (SAS Institute, Inc.). Statistical tests were two-sided, and P < 0.05 was considered statistically significant.

The distribution of cases and controls by race is shown in Table 1. Among Whites, controls were older and more educated, had a slightly lower mean BMI 1 year ago, and used more nonsteroidal anti-inflammatory drugs than cases. Among African-Americans, the mean age was less for cases than controls whereas cases had a higher mean BMI. In addition, a larger proportion of African-American cases had a family history of colorectal cancer compared with controls. All of these participant characteristics were significantly associated with the risk for rectal cancer in multivariate models, except annual income, smoking status, and family history (data not shown).

Tables 2 and 3 give the covariate-adjusted race-specific odds ratios and 95% CIs for each food group among Whites and African-Americans, respectively. The odds ratios presented are not mutually adjusted for the other food groups, although estimates were similar when we controlled for the other primary food groups. For Whites, high intakes of nonwhole grains and white potatoes were significantly positively associated with rectal cancer risk (Table 2). Conversely, fruit, dark-green vegetables, deep-yellow vegetables, other starchy vegetables, other vegetables, dairy foods, fish, and poultry were significantly associated with reduced risk for rectal cancer. High consumption of dark-green vegetables had the strongest inverse association (odds ratio, 0.41; 95% CI, 0.29-0.58). The highest quartile of red meat intake had an odds ratio <1 but was not statistically significant. High intake of other fruit and added sugar were associated with elevated risk in African-Americans (odds ratio, 3.25; 95% CI, 1.52-6.96 for other fruit; odds ratio, 2.65; 95% CI, 1.11-6.34 for added sugar; Table 3). There was a significant lower risk associated with the second quartile of intake of total vegetables, other vegetables, total meat, and discretionary fat in African-Americans.

Three dietary patterns were identified separately among White and African-American controls using principal components analysis. These three patterns explained 39% of the variance in Whites and 43% of the variance in African-Americans. Table 4 presents the factor loadings for the food groups on each dietary pattern for each race group. The first dietary pattern, high fat/meat/potatoes, was similar for both Whites and African-Americans and had strong positive loadings for discretionary fat, nonwhole grains, white potatoes, red and processed meat, cheese, and added sugar. The second and third factors were only slightly different for Whites and African-Americans. For Whites, the second dietary pattern was characterized by high loadings of most vegetables, as well as fish and poultry, and was therefore labeled the vegetable/fish/poultry pattern. The third dietary factor in Whites was labeled fruit/whole grain/dairy because of its high positive loadings of fruit, whole grains, and milk/yogurt. In African-Americans, fruits also loaded heavily on the second factor in addition to vegetables. This factor was labeled fruit/vegetables. The third factor in African-Americans had strong loadings of nuts, beans and peas, and milk/yogurt, and was labeled legumes/dairy.

Table 5 shows correlations of the three separate dietary patterns in Whites and African-Americans with selected participant characteristics and dietary variables. Age was inversely correlated with the high fat/meat/potatoes pattern for both race groups, whereas education and income were positively correlated with the vegetable/fish/poultry pattern in Whites. The dietary variables presented are those related to energy intake (that is, total energy, fat, carbohydrate, protein, alcohol). Folate and fiber were included because of their high content in fruits and vegetables. The high fat/meat/potatoes pattern had the highest correlation with total energy in Whites (r = 0.86) and African-Americans (r=0.82) whereas inversely related to carbohydrates, alcohol, folate, and fiber. The vegetable/fish/poultry pattern in Whites had a strong positive correlation with protein, and the fruit/vegetable pattern in African-Americans was highly correlated with folate and fiber.

Associations (odds ratios and their 95% CI) of the dietary patterns (according to quartiles of factor scores) with rectal cancer risk, stratified by race, are given in Table 6. Estimates based on race-specific quartile cut points are shown, although similar associations were observed when quartile cut points were matched across ethnic groups. Among Whites, high factor scores for the high fat/meat/potatoes pattern had odds ratios suggestive of elevated rectal cancer risk (odds ratio, 1.84; 95% CI, 1.08-3.15). The second and third patterns in Whites were significantly associated with reduced risk for rectal cancer. The odds ratios for the highest quartiles for the vegetable/fish/poultry and fruit/whole grain/dairy patterns were 0.47 (95% CI, 0.33-0.67) and 0.65 (95% CI, 0.45-0.93), respectively. In African-Americans, the high fat/meat/potatoes and legumes/dairy patterns were suggestive of reduced risk whereas the fruit/vegetables pattern suggested elevated risk. None of the quartile estimates reached statistical significance. There was, however, evidence of a positive linear trend for the fruit/vegetables pattern and an inverse dose-response for the legumes/dairy patterns (P < 0.0001 for both). We did not observe any effect modification by gender for any of the food group totals or dietary patterns.

This population-based case-control study examined the relationship of food groups and dietary patterns with the risk for rectal cancer in Whites and African-Americans. High intakes of fruit, vegetables, and dairy were associated with reduced rectal cancer risk in Whites, whereas African-Americans had an elevated risk associated with other fruit and added sugar. We identified three major dietary patterns and investigated the relationship between these patterns and rectal cancer. The first dietary pattern, high fat/meat/potatoes, was similar for Whites and African-Americans, whereas the other two patterns differed slightly. To our knowledge, this is the first study to examine these associations in African-Americans.

Increased consumption of whole grain foods, as well as fruit, vegetables, and dairy products, has generally been associated with reduced colon and rectal cancer risk in epidemiological studies (13-15), although results have not been entirely consistent. The potentially protective role of these food groups has been attributed to their fiber content and micronutrients such as vitamins, carotenoids, calcium, and folate (16-18). Our study showed that fruit, some vegetables, and dairy foods were associated with reduced risk in Whites. Our findings support evidence from a case-control study by Slattery et al. (15) that reported significant rectal cancer risk reductions for high consumption of fruit and vegetables in a predominantly White population.

The relationship between fruit and vegetables and rectal cancer risk in our study varied by race and food subgroups. Contrary to our results, which showed risk reductions associated with specific fruits and vegetables among Whites, Michels et al. (19) did not find a protective effect for total fruit and vegetable intake or any subgroups of fruit and vegetables on colon and rectal cancer incidence. High other fruit consumption in African-Americans correlated with significantly higher risk for rectal cancer. This strong positive association remained after adjustment for other dietary variables such as citrus fruit, added sugar, and total carbohydrate intake. The elevated risk may be due to high intakes of high-calorie fruit juice or low intakes of fresh fruit.

Interestingly, high intake of red meat in our study population was not significantly associated with rectal cancer risk. It has been hypothesized that the high heme iron content in red meat enhances free radical production and tumor cell proliferation (20, 21) and that the fat content of red meat may increase the production of bile acids, also causing cellular proliferation (22). Some studies have shown elevated rectal cancer risk to be associated with high consumption of red meat (23, 24) and processed meat (24, 25). Our results are in agreement with findings by Wei et al. (7), which also showed no association between increased consumption of red meat and rectal cancer risk, although our findings do suggest elevated risk for high intake of processed meat in Whites.

Fish and other seafood may play an important role in rectal cancer risk reduction perhaps because of their rich omega-3 polyunsaturated fatty acid content, which may reduce the production of pro-inflammatory eicosanoids (26, 27). Although the effects of fish and poultry on colon and rectal cancer risk have been examined less often compared with red meat, at least five studies have shown fish and poultry to be associated with reduced risk for colorectal cancer (23, 24, 28-30). Three of these studies reported an inverse relationship between these food groups and risk for rectal cancer specifically (23, 24, 28), as we did in this present study among Whites. Fish and poultry had a nonsignificant positive association with risk in African-Americans, which may reflect how these foods were prepared. However, the results did not change when we adjusted for total fat intake.

Three dietary patterns were identified separately among White and African-American controls using principal components factor analysis. The high fat/meat/potatoes dietary pattern was similar in both race groups. Comparable dietary patterns in some cohort studies have found no association of this pattern with colon or rectal cancer risk (31, 32). However, other studies in which this pattern was labeled “Western” and “red meat” have reported significant elevated risk for colon cancer and colorectal cancer, respectively (4, 33). Our results for Whites are consistent with these findings because high factor scores among Whites for the high fat/meat/potatoes pattern were associated with elevated risk.

In addition to a type of Western dietary pattern, researchers have often identified a presumably healthy pattern that has been labeled “healthy,” “prudent,” and “vegetable” patterns in some studies (34-38). Among Whites in our study, potentially healthy patterns emerged as two distinct dietary patterns, that is, the vegetable/fish/poultry and the fruit/whole grain dairy patterns; both were associated with reduced risk for rectal cancer. Interestingly, the vegetable/fish/poultry pattern had weak factor loadings for fruits and dairy products, and the fruit/whole grain dairy pattern had only weak to modest loadings for most vegetables. This suggests that it may not be appropriate to combine fruit and vegetables as an individual food group. In African-Americans, the two presumably healthy patterns were the fruit/vegetables pattern and the legumes/dairy pattern. There was a positive linear relationship between the fruit/vegetable pattern and rectal cancer. This could be due to the heavy loadings of fruit, especially citrus fruit, which also showed a significant positive trend in risk in the food group analysis. The legumes/dairy pattern in African-Americans suggested a protective effect on risk, as was expected.

These dietary patterns only accounted for 39% and 43% of the total variance in Whites and African-Americans, respectively, which suggests that other patterns exist. There were a total of five factors in Whites and seven factors in African-Americans that had eigenvalues >1.0, and together, these factors explained 50% and 65% of the variance, respectively. However, these factors not presented were difficult to interpret. The low proportion of variance explained by the three factors in each race group could also be due, in part, to the limited number of foods entered in the factor analysis or a reflection of the overall complexity of the diet.

Our findings provide evidence that rectal cancer risk differs between African-Americans and Whites for certain foods and dietary patterns. Unfortunately, there are virtually no studies on diet and colon and rectal cancer associations in African-Americans. Similar racial differences were reported by Satia-Abouta et al. (14) in a population-based study on food groups and colon cancer. Few studies have conducted comparisons of dietary patterns for Whites and African-Americans (39-41). The dietary patterns in our study were similar to those identified in the Multiethnic Cohort Study, which also used the U.S. Department of Agriculture food groups for the factor analysis (41). Bell and colleagues (39) reported that food patterns among Whites and African-Americans did not differ. Although the patterns were generally similar in both race groups in our study, there were some different associations with rectal cancer risk. The observed heterogeneity in risk may in part be due to racial variation in dietary intake of certain foods and nutrients, as reported in some studies (42-44). We used race-specific cut points for food groups and dietary patterns to account for possible differences in consumption, although this could have affected our assessment of racial differences in risk. We cannot exclude the possibility that socioeconomic status contributes to this racial disparity; however, we controlled for education and income in our analyses.

Our study has many strengths, including its population-based design and the inclusion of a large number of rectal cancer cases. In addition, the randomized recruitment strategy used to select participants minimized the possibility of selection bias in our results. Oversampling allowed us to increase the number of African-Americans in our study sample in an effort to assess racial differences. Food group analysis and factor analysis were examined in the same population and included the same covariates.

There are also some limitations to our study. The use of predefined food groups in the factor analysis may have introduced error in our risk estimates. Grouping foods prevents the food items within the group from having different loadings on the dietary patterns identified and may obscure differences in consumption. However, the consistent use of food groupings may enable us to better compare studies on dietary patterns. Food frequency questionnaires, like that used in this study, are subject to measurement error and may not have included some typically consumed Southern foods (45) or foods common to certain races/ethnicities. Due to our case-control study design, recall bias is a possibility. Response bias may also have been introduced in our study, especially because the response rate was lower among African-Americans than Whites and lower among controls compared with cases. Although we oversampled African-Americans, the sample size for this subpopulation was relatively small (n = 384). This resulted in less power to detect significant associations in African-Americans and unstable risk estimates. Therefore, these findings should be interpreted with caution and need to be confirmed in a larger sample of African-Americans.

In summary, this study used two different approaches to investigate the relationship between diet and rectal cancer risk: food group analysis and factor analysis. Our results showed that several food groups and dietary patterns are associated with rectal cancer risk. Some of the food groups yielded different associations with risk than the overall pattern with which it was highly correlated. Complex correlations between foods may be better captured by dietary patterns, which may also prove to be more amenable to translation into dietary recommendations and easier to apply to improve the efficacy of nutrition intervention and prevention programs. Notably, our results suggest that dietary risk factors may differ by race, which highlights the importance of examining diet and cancer associations in racially diverse study populations.

No potential conflicts of interest were disclosed.

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

We thank Dr. Marilyn Tseng for her contribution in the dietary pattern analyses.