Lifestyle and Colon Cancer: An Assessment of Factors Associated with Risk

American Journal of Epidemiology Copyright 1999 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 150, No. 8 Printed In U.SJ\. Lifestyle and Colon Cancer: An Assessment of Associated with Risk Martha L Slattery, 1 Sandra L. Edwards, 1 Kenneth M. Boucher, 1 Kristin Anderson, 2 and Bette J. Caan Studies of the etiology of colon cancer indicate that it is strongly associated with diet and lifestyle factors. The authors use data from a population-based study conducted in northern California, Utah, and Minnesota in 1991-1995 to determine lifestyle patterns and their association with colon cancer. Data obtained from 1,99 and 2,410 controls were grouped by using factor analyses to describe various aspects of lifestyle patterns. The first five lifestyle patterns for both men and women loaded heavily on dietary variables and were labeled: "," "moderation," "calcium/low-fat dairy," "meat and mutagens," and "nibblers, smoking, and coffee." Other important lifestyle patterns that emerged were labeled "body size," "medication and supplementation," "alcohol," and "physical activity." Among both men and women, the lifestyle characterized by high levels of physical activity was the most marked lifestyle associated with colon cancer (odds ratios = 0.42, 95% confidence interval: 0.2, and odds ratio = 0.52, 95% confidence interval: 0.9,, for men and women, respectively) followed by medication and supplementation (odds ratio = 1.68, 95% confidence interval: 1.29, 2.18 and odds ratio = 1.6, 1.2, 2.16, respectively). Other lifestyles that were associated with colon cancer were the lifestyle, the lifestyle characterized by large body size, and the one characterized by calcium and low-fat dairy. Different lifestyle patterns appear to have age- and tumor site-specific associations. Am J Epidemiol 1999;150:869-77. anti-inflammatory agents, nonsteroidal; aspirin; colonic neoplasms; diet; hormone replacement therapy; obesity; physical exercise Colon cancer is multifactorial in its etiology and is thought to result from a composite of host, diet, and other lifestyle factors that occur over many years. It is likely that there are many pathways to disease and that factors that increase risk do not work in isolation from other factors. Because of studies conducted among migrant populations, in which incidence rates of the host country are adopted after migration, it has been hypothesized that components of a lifestyle contribute to risk (1, 2). Studies seeking to identify causal factors focused on specific components of a lifestyle, such as diet, physical activity, obesity (1), and, in some instances, interactions, have been examined between exposures (-6). However, an integrated picture of lifestyle on risk of colon cancer is lacking. In this study, we attempt to identify lifestyle Received for publication November 11, 1997, and accepted for publication September 25, 1998. Abbreviations: HRT, hormone replacement therapy; NSAIDS, nonsteroidal anti-inflammatory drugs. 1 University of Utah, Department of Family and Preventive Medicine, Health Research Center, Division of Public Health Sciences, Salt Lake City, UT. 2 Department of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, MN. Division of Research, Kaiser Permanente Medical Care Program, Oakland, CA. Reprint requests to Dr. Martha L. Slattery, University of Utah, Division of Public Health Sciences, 91 Chipeta Way, Suite G, Salt Lake City, UT 84108. factors that may alter risk of colon cancer. We do this by merging many individual exposures associated with risk into lifestyle factors that explain variability in the population. We examine the impact of these lifestyle factors on colon cancer to determine which of them may be most important in the population. MATERIALS AND METHODS Study population Study participants were recruited from 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 (Anoka, Carver, Dakota, Hennepin, Ramsey, Scott, and Washington counties) of Minnesota. A rapid-reporting system was used to identify, with the majority of being interviewed within 4 months of diagnosis. Eligibility criteria for included diagnosis with first primary incident colon cancer (International Classification of Diseases for Oncology, 2nd edition, codes 18.0 and 18.2-18.9) between October 1, 1991, and September 0, 1994; aged 0-79 years at time of diagnosis; Black, White, or Hispanic background; and mentally competent to complete the interview. Cases with cancers of the rec- 869

870 Slattery et al. tosigmoid junction or rectum (defined as the first 15 cm from the anal opening), with known familial adenomatous polyposis, ulcerative colitis, or Crohn's disease were not eligible. Of those invited to participate in the study, approximately 76 percent cooperated. Methods used to ascertain controls have been reported and include random selection from Kaiser membership lists in California, random-digit dialing, and random selection from driver's license lists and Social Security lists in Utah; and random selection from driver's license lists in Minnesota (7). Of all controls asked to participate, approximately 64 percent cooperated. Reasons for nonparticipation have been described elsewhere (8): 29.5 percent of controls identified refused to participate, and 15.5 percent could not be located. Of those who could not be located, 88 percent were from Minnesota, where people obtaining a driver's license were added but those who moved out of the state or who no longer held a valid Minnesota driver's license were not deleted from lists. A total of 1,99 and 2,410 controls with complete data are included in the analyses presented. Data collection Data were collected from study participants by trained and certified interviewers using laptop computers. The study questionnaire was pretested on a randomly selected (using random-digit dialing) group of people over age 50 years living in Utah. The referent period that study participants were asked to recall was the calendar year 2 years prior to the date of selection (the date of diagnosis for or the date of selection for controls). The interview took approximately 2 hours to complete. Quality-control methods used in the study have been described elsewhere (9, 10). Dietary intake data were ascertained using an adaptation of the validated Coronary Artery Risk Development in Young Adult (CARDIA) diet history questionnaire (10-12). Nutrient values for specific foods were calculated using the Minnesota Nutrition Coordinating Center's nutrient database version 19 (1). Foods included in various food groups have been described previously (14-16). Servings within food groups were developed by assigning each food item a standard serving size; the number of standard servings consumed from a food group were summed for each individual. Dietary data included in these analyses were total energy intake, dietary fiber, calcium, dietary cholesterol, folic acid, sucrose, animal protein, vegetable protein, red meat, fish, poultry, eggs, low- and high-fat dairy products, whole and refined grains, vegetables, fruits, and sugar. Other data used in these analyses were age at the time of diagnosis or selection; body mass index (weight/height 2 for men and weight/height IJ for women) self-reported for the referent year; presence or absence of first-degree relatives with colorectal cancer; use of aspirin and/or nonsteroidal anti-inflammatory drugs on a regular basis (hereafter grouped together and referred to as NSAIDs); usual number of cigarettes smoked; amount of alcohol consumed during the referent year; use of vitamin/mineral supplements; long-term vigorous physical activity; education level; number of meals eaten per day; usual amount of coffee consumed per day; and a mutagen index that takes into account the amount of meat eaten, how it is prepared, and whether it is precooked in a microwave. Total number of times pregnant and use of hormone replacement therapy (HRT) were obtained as part of the reproductive history section of the questionnaire and were used as lifestyle variables for women. Statistical methods Lifestyle factors. Diet and lifestyle variables, as described above, were included in the factor analysis. These variables were chosen because of their previous associations with colon cancer, either in these data or in other published data. Lifestyle factors were developed using factor analysis as described elsewhere (14), using the Statistical Analysis Software principal components program (17). with eigenvalues of greater than 1.0 were further evaluated. Within a factor, a negative loading indicates that lifestyle variables are inversely associated with the factor, while a positive loading indicates a direct association with the factor. The larger the loading of a variable to the factor, the greater the contribution of that variable to a specific factor. Variables that loaded at 0.20 or greater were considered to be making a larger contribution to the factor and are italic in tables; the value for meaningful factor loading, however, is arbitrary. After a varimax rotation, factor scores were saved from the principal component analyses for each individual. All data presented are from the varimax rotation. These scores were used to estimate associations with colon cancer and determine which composite of factors may be most important in colon cancer etiology. Because of the exploratory nature of factor analyses, 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 many instances, the maximum likelihood method did not converge, although the principal component method did. In this paper, we present data generated from the principal components method. Labeling of lifestyle patterns was done based upon our interpretation of the data and does not represent a priori lifestyle patterns. However,

Lifestyle Patterns and Colon Cancer 871 the labels given were those that we believed were representative of the existing literature on colon cancer. It should be kept in mind that these labels are arbitrary, and others may be equally suited to the data. Analytic methods. Age-specific analyses used the median age of the controls, 67 years, as the cutpoint. Factor scores were categorized into quintiles based upon the distribution of the control population for men and for women separately. To determine the associations between dietary factors or eating patterns and colon cancer, odds ratios and 95 percent confidence intervals were calculated from unconditional logistic regression models. Age-adjusted estimates of association were generated as well as backward stepwise logistic regression models that included all factors. On the basis of these results, we developed models that best represented the data and constructed multiple logistic regression models that included all lifestyle factors that appeared to contribute to risk. Tumor site within the colon was classified as proximal (cecum through transverse colon) or distal (splenic flexure, descending, and sigmoid colon). RESULTS The first five factors were almost identical among men (table 1) and women (table 2) and loaded heavily on dietary components. We have labeled the first factor "" since it loaded heavily on most dietary items typical of a diet, including total energy intake, dietary cholesterol, animal protein, red meat, refined grains, high-fat dairy, and sugar; physical activity showed one of its lowest loadings to this factor. The second factor was labeled "moderation." Although it loaded heavily on dietary fiber, folic acid, vegetable protein, whole grains, and fruits and vegetables, it had high loadings on sucrose and total energy and moderate-to-low loadings on fish, poultry, animal protein, cholesterol, sugar, and physical activity (loadings of between 0.10 to 0.20). The next factor, characterized by low-fat dairy foods and calcium, although whole grains contributed to the factor in both men and women and sucrose contributed to the factor among women, was labeled "calcium/low-fat dairy." Animal protein, red meat, fish, poultry, and mutagen index loaded to the next factor we labeled "meat and mutagens." The fifth lifestyle factor, labeled "nibblers, smoking, and coffee" was characterized by smoking cigarettes, eating frequent meals, and drinking coffee. The remaining factors varied in order of contribution of variability to lifestyle among men and women. Among men, factor 6 loaded heavily on alcohol and was labeled "alcohol" (table 1). Factor 7 among men was typified by higher education and higher levels of vigorous activity; we called this factor "physical activ- ity." Factor 8 represented men who did not take multivitamins or use aspirin or NSAFDS and who were also more likely to have a family history of colorectal cancer and was labeled "medication and supplementation." Factor 9 among men was represented by larger body size. Among women, a larger body size and increased parity was associated with factor 6. Factor 7 (not shown in table) did not load heavily on any item but had low loadings on several items. Factor 8 among women included those who did not take multivitamins, aspirin and/or NSAIDS, or HRT and was called medication and supplementation. loaded heavily to factor 9 among women; family history of colorectal cancer was inversely associated with this factor (labeled physical activity). Age-adjusted estimates of association between the nine predominate lifestyle patterns in men and risk of colon cancer (table ) showed that, no drugs, and body size were all directly associated with colon cancer. and calcium/low-fat dairy were inversely associated with colon cancer. Among all women, age-adjusted associations showed that and medication and supplementation were directly associated with colon cancer risk, while calcium/low-fat dairy and physical activity were inversely associated with risk (table 4). We further evaluated lifestyle factors, taking into account other lifestyle factors by age at time of diagnosis and by tumor site within the colon for men and women. It appears that estimates of association from the fully adjusted logistic models (tables 5 and 6) are almost identical to those from the age-adjusted models (tables and 4). However, the importance of lifestyle factors varies by sex, age at diagnosis, and tumor site within the colon. More lifestyle patterns are importantly associated with colon cancer among men than among women, among those diagnosed at a younger age than among those diagnosed when older, and among those with distal tumors rather than proximal tumors. For both men and women,, calcium/low-fat dairy, physical activity, and medication and supplementation aspects of lifestyles were associated with risk (tables 5 and 6). Among men, body size also contributed to risk. Patterns observed for those diagnosed before age 67 years were similar to those observed for the total population, but was slightly more strongly associated with risk in this younger age group. Among individuals who were older at the time of diagnosis, physical activity and medication and supplementation were the only factors associated with risk among women; among men, and body size continued to contribute to risk. and medication and supplementation were associated with

872 Slattery et al. TABLE 1. Assessment of factor-loading matrix of lifestyle factors among male controls (n = 1,290)', United States, 1991-1995 Education BMIf Family history Cigarettes smoked Meal frequency Dietary intake Energy Dietary fiber Calcium Cholesterol Folic acid Sucrose Animal protein Vegetable protein Red meat Fish Poultry Eggs Low-fat dairy High-fat dairy Whole grains Refined grains Vegetables Fruit Sugar Coffee Alcohol Mutagen index Multivitamins Aspirin/NSAIDS Proportion of variability Factor 1 (western) 0.56 0.87 0.2 0.58 0.7 0.56 0.68 0.45 4.84 Factor 2 (moderation) 0.52 0.91 0. 0.51 0.2 0.29 0.4 0. 4.29 Factor (calcium/ low-fat dairy) 0.67 0.25 0.7 0.92 0.50 1.89 Factor 4 (meat and mutagera) 0.20 0.28 0.45 0.4 0.46 0.40 0.68 0.1 0.54 1.79 (nibbling, smoking, and coffee drinking 0.57-0.2 1.57 (alcohol) 0.27-0.29 0.6-0.24-0.2-0.1 0.20 1.7 Factor 7 (physical activity) 0.67\ 0.7-0.41-0.20-0.22 1.5 Factor 8 and supplementation) 0.6-0.24 0.75 1.16 Factor 9 (body size) Values of <0.20 have been excluded from the table for simplicity. t Values with loadings of 0.40 or greater are italicized. Positive loading with multivitamins, aspirin, and/or nonsteroidal anti-inflammatory drugs (NSAIDS) indicates no usage. Positive loading with family history of coiorectal cancer indicates a positive family history. % BMI, body mass index. -0.67 0.78-0.2 0.8 1.15 risk among those with proximal tumors in both men and women. It is of interest to note that the upper bound of the confidence interval associated with physical activity among men was 0.50. Among women with proximal tumors, moderation is inversely associated with risk, while among men, is directly associated with risk. The associations between and body size and distal colon tumors is stronger than that observed for proximal tumors among men. DISCUSSION This study was designed to examine associations between dietary intake, including nutrients as well as foods, physical activity, and other lifestyle factors and risk of colon cancer. Several lifestyle patterns emerged that explained variability in the data. Of these factors, the first five were related to diet. Although this study focused on diet and therefore included many dietary exposures in developing lifestyle patterns, this observation suggests that differences in diet explain much of the variability in the population. On the other hand, exposures such as physical activity explained little variability in the population, possibly because few people regularly engage in vigorous activities. Family history of coiorectal cancer did not generally load to any factor, suggesting that it explains very little of the variability in the spread of the data, even though it has been shown repeatedly to be an important risk factor for colon cancer (1). Thus, our examination of lifestyle patterns includes factors that explain variability in the population, but does not include all risk factors for colon cancer. In instances in which interpretation of the data is not improved

Lifestyle Patterns and Colon Cancer 87 TABLE 2. Assessment of factor-loading matrix for lifestyle factors among female controls (n = 1,120)*, United States, 1991-1995 Education BMIf Family history Cigarettes smoked Meal frequency Dietary intake Energy Dietary fiber Calcium Cholesterol Folic acid Sucrose Animal protein Vegetable protein Red meat Fish Poultry Eggs Low-fat dairy High-fat dairy Whole grains Refined grains Vegetables Fruit Sugar Coffee Alcohol Mutagen index Multivitamins Aspirln/NSAIDS HRT Parity Proportion of variability giggg.l'g'gg 0.29 0.50 0.8 0.0 0.66 0.67 0.48 0.57 0.52 0.70 4.86 0.50 0.91 0.28 0.40 0.78 0.20 0.5 0.9 0.64 4.0 0.26 0.75 0.24 0.47 0.91 0.40 2.17 0.2 0.48 0.6 0.62 0.66 0.1 0.44 1.84 0.56 0.65-0.22 1.52-0.1 0.68$ 0.26 0.4-0. 0.5 0.66 0.7 0.4 1.4 1.22 0.27-0.57 * Values of <0.20 have been excluded from the table for simplicity. t BMI, body mass index. X Values with loadings of 0.40 or greater are italicized. Positive loading with multivitamins, aspirin, and/or nonsteroidal anti-inflammatory drugs (NSAIDS), and hormone replacement therapy (HRT) indicates no usage. Positive loading with family history of colorectal cancer indicates a positive family history. 0.25 1.18 beyond that by an individual variable, lifestyle factors are not generated. Each lifestyle pattern contains elements of what we have traditionally considered good and bad factors. This is most evident when we examine lifestyle patterns loading heavily on dietary variables and may account for the somewhat low overall association between these patterns and the risk of colon cancer. Thus, the risk associated with each lifestyle factor is influenced by how heavily it is loaded with good and bad factors. The lifestyle patterns that contribute most consistently to risk are those that load heavily on a few variables, such as those that we labeled as physical activity and medication and supplementation, yet describe more variability in the population than each variable independently. Studies have repeatedly shown that risk factors associated with colon cancer may vary by age at diagnosis, tumor site within the colon, and sex of the case (1, 18-20). In our evaluation of lifestyle patterns, it appears that certain lifestyle patterns, such as those involving high levels of physical activity and those that are associated with using drugs such as NSAIDs, are important for all subgroups. Others lifestyle patterns have stronger or weaker associations depending on the subgroup being assessed. For instance, factors associated with a lifestyle have been thought

TABLE. Age-adjusted colon cancer risk associated with lifestyle factors for all men, United States, 1991-1995 CO Moderation Calclum/low-fat dairy Meat and mutagens Nibbling, smoking, and coffee drinking Alcohol 26/185 261/220 269/270 266/26 265/224 265/224 264/14 262/177 264/189, odds ratio; Cl, confidence interval. * 259/184 26/207 255/207 254/216 261/219 261/219 258/249 258/188 255/206 1.01 0.9 0 81 0.96 1.09 0.99 0.81 1.08 1.1 * 0.78, 1.2, 1.20 0.6, 1.04 0.75, 1.2 0.85, 1.40, 1.28 0.64, 1.0 0.82, 1.41 0.87, 1.46 258/220 249/28 254/29 258/226 25/188 25/188 25/22 257/24 260/207 1.22 1.1 0.94 0.99 1.05 0.88 1.5 Quintile 0.94, 1.59 0.88, 1.46 0.7, 120, 1.27 0.82, 1.5 0.68, 0.59, 0.97 1.04, 1.75 0.85, 1.44 No of 252/246 260/210 255/196 259/196 255/222 255/222 260/174 260/220 257/210 1.40 0.96 0.86 1.0 1.25 1.08, 1.62, 1.2 0.60, 0.98 0.66, 0.78, 1.0, 1. 0.4, 0.97, 1.6 0.88, 1.48 258/264 257/224 257/187 25/225 256/246 256/246 255/10 25/280 254/287 5 (high) 1.49 1.0 0.7 1.01 0.97 0.41 1.65 1.58 1.15, 1.9, 1. 0.56, 0.9 0.78, 1.0 0.75, 1.26 0.89, 1.47 0.1,0.54 1.28,2.1 1.22,2.0 P for trend 0.70 0.02 0.75 0.58 0.0 co.01 I 0) TABLE 4. Age-adjusted colon cancer risk associated with lifestyle factors for all women, United States, 1991-1995 Quintile 1 (low) 2 4 5 (high) rn < o p Moderation Calclum/low-fat dairy Meat and mutagens Nibbling, smoking, and coffee drinking 28/177 25/20 28/195 29/182 26/167 29/198 26/150 28/20 * 22/142 224/189 218/212 219/164 22/180 220/170 224/15 216/216 0.86 0.98 1.19 0.98 0.9 1.07 1.0 * 0.64, 0.75, 1.28 0.91, 1.55, 1.0 0.88, 1.50, 1.2, 1.4, 1.4 217/179 221/181 221/196 217/216 22/190 217/156 218/154 225/175 0.95 1.08 1.20 0.87 0.84, 1.47 0.7, 1.25 0.8, 1.42,1.71 0.91, 1.58 0.66, 1.15 0.8, 1.48 0.61, 1.04 221/181 220/150 222/146 224/171 219/172 22/155 221/211 219/154 1.10 0.99 1.10 0.84 1.50 0.70 0.8, 1.45 0.60, 1.05 0.60, 1.06 0.75. 0.8, 1.47 0.64, 1., 1.98 0.5, 0.9 221/215 220/171 221/145 221/161 219/185 221/215 221/226 222/119 0.90 0.95 1.19 1.18 1.61 0.5, 1.72 0.68, 1.18 0.60, 1.05, 1.26 0.90, 1.57 0.90, 1.5 1.22,2.12 0.9, P for trend 0.02 0.16 0.01 0.70 0.42 0.44 9 *, odds ratio; Cl, confidence Interval. (O CO CO

Lifestyle Patterns and Colon Cancer 875 TABLE 5. Best-m models* associations between lifestyle factors and colon cancer risk In men, United States, 1991-1995 Quintfle 1 (row) (t) 2 95%Crf 4 >(hlgh) All subjects Calcium/low-fat dairy 1.02 0.84 1.10 1.16 0.78, 0.65, 0.6, 0.84, 0.89, 1.4 1.08 1.02 1.45 1.52 0.95 1.2 1.12, 1.71, 1.22 0.58, 0.95 1.01, 1.72 0.86, 1.47 1.9 1.0 1.15 1.07,1.81 0.61, 1.02 0.4, 0.99, 1.70 0.88, 1.50 1.44 0.42 1.68 1.54, 1.88 0.56, 0.94 0.2, 1.29,2.18 1.19, 2 Subjects age <67 years Calcium/low-tat dairy 1.9 0.97 0.90 1.21 0.91,2.14 0.67, 1.40 0.54, 1.17 0.60,1.5 0.8,1.78 1.57 1.01 1.56 0.98 1.04,2.8, 1.4 0.49, 1.02 1.05,2.0 0.67, 1.45 1.9 0.59 1.21 1.27 0.9, 2.08 0.5, 1.10 0.40, 0.86 0.82,1.79 0.87, 1.85 1.64 0.68 0.40 1.68 1.49 1.12,2.42 0.47, 0.99 0.27, 0.58 1.16, 2.44 1.04, 2.15 Subjects age 67 years Proximal tumors Distal tumors Calcium/low-fat dairy 0.82 0.8 1.28 0.95 1.24 1.09 0.82 0.98 1.29 0.57,1.17 0.61, 0.88, 1.87,1.67 0.68,1.5 0.57,1.04 0.87, 1.77, 1.54 0.52, 0.99 0.61,1.10,1.8 0.91, 1.82 1.15 1.12 1.2 1.55 0.82 1.16 1.9 0.81,1.65 0.57,1.12,1.6 0.90,1.9 0.94, 1.8 0.54, 0.98 1.10,2.18 0.9, 1.84 0.60,, 1.01 0.8, 1.6 0.99, 1.97 1.49 0.50 1.40 1.0 1.4 0.54 1.8 1.6 1.18 1.46 1.04, 2.1 0.5, 0.96, 2.04 0.70,1.5 1.0,1.98 0.9, 0.75 0.98,1.95 0.97, 1.90 0.5, 0.99 0.40,, 0.85, 1.66 1.0,2.06 1. 0.47 1.65 1.60 1. 0.5 1.7 1.69 0.48 1.58 1.82 0.91, 1.94 0.0, 1.1, 2.40 1.09,2.4 0.91, 1.94 0.24, 0.50 1.24,2.4 1.22,2.4 0.51,0.95 0.4, 0.67,2.18 1.0,2.54 Ail factors in the same model. Best-fit models were determined from those factors that significantly improved the fit of the logistic regression model. t, odds ratio; Cl, confidence interval. to be more important for distal tumors (21). In our data, and calcium/low-fat dairy lifestyles were more associated with distal tumors than with proximal tumors. On the other hand, proximal tumors have been thought to be more related to endogenous factors than to exogenous factors (21); the lifestyle patterns most associated with proximal tumors were physical activity, medication and supplementation (including HRT among women), and moderation among women and among men. In general, we observed that more lifestyle patterns appeared to be important for men than for women in relation to colon cancer. This finding is supported by the literature that suggests that men who migrate adopt the rates of the host country sooner than do women (2). loading on dietary components have a greater influence on risk of colon cancer among people diagnosed at a younger age than among those diagnosed at an older age. As in other studies, few factors appear to be associated with risk among older women (4, 5,19, 22). There are limitations to this study. Direct measurement of lifestyle patterns is impossible, and therefore, our assessment of lifestyle patterns relies on statistical manipulation of data. However, it is built on our knowledge of colon cancer and of associations previously identified within this and other databases. Patterns that emerged were, for the most part, interpretable and consistent for men and women. Labels assigned to these lifestyle patterns were arbitrary and

876 Slattery et al. TABLE 6. Best-fit models* associations between lifestyle factors and colon cancer risk in women, United States, 1991-1995 Qulntile 1 (low) (t) 2 t 4 5 (high) All subjects CalciurrVlow-fat dairy 0.86 1.16 0.97 0.65, 1.16 0.88, 1.52, 1.27 0.86, 1.51 0.84, 1.46 0.54, 0.94 1.1 0.84,1.52 1.15 0.86,1.55 1.12 1.60 0.84, 1.48 0.60, 1.07 0.54, 0.94 1.7 0.52 1.04, 1.81 0.60, 1.07 0.9, 1.20,2.12 1.6 1.2,2.16 Subjects age <67 years Calcium/low-fat dairy 0.85 1.28 0.5, 1.5 0.86, 1.91 0.50, 1.17 1.47 0.57 0.85, 2.02 0.98, 2.21 0.7, 0.86 0.89 0.66 0.85, 2.01 0.59, 1.5 0.44, 1.57 0.78 0.6 1.04, 2.6 0.52, 1.17 0.24, 0.97 0.64,1.48 1.15,1.76 1.55 1.0,2.4 1.40 0.9,2.11 Subjects age 67 years Proximal tumors Moderation Distal tumors Calcium/low-fat dairy 1.17 0.8,1.65 0.9 0.64,1.5 0.6 0.42,0.95 0.68 0.44,1.06 1.4 0.88,2.0 0.75,1.74 1.70,2.52 1.92 1.0,2.85 1.10 0.91 1.5 1.04, 1.54 0.66, 1.26 1.09 0.78, 1.5 0.49, 0.96 1. 0.91, 1.95 1.8 0.94,2.01 0.5, 0.96, 1.90, 1.46 1.15 0.75 0.81, 1.65, 1.58 0.5, 1.07 0.7 0.59 1.82 1.05 0.50, 1.05 0.41, 0.8 0.7, 1.51 0.51, 1.08 0.56, 1.15 0.44 1.29 0.78 0.56 0.90 0.62, 0.94 0.65,1.7 1. 0.9,1.90 1.46 0.52, 1.09 0.1,0.65 1.26,2.62 1.82 1.26,2.61 0.91, 1.8 0.54, 1.1 0.8, 0.81 1.0,2.06 Ail factors in the same model. Best-fit models were determined from those factors that significantly improved the fit of the logistic regression model. t, odds ratio; Cl, confidence interval. represent our interpretation of the data; others may label these lifestyle patterns differently. These data represent aspects of lifestyle patterns identified within the population being studied. The importance of these patterns may vary between populations. The patterns were developed based on their ability to explain the variability in the data and, as previously mentioned, do not account for all individual items that may be associated with colon cancer risk. Other studies have used similar statistical methods to evaluate eating patterns and health (14, 2-25). In contrast to our current study of lifestyles, our evaluation of eating patterns from this population included a more detailed listing of foods and did not take into consideration nutrients (other than total energy) and environmental exposures in their development (14). As in these analyses, a dietary pattern typified as a diet was directly associated with risk, as was the lifestyle pattern we defined as. The eating pattern we formerly identified as the "substitutor" may model the lifestyle of "moderation," in which changes were made that were considered healthy, but did not carry the same weight as other eating and lifestyle patterns reflective of more stringent patterns, such as the "prudent" eating pattern. Development of lifestyle patterns using our previously constructed eating patterns (14), rather than nutrients and foods, yielded slightly different results. The primary difference was a stronger increase in risk associated with factor 1 (), while loaded heavily with the diet. In men, the risk estimate was 2.05 (95 percent confidence interval: 1.55, 2.71), while in women, it was 1.50 (95 percent confidence interval: 1.1, 1.99), comparing the upper with the lower quartile of intake. The prudent diet, which we previously identified as being protective, did not dominate any factor and had loadings of approximately 0.25-0.5 for most factors. Associations with physical activity and other factors were similar to

Lifestyle Patterns and Colon Cancer 877 those presented here. In general, however, our ability to describe lifestyle patterns was easier when we utilized foods and nutrients as the primary input data in constructing the factors than when we used previously generated scores. While there are benefits to the assessment of lifestyles in conjunction with public health practice, it is important to evaluate these factors in terms of identification of factors that may contribute to colon cancer risk beyond the individual components within the factors. Comparison of individual contributors with factor loadings of more than 0.50 to lifestyle factors shows that in most instances the lifestyle pattern reinforces previously identified factors for all subjects; however, there is variability in subgroups of the population for individual exposures, as shown for lifestyle patterns. Some lifestyle factors, such as physical activity, appear to be slightly stronger, while individual contributors to the lifestyle of moderation appear to be slightly weaker than some of the contributors to the lifestyle, such as dietary fiber and vegetable intake. This is in contrast to previously reported eating patterns, developed based solely on food intake, that showed stronger and more consistent associations for overall patterns than for individual components of patterns (14). In summary, these data support the hypothesis that lifestyle patterns are important determinants of colon cancer risk. These data stress the importance of incorporating physical activity into one's lifestyle, not becoming overweight, and eating a diet that is low in animal products and high in plant foods and low-fat dairy products. A lifestyle that incorporates using drugs such as aspirin, NSATDS, and HRT also appears to be beneficial. While certain lifestyle patterns appear to be related to certain subgroups within the population, the importance of developing a lifestyle that incorporates vigorous physical activity appears to be universal to all groups evaluated and may be the most important component of a healthy lifestyle that decreases risk of colon cancer. Assessment of lifestyle patterns has direct implications for public health practice, although they contribute less to identification of new, previously unrecognized risk factors for colon cancer. ACKNOWLEDGMENTS This study was funded by grant RO1 CA48998 to Dr. Martha L. 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 Dr. John Potter and Dr. Debra Duncan to the 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; 15:499-545. 2. 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;40:4-68.. Giovannucci E, Rimm EB, Ascherio A, et al. Alcohol, lowmethionine-low-folate diets, and risk of colon cancer in men. J Natl Cancer Inst 1995;87:265-7. 4. Slattery ML, Potter JD, Duncan D, et al. Dietary fats and colon cancer assessment of risk associated with specific fatty acids. Int J Cancer 1997;7:670-7. 5. Slattery ML, Caan BJ, Potter JD, et al. Dietary energy sources and colon cancer risk. Am J Epidemiol 1997; 145:199-210. 6. Whittemore AS, We-Williams AH, Lee M, et al. Diet, physical activity, and colorectal cancer among Chinese in North America and China. J Natl Cancer Inst 199O;82:915-26. 7. Slattery ML, Edwards SL, Caan BJ, et al. Response rates among control subjects in case-control studies. Ann Epidemiol 1995;5:245-9. 8. Slattery ML, Potter JD, Caan BJ, et al. Energy balance and colon cancer beyond physical activity. Cancer Res 1997;57:75-80. 9. Edwards S, Slattery ML, Mori M, et al. Objective system for interviewer performance evaluation for use in epidemiologic studies. Am J Epidemiol 1994; 140:1020-8. 10. Slattery ML, Caan BJ, Duncan D, et al. A computerized diet history questionnaire for epidemiologic studies. J Am Diet Assoc 1994;94:761-6. 11. McDonald A, Van Horn L, Slattery ML, et al. The CARDIA dietary history: development and implementation. J Am Diet Assoc 1991;91:1104-12. 12. Liu K, Slattery ML, Jacobs D Jr, et al. A study of the reliability and comparative validity of the CARDIA dietary history. EthnDis 1994;4:15-27. 1. Dennis B, Ernst N, Hjortland M, et al. The NHLBI nutrition data system. J Am Diet Assoc 1980;77:641-7. 14. Slattery ML, Boucher KM, Caan BJ, et al. Eating patterns and risk of colon cancer. Am J Epidemiol 1998;148:4 16. 15. Kampman E, Slattery ML, Bigler J, et al. Meat consumption, genetic susceptibility, and colon cancer risk: A U.S. multicenter case-control study. Cancer Epidemiol Biomarkers Prev 1999 8;15-24. 16. Slattery ML, Potter JD, Coates A, et al. Plant foods and colon cancer: an assessment of specific foods and their related nutrients. Cancer Causes Control 1997;8:575-9O. 17. SAS Institute, Inc. SAS user's guide: basics statistics. Cary NC: SAS Institute, Inc., 1985. 18. Slattery ML, Potter JD, Sorenson AW. Age and risk factors for colon cancer are there implications for understanding differences in case-control and cohort studies? Cancer Causes Control 1994;5:557-6. 19. West DW, Slattery ML, Robison LM, et al. Dietary intake and colon cancer: sex and anatomical site-specific associations. Am J Epidemiol 1989; 10:88-94. 20. Young TB, Wolf DA. Case-control study of proximal and distal colon cancer and diet in Wisconsin. Int J Cancer 1988;42:167-75. 21. Bulfill JA. Colorectal cancer, evidence for distinct genetic categories based on proximal and distal tumor location. Ann Intern Med 1990; 114:779-88. 22. Bostick RM, Potter JD, Kushi LH, et al. Sugar, meat, and fat intake, and non-dietary risk factors for colon cancer incidence in Iowa women. Cancer Causes Control 1994;5:8-52. 2. 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 1982;5:119-25. 24. Randall E, Marshall JR, Graham S, et al. High-risk health behaviors associated with various dietary patterns. Nutr Cancer 1991;16:15-51. 25. Slattery ML, Bishop DT, French TK, et al. Lifestyle and blood pressure levels in male twins in Utah. Genet Epidemiol 1988;5:277-87.