ORIGINAL CONTRIBUTIONS. Eating Patterns and Risk of Colon Cancer

Transcription

1 American Journal of Epidemiology Copyright 1998 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 18, No. 1 Printed in U.S.A. ORIGINAL CONTRIBUTIONS Eating Patterns and Risk of Colon Cancer Martha L. Slattery, 1 Kenneth M. Boucher, 1 Bette J. Caan, John D. Potter, and Khe-Ni Ma 1 Colon cancer has been associated with several nutrients and foods. The authors used data from a population-based study conducted in Northern California, Utah, and Minnesota to examine associations between dietary eating patterns and risk of developing colon cancer. Through factor analysis, detailed dietary intake data obtained from 1,99 cases (diagnosed in ) and,10 controls were grouped into factors that were evaluated for relations with lifestyle characteristics and colon cancer risk. Several dietary patterns emerged. The dietary patterns with the most variation were labeled "Western," "prudent," "high fat/sugar dairy," "substituters," and "drinkers." The "Western" dietary pattern was associated with a higher body mass index and a greater intake of total energy and dietary cholesterol. The "prudent" pattern was associated with higher levels of vigorous leisure time physical activity, smaller body size, and higher intakes of dietary fiber and folate. Persons who had high scores on the "drinker" pattern were also more likely to smoke cigarettes. The "Western" dietary pattern was associated with an increased risk of colon cancer in both men and women. The association was strongest among people diagnosed prior to age 67 years (for men, odds ratio (OR) = 1.96, 95% confidence interval (Cl) ; for women, OR =.0,95% Cl ) and among men with distal tumors (OR =.5, 95% Cl ). The "prudent" diet was protective, with the strongest associations being observed among people diagnosed prior to age 67 years (men: OR = 0.6, 95% Cl ; women: OR = 0.58, 95% Cl ); associations with this dietary pattern were also strong among persons with proximal tumors (men: OR = 0.55, 95% Cl ; women: OR = 0.6, 95% Cl ). Although "substituters" (people who substituted low fat dairy products for high fat dairy products, margarine for butter, poultry for red meat, and whole grains for refined grains) were at reduced risk of colon cancer, the reduction in risk was not statistically significant. These data support the hypothesis that overall dietary intake pattern is associated with colon cancer, and that the dietary pattern associated with the greatest increase in risk is the one which typifies a Western-style diet. Am J Epidemiol 1998; 18:-16. colonic neoplasms; diet; eating; factor analysis, statistical; food habits; meat; vegetarianism Editor's note: An invited commentary on this article appears on page 17 and is followed by the authors' response. Numerous components of dietary intake have been associated with colon cancer in several populationbased case-control and cohort studies (1). While both nutrients and individual foods have been assessed for evaluation of their association with colon cancer, few Received for publication January 0,1997, and in final form June, Abbreviations: CARDIA, Coronary Artery Risk Development in Young Adults; NSAIDs, nonsteroidal antiinflammatory drugs. 1 Department of Oncological Sciences, University of Utah Medical Center, Salt Lake City, UT. Division of Research, Kaiser Permanente Medical Care Program, Oakland, CA. Fred Hutchinson Cancer Research Center, Seattle, WA. Reprint requests to Dr. Martha L. Slattery, University of Utah Medical Center, 56 Chipeta Way, Suite 1100, Salt Lake City, UT attempts have been made to identify a broader eating pattern which may be associated with colon cancer risk (, ). Nonetheless, it is intuitive that eating patterns may alter risk of disease, since foods (and especially nutrients) are generally not eaten in isolation, and the overall pattern of a diet may have a greater effect on health than any one food. However, a major difficulty in studying eating patterns is the difficulty of establishing a classification system for characterization of usual intake. Food frequency questionnaires, which are generally used to estimate individual consumption of certain foods and nutrients, do not directly lend themselves to studies of broader eating patterns. Thus, attempts at studying eating patterns have generally used statistical methods to characterize eating patterns (-6). Schwerin et al. () divided dietary data obtained from -hour recalls into 15 food groups. They then used factor analysis to define eating habits and eval- Downloaded from by guest on 0 November 018

2 Eating Patterns and Colon Cancer 5 uated how these habits related to the presence of biochemical and clinical signs of nutritional deficiency. Randall et al. (5) used food frequency data obtained from controls in a case-control study of colon cancer to generate seven dietary patterns for men and women which were then compared with other healthrelated behaviors such as cigarette smoking. Slattery et al. (6) used similar techniques to evaluate diet and lifestyle patterns in relation to blood pressure among monozygotic and dizygotic twin pairs. However, to our knowledge, these techniques have not previously been applied to dietary associations with colon cancer. In this study, we attempted to identify eating patterns that may alter risk of colon cancer. We did this by delineating major food groups with a detailed diet history questionnaire and translating these groups into identifiable dietary patterns using factor analysis. Dietary factors or eating patterns were then evaluated in conjunction with other lifestyle factors to determine whether a particular eating pattern could characterize risk of colon cancer beyond that described by any one food item. MATERIALS AND METHODS Study population Study participants were black, white, or Hispanic. They were recruited from 1) the Kaiser Permanente Medical Care Program of Northern California, ) an eight-county area in Utah (Davis, Salt Lake, Utah, Weber, Wasatch, Tooele, Morgan, and Summit counties), and ) the metropolitan Twin Cities area of Minnesota (Anoka, Carver, Dakota, Hennepin, Ramsey, Scott, and Washington counties). A rapid reporting system was used to identify cases, with the majority of cases being interviewed within months of diagnosis. Eligibility criteria for cases included diagnosis with first primary incident colon cancer (International Classification of Diseases for Oncology, second edition (7), codes 18.0 and ) between October 1, 1991, and September 0, 199, being between 0 and 79 years of age at the time of diagnosis, and being mentally competent to complete the interview. Patients with cancers of the rectosigmoid junction or rectum (defined as the first 15 cm from the anal opening) or with known familial adenomatous polyposis, ulcerative colitis, or Crohn's disease were not eligible. Of the cases invited to participate in the study, approximately 76 percent agreed. Methods used to identify controls have been reported previously (8). Controls included persons who were randomly selected from Kaiser Permanente membership lists, persons contacted through random digit dialing, and persons randomly identified from driver's license lists (in Utah and Minnesota) and Social Security lists (in Utah). Of all controls asked to participate, approximately 6 percent agreed. Reasons for nonparticipation have been given elsewhere (9). The major reason for nonparticipation among controls was refusal (9.5 percent of controls identified) and our inability to locate them (15.5 percent). Of those we were unable to locate, 88 percent were from Minnesota, where people obtaining a driver's license were added to driver lists but those who moved out of the state or no longer held a valid Minnesota driver's license were not deleted. A total of 1,99 cases and,10 controls with complete data were included in the results presented here. Data collection Data were collected from study participants by trained and certified interviewers using laptop computers. The study questionnaire was pretested in a randomly selected (by random digit dialing) group of Utah residents over age 50 years. The referent period which study participants were asked to recall was the calendar year years prior to the date of selection (date of diagnosis for cases or date of selection for controls). The interview took approximately hours to complete. Quality control methods used in the study have been described elsewhere (10, 11). Dietary intake. Dietary intake data were ascertained using an adaptation of the validated diet history questionnaire of the Coronary Artery Risk Development in Young Adults (CARDIA) Study (11-1). Participants were asked to report which foods were eaten (using brand names when possible), the frequency with which the foods were eaten, and the amounts of fat used in preparation of other foods. Threedimensional food models were used to help participants estimate their usual serving sizes. Cue cards were used to aid in the consistent identification of foods within broad categories. For certain categories that could comprise many different types of items (such as breakfast cereal), participants were asked to report the three mostly commonly eaten items. As part of the diet history, detailed information was obtained on foods eaten as additions to other foods (such as sugar added to cereal); standard amounts of additions were assigned per unit of the food item they accompanied. Nutrient values for specific foods were calculated using the University of Minnesota Nutrition Coordinating Center's nutrient database (version 19) (1). Food groups. Thirty-five separate food groups were used in these analyses to describe eating patterns. The food groups used were: processed meats (including hot dogs, luncheon meats, and sausages); organ meats; red meat; fish; poultry; commercially prepared Downloaded from by guest on 0 November 018

3 6 Slattery et al. fast-food meat (including any fast-food beef, chicken, or fish); eggs; butter; margarine; yogurt; low fat dairy products; high fat dairy products; liquor; beer; wine; tea; coffee; dried fruit; canned fruit; fresh fruit; fruit juice; vegetable juice; nuts (including peanut butter); legumes; potatoes; cruciferous vegetables; green salads; dark yellow vegetables such as carrots, yams, and tomatoes; other vegetables (including corn, peas, and green beans); whole grains; refined grains; added sugar; high sugar dairy products (including ice cream and dairy desserts); high sugar drinks (such as sodas, Hi-C (Coca-Cola Company, Atlanta, Georgia), fruit drinks, and Kool-Aid (General Foods Corporation, White Plains, New York)); and desserts (including cookies, candy, cakes, and pies). The food groups were developed by assigning each food item a standard serving size; the broader food group categories were chosen to represent areas that have been associated with colon cancer. The numbers of standard servings consumed from a food group were summed for each individual. Eating patterns. Eating patterns were developed using factor analysis, as described elsewhere (-6). Food items included in the factor analysis were those that loaded high (>) on at least one factor, or those for which the variability in the food item was largely explained by the factors generated. All individual food groups were used to determine patterns of intake that explained the widest spread in the data yet had face validity. Labeling of eating patterns was done on the basis of our interpretation of the data and does not represent a priori patterns of intake. However, the labels given were those that we believed were representative of the existing literature on diet and colon cancer. It should be kept in mind that these labels are arbitrary, and others may be equally suited to the data. Other data. Other data obtained and used in these analyses were: age at the time of diagnosis or selection; body mass index (weight (kg)/height (m ) for men and weight (kg)/height 5 (m 15 ) for women), as self-reported for the referent year; presence or absence of a first degree relative with colorectal cancer; use of aspirin and/or nonsteroidal antiinflammatory drugs (NSAIDs) on a regular basis (hereafter grouped together); usual number of cigarettes smoked per day; having had a sigmoidoscopy during the past 10 years; and long term vigorous leisure time activity (9). The body mass index of weight/height 15 was used for women because it has been shown to be more independent of height than weight/height among women (15). Physical activity performed at home and at leisure was ascertained using an adaptation of the validated CARDIA physical activity history (16, 17). Possible scores on the physical activity scale ranged from to 1. The scale was based on calculated levels of long term vigorous physical activity, where represents no long term vigorous physical activity and 1 represents high levels of vigorous physical activity for the referenced periods, 10 and 0 years previously (9). within the colon was classified as proximal (cecum through transverse colon) or distal (splenic flexure, descending colon, and sigmoid colon). Statistical methods Eating patterns were identified using the principal components program for factor analysis in SAS (version 6.11; SAS Institute, Cary, North Carolina). Although a common practice is to focus on factors with eigenvalues greater than 1.0, indicating that the factor describes more of the variability in the data than the average variable for any individual item within the factor, we focused on factors whose eigenvalues were greater than 1.5. We did this to limit the number of factors, as well as to better identify more meaningful factors. Within a factor, negative loading indicates that foods are inversely associated with the factor, while positive loading indicates a direct association with the factor. The larger the loading of a given food group to the factor, the greater the contribution of that food to a specific factor. Foods which loaded at or greater were considered to be making a contribution to the factor, although the value for meaningful factor loading is arbitrary. After a varimax rotation, factor scores were saved from the principal components analysis for each individual. All data presented here are from the varimax rotation. These scores were used for comparison with other lifestyle factors and to estimate associations with colon cancer. Because of the exploratory nature of factor analysis, we compared factors identified from the principal components factor analysis method with those obtained from the maximum likelihood method, as further verification of the appropriateness of the factors. In most instances, factors identified using the principal components method were almost identical to those obtained from the maximum likelihood method. While a pattern typified as "drinker" was identified by the principal components method, alcohol loaded most heavily on the "Western-style" dietary pattern by the maximum likelihood method. Because factors identified via the principal components method were similar to those generated from the maximum likelihood method, and because other researchers who have used factor analyses have used the principal components method, we present data generated from the principal components method. Age-specific analyses used the median age of the controls, 67 years, as the cutpoint. Factor scores were categorized into quintiles based on the distribution of the control population for men and women separately. Downloaded from by guest on 0 November 018

4 Eating Patterns and Colon Cancer 7 To determine the associations between dietary factors or eating patterns and colon cancer, we calculated odds ratios and 95% confidence intervals from unconditional logistic regression models. In these analyses, total energy intake, age at selection, body mass index, and long term vigorous physical activity were used as covariates, since they were associated with eating patterns and colon cancer risk in this population; these factors slightly confounded the observed associations. Other factors such as alcohol, dietary folate, calcium, dietary fiber, cigarette smoking, use of NSAIDs, and family history of colorectal cancer did not appear to confound the reported associations. RESULTS The majority of the study participants were over 60 years of age (table 1). Cases tended to have lower levels of income and education than controls. A slightly larger proportion of controls were members of the Church of Jesus Christ of Latter-day Saints, while a slightly larger proportion of cases reported being Catholic. Six dietary factors were generated for men (table ) and seven dietary factors were generated for women (table ) that fitted the criterion of an eigenvalue greater than 1.5. The patterns identified for men and women were similar, although the order of their importance varied, and in some instances specific food items did not load equally for men and women. Foods with a very high factor loading (arbitrarily set at 0.0 or greater) are underlined in tables and. For both men and women, the first dietary pattern was one which loaded food groups that may typify a "Western"-type diet: high levels of red meat, pro- TABLE 1. Description of the study population used in a factor analysis of dietary intake patterns and colon cancer risk, Sex Male Female Annual income (1990s dollars) <0,000 0,001-5,000 5,000-50,000 50,000-60,000 >60,000 Education Less than high school High school graduate Some college education College graduate Postoollege education Household size 1 > Religion Catholic Protestant Latter-day Saints (Mormon) Jewish Seventh-day Adventist Other religion No religious preference No. 1, , Cases (n = 1,99) % No. 1,90 1, , , Controls (n =,10) % X * * Downloaded from by guest on 0 November 018 ' Age and sex were matching variables in this study.

5 8 Slattery et al. TABLE. Assessment of the factor-loading matrix for male controls (n = 1,90) in a factor analysis of dietary intake patterns and colon cancer risk, * Processed meats Organ meats Red meat Fish Poultry Fast-food meats Eggs Butter Margarine Yogurt Low fat dairy foods High fat dairy foods Liquor Beer Wine Tea Coffee Dried fruit Canned fruit Fresh fruit Fruit juice Vegetable juice Nuts Legumes Potatoes Cruciferous vegetables Salad Carrots/tomatoes Other vegetables Whole grain Refined grain Added sugar High sugar dairy foods High sugar drinks High sugar desserts Proportion of variability (%) 1 ("Western" diet) QJ ("prudent" diet) Oil 8. ("Nghfal/ sugar dairy") QJ1 5. Factor ("drinker) r O ("substituter") ("tajlt juice") * Values of < were excluded from the table for simplicity; those with loadings of 0.0 or greater are underlined. cessed meat, fast food, refined grains, and sugarcontaining foods, and low levels of vegetables (other than potatoes) and fruits, with the predominant fruit being canned fruit. The second dietary pattern was one in which all types of fruits and vegetables were consumed and fish and poultry were eaten more often than red meat or processed meat; this dietary pattern was inversely associated with most high sugar foods and was designated "prudent." The third pattern, "high fat/sugar dairy," was one which loaded heavily on yogurt and high fat/high sugar dairy products. The "drinker" pattern, in which alcoholic beverages loaded highest, was the fourth dietary pattern we identified in men and the sixth pattern we identified in women. Fish and poultry loaded heavier in this pattern than did red meat, although red meat was also consumed, as were salads among men and coffee among women. We labeled factor 5 for men and factor for women the "substituter" pattern. This dietary pattern was one in which margarine appeared to be substituted for butter, whole grains were substituted for refined grains, low fat dairy products were substituted for high fat dairy products, and poultry was substituted for red meat. Although fruit was consumed by this group, it appeared that they avoided most vegetables and still consumed substantial amounts of added sugar. The remaining patterns, factor 6 for men and factors 5 and 7 for women, were more difficult to interpret. We arbitrarily classified factor 6 for men as "fruit juice"; for women, we classified factor 5 as "coffee.1 Downloaded from by guest on 0 November 018

6 Eating Patterns and Colon Cancer 9 TABLE. Assessment of the factor-loading matrix for female controls (n = 1,10) in a factor analysis of dietary intake patterns and colon cancer risk, * Processed meats Organ meats Red meat Fish Poultry Fast-food meats Eggs Butter Margarine Yogurt Low fat dairy foods High fat dairy foods Liquor Beer Wine Tea Coffee Dried fruit Canned fruit Fresh fruit Fruit juice Vegetable juice Nuts Legumes Potatoes Cruciferous vegetables Salad Carrots/tomatoes Other vegetables Whole grain Refined grain Added sugar High sugar dairy foods High sugar drinks High sugar desserts 1 ("Western" fprudert" ("high fat/ diet) diet) sugar dairy") QJ e^i Proportion of variability (%) Factors QZ ("coffee ^ ro r) * Values of <0 were excluded from the table for simplicity; those with loadings of 0.0 or greater are underlined. and roll." This factor in women was highest in butter and added sugar, although coffee, high fat dairy foods, and refined grains contributed to the pattern. The seventh dietary factor identified in women was not further explored, because it could not be adequately characterized. The names given above to these dietary patterns are those used to describe the dietary factors in the remainder of this paper. Evaluation of the proportion of variability accounted for by each factor showed that the "Western" diet accounted for the greatest spread in the data among both men and women, although the proportion of the variability accounted for by the composite of the dietary factors might still be considered low Among both men and women, those who had a higher score on the "Western" diet pattern tended to be younger and to have lower levels of income and education (data not shown). On the other hand, those who had a higher score for the "prudent," "substituter," or "drinker" dietary pattern were more likely to have higher levels of income and education. Among men, the "fruit juice" pattern was significant only for age, with younger people being more likely to score high on this pattern. Among women, the "coffee and roll" dietary pattern was more likely to be present among persons with lower incomes, although it was equally present among those with the lowest and highest levels of education. Comparison of these factors with lifestyle charac-.1 Downloaded from by guest on 0 November 018

7 10 Slattery et al. TABLE. Pearson correlation coefficients (r) for relations between dietary factors and other lifestyle characteristics in male controls (n = 1,90), Body mass index* Physical activity Daily dietary intake Energy Cholesterol Fiber Calcium Cigarette smokingt Family history of colon cancerf Use of aspirin/nsaidst Sigmoidoscopy screeningt, Western Prudent < High fat/ sugar dairy Dietary pattern Drinker O.00 Substituter Fruit juice * Weight (kg)/height? (m*). t Spearman correlation coefficients are presented for these variables. Positive associations indicate a higher category of smoking, a greater number of family members with a history of colon cancer, and less use of aspirin/ NSAIDs. $ NSAIDs, nonsteroidal antiinflammatory drugs. Data were available for 968 male controls. Positive associations indicate less screening. teristics was carried out for both cases and controls; data obtained from the controls are presented in tables and 5, although data obtained from cases were similar to those reported. Factor 1, the Western diet, was associated (as indicated by correlation coefficients) with higher body mass indices and a higher consumption of total energy and cholesterol than any of the other factors. Factor, the prudent diet, was associated with higher levels of physical activity, lower body mass index, and, as might be expected, higher levels of dietary fiber and folate. Among both men and women, the factor we identified as the drinker pattern was positively associated with cigarette smoking. Cigarette smoking was also positively associated with the Western diet and inversely associated with the substituter pattern and the fruit juice pattern TABLE 5. Pearson correlation coefficients (r) for relations between dietary factors and other lifestyle characteristics in female controls (n = 1,10), Body mass index* Physical activity Daily dietary intake Energy Cholesterol Fiber Calcium Cigarette smokingt Family history of colon cancerf Use of aspirin/nsaidsf,* Sigmoidoscopy screeningt, Western O.0 - Prudent High fat/ sugar dairy Dietary pattern Substituter Coffee and roll Drinker * Weight (kg)/height«t Spearman correlation coefficients are presented for these variables. Positive associations indicate a higher category of smoking, a greater number of family members with a history of colon cancer, and less use of aspirin/ NSAIDs. $ NSAIDs, nonsteroidal antiinflammatory drugs. Data were available for 98 female controls. Positive associations indicate less screening. 0. -O.0 -O Downloaded from by guest on 0 November 018

8 Eating Patterns and Colon Cancer 11 TABLE 6. levels of body mass index, long term physical activity, and total energy intake, by eating pattern, among male controls (n = 1,90), Dietary pattern Eating pattern (quintile) Western «Prudent High fat/ sugar dairy Drinker Substituter Fruit juice Body masslndexf Long term physical activity} Total energy intake (kcal/day) ,895,076,87,816, ,58,0,99,800, ,98,8,0,76,589, standard error. t Weight (kgyneight* (m*). \ Possible scores on the physical activity scale ranged from to 1 (9) (see text). observed in men. Family history of colorectal cancer was not associated strongly with any factor, nor was use of aspirin or other NSAIDs. The Western pattern was associated with less sigmoidoscopy screening in men. Evaluation of mean levels of body mass index, long term physical activity patterns, and total energy intake showed that those who had high factor scores for the Western diet had higher levels of body mass index and total energy intake and a lower level of physical activity (tables 6 and 7). Those with higher scores on the prudent diet pattern had higher levels of physical activity and energy intake, although there was no variation in body mass index across this group. Those who scored highest on the substituter pattern were more likely to have a higher level of physical activity than those with low scores on the substituter pattern. There was an increase in total energy intake associated with all patterns except the drinker and fruit juice patterns, where persons at both ends of the distribution had the highest levels of energy intake. Evaluation of dietary patterns with regard to risk of colon cancer showed that a Western-type diet increased the risk of colon cancer in both men and women (tables 8 and 9), although associations were generally stronger for men than for women. Associations were slightly stronger for people diagnosed at a ,006,9,0,567, ,669,50,56,78, ,067,7,6,515, younger age. A significant linear trend was observed for all subgroups except older women. The prudent dietary pattern was associated with decreased risk of colon cancer, with the strongest associations being observed for younger individuals and those with proximal tumors. Risk estimates for the substituter and drinker patterns are shown in tables 8 and 9, along with those for the Western and prudent dietary patterns. Although associations were generally inversely associated with the substituter pattern, they were generally weak in magnitude. The drinker pattern was not consistently associated with colon cancer, although there was a suggestion of a weak positive association among men with distal tumors. None of the other factors consistently altered risk. DISCUSSION Migrant studies have shown that as people move from areas of low incidence of colon cancer to areas of high incidence, the disease rates of the migrant populations become more like those of the host countries (18). A Western-style diet has often been hypothesized as contributing to these patterns, including a higher rate of distal tumors in developed countries as compared with developing countries. In investigation of this hypothesis, numerous individual foods and nutri Downloaded from by guest on 0 November 018

9 1 Slattery et al. TABLE 7. levels of body mass controls (n = 1,10), index, long term physical activity, and total energy intake, by eating pattern, among female Eating pattern (quintlle) Western * Prudent High fat/ sugar dairy Dietary pattern Substituter Coffee and roll Drinker Body mass Indexf Lang term physical activity}: Total energy Intake (kcal/day) , ,10, ,896 1,85 1,90,07, ,80 1,711 1,918,057,65 *, standard error. tweight (kg)/hekjhti-5 (mis). t Possble scores on the physical activity scale ranged from to 1 (9) (see text). ents have been studied; however, few attempts have been made to evaluate dietary patterns, which represent a broader picture of dietary consumption and may be more associated with risk than any individual food or nutrient. In this study, we identified dietary patterns which were associated with other lifestyle characteristics as well as with risk of colon cancer. A Western-style diet, along with other characteristics of developed societies such as less physical activity and a greater tendency to be overweight, has been suspected to contribute to higher rates of many chronic diseases (19). The dietary pattern which we labeled "Western" in this study was associated with the other relevant lifestyle characteristics of the developed world. The dietary pattern which we labeled "prudent" was associated with higher levels of physical activity and lower body size measurements. Thus, this dietary pattern may represent individuals who exhibit healthier behaviors. Persons with dietary patterns such as the "drinker" pattern were also more likely to smoke cigarettes. On the other hand, a "substituter" was more likely to have healthy behaviors such as sigmoidoscopy screening, which goes along with the dietary pattern of substituting low fat foods for high fat foods and whole grain products for refined grain products. Our ability to associate the dietary patterns we identified with other lifestyle characteristics helps us feel more <0.0' I ,79 1,759 1,90.077, I ,99 1,796 1,768 1,99, ,0 1,89 1,80 1,95, confident that we identified meaningful eating patterns. One question that we wished to address is, "Can dietary patterns characterize disease associations beyond that described by any individual food item?" The data suggest that they do. The literature is mixed in terms of the associations found between meats, sugar, poultry, and fish and colon cancer (1). In our study, we observed that consumption of animal products had a minimal effect on risk (E. Kampman et al., Wageningen Agricultural University (Wageningen, The Netherlands), unpublished manuscript). However, the composite eating pattern suggests that a more Western diet does increase risk beyond that observed for the individual items included in the dietary pattern. On the other hand, we observed that vegetables were protective, as was dietary fiber (0), which is similar to estimates of risk for the prudent diet. However, the colon cancer associations observed for the prudent dietary pattern were more consistent between the sexes, for age at diagnosis, and for tumor site within the colon than were individual foods or nutrients. The substituters, while adopting many healthy eating patterns, did not include vegetables as part of their dietary pattern and did not experience the reduced risk of colon cancer observed in those with the prudent diet. Thus, we believe that these data further support our previous observations that vegetable consumption is 1 Downloaded from by guest on 0 November 018

10 TABLE 8. Colon cancer risk associated with dietary eating patterns in men, * O 00 o 8CO Western diet No. of controls/no, of cases All subjects <67 67 Distal Proximal Prudent diet No. of controls/no, of cases All subjects <67 67 Distal Proximal youp) ORf 95% Clt 58/1 50/6 Drinker No. of controls/no, of cases 60/00 All subjects <67 67 Distal Proximal Substituter No. of controls/no, of cases 56/0 All subjects <67 67 Distal Proximal / / / / Quintile OR 95% Cl OR 60/ / / / / /1 56/ 95% Cl OR / /76 95% Cl / / / * Odds ratios were adjusted for age, body mass index, long term physical activity, and total energy intake, Numbers of subjects for age and tumor subsite are approximately half of those given for the total sample. t OR, odds ratio; Cl, confidence interval. p for trend m * to (0 0> Q. O o I Downloaded from by guest on 0 November 018

11 TABLE 9. Colon cancer risk associated with dietary eating patterns in women, * o 1 S-' iol o. CO Z O 1998 Western diet No. of controls/no, of cases All subjects <67 67 Distal Proximal Prudent diet No. of controls/no, of cases All subjects <67 :67 Distal Proximal Drinker No. of controls/no, of cases All subjects <67 67 Distal Proximal Substituter No. of controls/no, of cases All subjects <67 St67 Distal Proximal (reference group) /119 1/15 /175 19/16 ORt /166 /18 /17 8/0 95% Clf OR * Odds ratios were adjusted for age, body mass index, long term physical activity, those given for the total sample. t OR, odds ratio; Cl, confidence interval. 0/190 /175 /00 1/19 95% Cl Qulntlle OR /1 /155 1/18 /177 95% Cl and total energy intake. Numbers of subjects for age and tumor subsite are approximately half of OR /00 /161 5/156 /15 95% Cl p (or trend Slattc a Downloaded from by guest on 0 November 018

12 Eating Patterns and Colon Cancer 15 important in reducing risk of colon cancer (0). Identification and evaluation of dietary patterns as they are associated with disease is difficult to achieve, given our current methods of dietary intake data collection. Thus, in order to study eating patterns rather than isolated foods and nutrients, we must use statistical methods which formulate patterns. These methods are subjective, although we believe that they may prove to be useful in studying dietary associations with disease. The dietary patterns generated in this study were not established a priori. The emergence of dietary patterns which fitted our a priori assumptions of diet and disease lends further support to the commonly held belief that a Western-style diet contributes to colon cancer risk and that a prudent diet, one enriched with vegetables, refined grains, fruits, fish, and poultry, may help prevent colon cancer. We have also shown that some dietary patterns may be more strongly and consistently associated with colon cancer than individual food items or nutrients. There are some limitations in our data. First, we were unable to measure eating patterns directly, and we relied on statistical manipulation of the data to develop dietary patterns. However, similar patterns were generated using the maximum likelihood method and for men and women in sex-specific factor analyses. Thus, we believe that the dietary patterns we identified are plausible. However, these patterns do not represent all possible patterns, as is indicated by the proportion of variability in the diet that these patterns represent. Likewise, it should be kept in mind that while people may fit into more than one pattern, the association with colon cancer comes from the degree to which they are associated with any particular pattern. When developing the dietary patterns, we grouped the data into predefined groups. While we could have used all of the food items listed on the questionnaire, it was impractical to approach the problem in that manner given the fact that the questionnaire included over 800 possible individual food items. Likewise, our data are based on recall of diet for a period - years prior to the date of interview; it is always possible that errors in recall occurred in both cases and controls, or that the referent period of the study did not capture the relevant etiologic time of exposure. While we used factor analysis to identify dietary patterns that may be associated with colon cancer, factor analysis may also be useful in identifying lifestyle patterns associated with disease. Our brief assessment of lifestyle factors based on variables such as physical activity, body size, cigarette smoking, and use of NSAIDs in conjunction with dietary variables showed clear patterns of association between physical activity, body size, the Western diet, and the prudent diet. However, we did not adjust for physical activity or body size, since it appeared that little would be added in terms of an adjustment variable beyond that achieved by adjusting for the individual indicators of these variables. In summary, we believe that our evaluation of associations between dietary patterns and colon cancer provides insight into eating patterns which increase and decrease risk of the disease. Our data support the hypothesis that a Western-style diet increases risk of colon cancer while a dietary pattern that is richer in vegetables and lower in red meat reduces risk. The associations we observed appear to be more consistent and stronger than those identified for individual food items. ACKNOWLEDGMENTS This study was funded by National Cancer Institute grant RO1 CA8998 to Dr. Martha Slattery. Case identification and verification was supported by the Utah Cancer Registry, the Northern California Cancer Registry, the Sacramento Tumor Registry, and the Minnesota Cancer Surveillance System. The authors acknowledge the contributions and support of S. Edwards, T. Dennis Berry, Dr. Richard Kerber, Dr. Kristin Anderson, and Dr. Debra Duncan in data collection and study supervision. REFERENCES 1. Potter JD, Slattery ML, Bostick RM, et al. Colon cancer: a review of the epidemiology. Epidemiol Rev 199; 15: McCann, Randall E, Marshall JR, et al. Diet diversity and risk of colon cancer in western New York. Nutr Cancer 199;1:1-1.. Fernandez E, D'Avanzo B, Negri E, et al. Diet diversity and the risk of colorectal cancer in northern Italy. Cancer Epidemiol Biomarkers Prev 1996;5:-6.. Schwerin HS, Stanton JL, Smith JL, et al. Food, eating habits, and health: a further examination of the relationship between food eating patterns and nutritional health. Am J Clin Nutr 198;5(suppl): Randall E, Marshall JR, Graham S, et al. High-risk health behaviors associated with various dietary patterns. Nutr Cancer 1991;16: Slattery ML, Bishop DT, French TK, et al. Lifestyle and blood pressure levels in male twins in Utah. Genet Epidemiol 1988; 5: World Health Organization. International classification of diseases for oncology. nd ed. Geneva, Switzerland: World Health Organization, Slattery ML, Edwards SL, Caan BJ, et al. Response rates among control subjects in case-control studies. Ann Epidemiol 1995;5: Slattery ML, Potter J, Caan B, et al. Energy balance and colon cancer beyond physical activity. Cancer Res 1997;57: Downloaded from by guest on 0 November 018

13 16 Slattery et al. 10. Edwards S, Slattery ML, Mori M, et al. Objective system for interviewer performance evaluation for use in epidemiologic studies. Am J Epidemiol 199;10: Slattery ML, Caan BJ, Duncan D, et al. A computerized diet history questionnaire for epidemiologic studies. J Am Diet Assoc 199;9: McDonald A, van Horn L, Slattery ML, et al. The CARDIA dietary history: development, implementation, and evaluation. J Am Diet Assoc 1991;91: Liu K, Slattery ML, Jacobs D Jr, et al. A study of the reliability and comparative validity of the CARDIA dietary history. Ethn Dis 1995;: Dennis B, Ernst N, Hjortland M, et al. The NHLBI nutrition data system. J Am Diet Assoc 1980;77: Micozzi MS, Albanes D, Jones DY, et al. Correlations of body mass indices with weight, stature, and body composition in men and women in NHANES I and II. Am J Clin Nutr 1986;: Jacobs DR Jr, Hahn LP, Haskell WL, et al. Validity and reliability of a short physical activity history: CARDIA and the Minnesota Heart Health Program. J Cardiopulm Rehabil 1989;9: Slattery ML, Jacobs DR Jr. Assessment of ability to recall physical activity of several years ago. Ann Epidemiol 1995; 5: Haenszel W, Kurihara M. Studies of Japanese migrants. I. Mortality from cancer and other diseases among Japanese in the United States. J Natl Cancer Inst 1968;0: Adlercreutz H. Western diet and Western diseases: some hormonal and biochemical mechanisms and associations. Scand J Clin Lab Invest Suppl 1990;01:-. 0. Slattery ML, Potter JD, Coates A, et al. Plant foods and colon cancer: an assessment of specific foods and their related nutrients (United States). Cancer Causes Control 1997;8: Downloaded from by guest on 0 November 018