This article was downloaded by: [130.132.123.28] On: 04 February 2015, At: 10:54 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Nutrition and Cancer Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/hnuc20
Dietary patterns and colon cancer in western New York a
b
b
Elizabeth Randall , James R. Marshall , John Brasure & Saxon Graham
b
a
Nutrition Program , State University of New York , 301 Parker Hall, Buffalo, NY, 14214 b
Department of Social and Preventive Medicine , State University of New York , Buffalo, NY, 14214 Published online: 04 Aug 2009.
To cite this article: Elizabeth Randall , James R. Marshall , John Brasure & Saxon Graham (1992) Dietary patterns and colon cancer in western New York, Nutrition and Cancer, 18:3, 265-276, DOI: 10.1080/01635589209514227 To link to this article: http://dx.doi.org/10.1080/01635589209514227
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
Dietary Patterns and Colon Cancer in Western New York Elizabeth Randall, James R. Marshall, John Brasure, and Saxon Graham
Downloaded by [] at 10:54 04 February 2015
Abstract Seven dietary patterns were identified among control subjects in the Western New York Diet Study (1975-1986) by application of principal components analysis to data from a 95-item food frequency interview. The results of case-control analyses of colon cancer risk for these patterns are presented. Cases were matched with neighborhood controls on the bases of age and sex; 205 colon case-control male and 223 female pairs were obtained. The dietary patterns and intakes of energy, total fat, and dietary fiber were examined with logistic regression for their individual contributions to risk. In males, three of these dietary patterns were associated positively with fat and energy consumption; they elevated risk for colon cancer and accountedfor more risk than did the specific nutrients. Control for energy and fat intakes allowed the protective influences of additional dietary patterns to be expressed. No patterns elevated risk in women; two patterns were protective for colon cancer. Controlling for energy and fat intake enhanced the protection afforded by one of these patterns but had no influence on that of the other. Measures of foods rather than single nutrients may be more inclusive of dietary exposures to risk as well as being related more directly to underlying health behaviors. Therefore they may be better able to account for risk in diseases with multiple causation. (Nutr Cancer 18, 265-276, 1992)
Introduction
People eat meals consisting of a variety of foods, and as we have discovered in our previous inquiries, habitual food use can be expressed as dietary patterns (1-3). It is obvious that individuals do not ingest single dietary constituents in isolation but, rather, typically ingest several at the same time. Inasmuch as combinations of nutrients, as well as additional nonnutrient dietary constituents, may have different metabolic consequences depending on the combinations in which they are consumed, it is important to determine whether the patterns in food use are associated either positively or negatively with cancer. Previously we demonstrated that locating individuals within quartiles of nutrient intakes simultaneously arrays them on other dietary characteristics as well (1). We identified gender-specific dietary patterns (2,3) among the 1,475 male and 780 female normal healthy subjects in the complete control series of the Western New York Diet Study. These dietary patterns can be characterized by their degree of correlation with the various single nutrients linked to cancer risk (2) as well as by their associations with exposures to nonnutrient dietary factors and to such discretionary health behaviors as alcohol consumption and cigarette E. Randall is affiliated with the Nutrition Program and J. R. Marshall, J. Brasure, and S. Graham with the Department of Social and Preventive Medicine, State University of New York at Buffalo, Buffalo, NY 14214.
Copyright © 1992, Lawrence Erlbaum Associates, Inc.
smoking (4). Multiple risk factors coalesce within several of the dietary patterns, notably those associated with high fat intake, in ways that suggest that putative dietary risks may exceed those of single nutrients (3,4). In earlier analyses using this data set (5), energy, total fat, and dietary fiber were examined separately as risk factors for colon cancer. Among men, both energy and total fat intakes were found to elevate risk. Energy and fat were so collinear that their individual contributions to risk were indistinguishable. Total dietary fiber intake was unrelated to colon cancer risk. For women, total fat elevated risk, but less strongly than for men. The purpose of this study was to examine the role of the dietary patterns in the risk of colon cancer to determine whether the patterns identified were correlated with risk of this high-incidence cancer. We found that certain dietary patterns were more strongly associated with risk of colon cancer than any single nutrient, including fat. Methods
Downloaded by [] at 10:54 04 February 2015
Sample Characteristics
The Western New York Diet Study is a case-control study of diet and cancer of the lung, esophagus, larynx, mouth, stomach, bladder, rectum, and colon that was conducted in Erie, Niagara, and Monroe Counties of Western New York in 1975-1986. The procedures followed have been reported in detail elsewhere (5). Cases were identified by nurse interviewers using hospital records. Those cases whose participation was approved by their physicians and who agreed to be interviewed were recruited and matched to control subjects on the bases of age, gender, and neighborhood of residence. Each subject completed an extensive 2.5-hour interview, including a 128-item semiquantitative food frequency instrument supplemented by questions concerning food preparation and storage practices. Validity and reliability data for the interview are available elsewhere (5). The colon cancer case-control series was completed during 1975-1984 and consisted of 410 males and 446 females of whom equal numbers were case and control subjects. Dietary Data Subjects were queried regarding their frequency of use of 128 food items during the 12-month period preceding the interview for control subjects or the 12 months before cancer diagnosis for the cases. Nutrient intakes were calculated on the basis of the complete food list. Dietary patterns were identified in the full series of control subjects by use of principal components analysis with orthogonal rotation of the factors. This technique identifies factors in which cohesiveness of the constituent foods, expressed as factor weights, is maximized. To enhance their definition, factors were extracted from a subset of the 95 food items from the original instrument that had a mean frequency of consumption of >0.5 times a month, so that foods consumed only infrequently could not be influential in identifying patterns. Foods consumed rarely could be strongly cohesive within individual factors and, therefore, could be highly influential in pattern identification without making any significant contribution to the dietary status of the individual consuming them. In addition, only actual foods (rather than sauces, condiments, or spreads) were included in pattern definition. Full descriptions of these procedures are provided elsewhere (2,3). Although some items were excluded from the original list of 128 food items, associations between their use and pattern scores have been investigated separately (3,4). The seven patterns identified for males and for females are summarized in Table 1. They are listed in order of their contributions to variance in food use, and foods within patterns are listed in the order of the strength of their associations with the specific patterns. Names were assigned to patterns merely to describe the dominant foods within them. High-fat patterns
266
Nutrition and Cancer 1992
Table 1. Dietary Patterns Identified, Western New York Diet Study, 1975-1986 Dietary Pattern Salad Fruit Healthful Traditional Snacks High fat
Downloaded by [] at 10:54 04 February 2015
Whole grain
Salad Healthful
Low cost Fruit High fat Light Whole grain
Foods" Use Males (n = 1,475) Lettuce, celery, green pepper, cucumber, tomatoes, radishes, onions, carrots Peaches, plums, pears, apples, oranges or tangerines, cherries, grapes, melons, bananas, berries Broccoli, spinach, cauliflower, rice, poultry, mushrooms, green beans, summer squash, asparagus Wax beans, green beans, potatoes, peas, beets, cabbage, roast beef, cakes, pies, corn Cookies, candy, crackers, donuts or pastries, hamburger, ice cream, baked beans Eggs, bacon, sausage, steak, beer, salami or pepperoni, distilled alcohol Whole wheat bread, cooked cereal, oatmeal
Females (n = 780) Lettuce, onion, cucumber, tomato, celery, green pepper, radishes, carrot, apples, broccoli Green beans, cauliflower, spinach, beets, summer squash, pineapple, liver, wax beans, broccoli, winter squash, cottage cheese, mushrooms, peas, cabbage Macaroni, spaghetti, bologna, hot dogs, french fries, noodles, hamburger, processed cheese, salami or pepperoni, canned fish, coffee Pears, peaches, plums, oranges or tangerines, apples, bananas, grapes, grapefruit Pies, candy, cake, potatoes, roast beef, donut or pastries, bacon, rolls, potato chips Mushrooms, lemons or limes, hard cheese, fin fish, yogurt, melon, broccoli Cooked cereal, cookies, dried fruit, unsweetened cereal, crackers, oat products, whole wheat bread, nuts, yogurt
Nonuse* None identified Beer Potatoes, cookies, white bread None identified None identified Unsweetened cereal, winter squash, summer squash, cookies White bread, spaghetti, macaroni, coffee, french fries, hamburger, noodles, cola None identified None identified
Asparagus Distilled alcohol, beer, wine Poultry, yogurt Corn, white bread, peas, potatoes Coffee, white bread
a: Listed in order of factor weight. b: Use of these foods is negatively related to the pattern.
were so named because of the fat density of the constituent foods. The traditional pattern includes meats, several common vegetables, and baked desserts; the low-cost pattern includes foods that are relatively inexpensive. The healthful patterns appear to be consistent with the current trend toward the consumption of lean and fiber-rich foods. Data Analysis Preliminary analyses in which pattern scores were calculated using all 95 food items were highly similar in their abilities to estimate colon cancer risk to patterns composed only of their strongest elements. Therefore, we deviated from the standard procedures of factor analysis by including in the patterns only those foods that load most heavily on each factor.
Vol. 18, No. 3
267
Downloaded by [] at 10:54 04 February 2015
To do otherwise would have required including all 95 food items in each pattern and would have resulted in data unwieldy -both conceptually and economically. As a consequence, however, some dietary patterns are intercorrelated. This effect is expressed most strongly in men, with positive correlations among the salad, fruit, healthful, and traditional patterns (0.18 < r < 0.38) and negative correlations between the whole-grain and the snacks patterns (r = -0.22) as well as with the high-fat pattern (r = -0.18). For women, positive intercorrelations were strongest among the salad, healthful, fruit, and light patterns (0.18 < r < 0.42), whereas the only strong negative association was between the high-fat and light patterns (r = -0.24). Such relatedness of the patterns is consistent with the propensity of human beings to change food behaviors in response to changes in living situations; individuals would be likely to exhibit more than one dietary pattern. Pattern scores were calculated in the colon series, for both case and control subjects, by multiplying the standardized reported frequency of intake of individual foods by the factor weight associated with that food among the full control series and summing over all foods included in the specific patterns. Each subject was assigned scores indicative of the degree to which his/her diet adhered to each of the seven sex-specific dietary patterns. A previous study of the colon cancer series (5) documented no differences in findings for matched and unmatched analyses. We therefore opted for unmatched analyses to maximize the number of observations. In these earlier analyses, energy, total fat, and dietary fiber were examined separately as risk factors for colon cancer. The authors reported that they had obtained essentially the same values whether or not they controlled for age and education in their analyses and, thus, presented crude odds ratios. Using logistic regression, we have analyzed the same data and have controlled for these two additional variables. Therefore, any discrepancies, between these data and those presented previously are minor and result from these procedural changes. We estimated colon cancer risk using standard procedures for logistic regression analysis, with dietary patterns as well as energy, total fat, and dietary fiber intakes entered as independent variables of interest. Results
Table 2 describes the associations in the full set of control subjects between each of the patterns and intakes of nutrients commonly examined in studies of diet and cancer (energy, total fat, dietary fiber, and vitamins A and Q and for measures of diversity in fruit and vegetable use. The design of the original food frequency instrument is particularly strong in its ability to examine variety in use of fruits and vegetables as advocated in national dietary guidelines because of their potential abilities to lower cancer risk. We have defined diversity as the numbers of different fruits or vegetables included on the food frequency instrument that were consumed at least monthly (3). Among men, of those patterns associated most strongly (r 10.4) with fat intake (traditional, snacks, high-fat), the traditional pattern is unique in the strength of its associations with each of the other nutrients. The snacks pattern, associated positively with fiber intake, is unrelated to intakes of vitamins A and C, and the high-fat pattern is weakly associated with intakes of these three nutrients. The high-fat pattern is noteworthy for its lack of association with diversity in either vegetable or fruit consumption, because previous work with diet diversity suggests that this measure is higher among persons with higher nutrient intakes, including fat (6). The traditional pattern, on the other hand, is strongly positively related to diversity in vegetable consumption. The observations regarding diet diversity are consistent with the observed differences in intakes of dietary fiber and vitamins A and C. Among women, the patterns associated most strongly with high fat intakes (low-cost and high-fat) are less readily distinguishable. Both relate moderately to fiber intake and are either unrelated or only slightly related to intakes of vitamins A and C. The low-cost pattern is
268
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 2. Coefficients for Dietary Pattern Scores Correlated With Measures of Nutrient Intake and Diet Diversity Among Control Subjects, Western New York Diet Study, 1975-1986"* Nutrient Intake Measure
Dietary Pattern
Energy
Fat
Salad Fruit Healthful Traditional Snacks High fat Whole grain
0.19 0.32 0.06* 0.45 0.51 0.27 -0.32
0.16 0.18 0.04* 0.40 0.41 0.41 -0.31
Salad Healthful Low cost Fruit High fat Light Whole grain
0.22 0.29 0.43 0.29 0.51 0.06* 0.26
0.17 0.25 0.43 0.16 0.49 0.07* 0.16
Dietary fiber
Vitamin A
Diversity Measure Vitamin C
Vegetable
Fruit
Males (n = 1,475) 0.38 0.47 0.57 0.37 0.40 0.27 0.49 0.36 0.36 0.08* 0.14 0.03* -0.06* -0.02*
0.51 0.42 0.41 0.28 0.06* 0.09* 0.01*
0.51 0.31 0.57 0.50 0.13 0.03* -0.08*
0.32 0.73 0.32 0.24 0.16 -0.06* 0.09
Females (it = 780) 0.54 0.48 0.45 0.61 0.20 0.12 0.41 0.53 0.25 0.09* 0.20 0.31 0.37 0.17
0.59 0.47 0.14 0.46 0.02* 0.37 0.11*
0.57 0.66 0.23 0.29 0.18 0.25 0.11*
0.39 0.34 0.04* 0.63 0.12 0.34 0.21
a: Values are Pearson product-moment correlation coefficients. b: Statistical s ignificance is as follows *, not statistically significant at p < 0.001 (2 tailed).
unrelated and the high-fat pattern is only weakly related to diversity in fruit consumption; both patterns are only modestly related to vegetable diversity. Dietary Patterns and Colon Cancer Risk in Men Data in Tables 3-5 provide a ready comparison of individual nutrients and dietary patterns as risk factors for colon cancer in men. The odds ratios in the bivariate analyses indicate the level of risk associated with the individual variables alone. With both pattern score and nutrient intake in the logistic regression analyses, risk associated with each of the dietary patterns is estimated, controlling for the effects of energy, total fat, and dietary fiber, respectively. Likewise, risks associated with intakes of energy, fat, and dietary fiber are estimated, controlling for differences in dietary patterns. In the bivariate analyses, with dietary parameters treated as continuous variables, high-energy and total fat intakes are not identified as risk elevating; high-fat intake does approach statistical significance, however. The levels of risk estimated for dietary patterns tend to be stronger than those for nutrient intakes. The traditional, snacks, and high-fat patterns elevate risk significantly and the salad pattern approaches statistical significance as a protective factor. The three dietary patterns in men that are related to high intakes of energy and fat (traditional, snacks, and high-fat) elevate risk for colon cancer, but they contribute more to this risk than does either energy or total fat alone. In fact, controlling for the effects of these dietary patterns substantially lessens the colon cancer risk previously associated with energy intake. On the other hand, controlling for the effects of energy intake has virtually no effect on risk associated with these dietary patterns, except that it increases the apparent strength of the protective effects of the salad pattern. This latter observation supports our previous suggestion (4) that the salad pattern may not be indicative of a prudent eating style but may be associated with life-styles as different as health spas and steak houses.
Vol. 18, No. 3
269
Downloaded by [] at 10:54 04 February 2015
Table 3. Relative Risk of Colon Cancer Among Males Associated With Dietary Patterns and Energy, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.81 1.19
0.66-0.99 0.97-1.46
0.90
0.74-1.10
0.84 1.21
0.68-1.05 0.98-1.51
Healthful pattern Energy
0.96
0.79-1.17
0.94 1.15
0.77-1.15 0.94-1.41
Traditional pattern Energy
1.28
1.04-1.57
1.26 1.03
1.00-1.58 0.83-1.29
Snacks pattern Energy
1.31
1.07-1.60
1.32 0.98
1.04-1.69 0.77-1.24
High-fat pattern Energy
1.28
1.05-1.58
1.26 1.08
1.02-1.56 0.88-1.33
Whole-grain pattern Energy
0.89
0.73-1.09
0.93 1.11
0.75-1.15 0.90-1.38
Dietary Variables
OR*
95% CI
Energy intake
1.14
0.94-1.02
Salad pattern Energy
0.84
Fruit pattern Energy
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
When the effects of the traditional, snacks, and high-fat patterns are controlled, risk associated with high fat intake is diminished. Again, controlling for the effects of high fat intake allows the protective influence of the salad pattern to be expressed more clearly. It also diminishes the risk associated with the three patterns linked to high intakes of energy and fat. Although the elevations of risk associated with the traditional and high-fat patterns approach significance, only the snacks pattern continues to be statistically significant in elevating risk. We infer from these observations that in men several dietary patterns are better able to account for colon cancer risk than is either high-energy or total fat intake alone. There are risk factors incorporated into the patterns, in addition to the greater energy and fat content, and their influence may be most profound in the snacks pattern. The salad pattern includes protective factors that may be masked in persons consuming diets high in energy and total fat. Dietary fiber was not found to be protective for colon cancer for either gender, a finding consistent with previous findings by Graham and co-workers (5). In the current analyses we observed that controlling for the effects of the traditional and snacks patterns among men resulted in the protective effects of dietary fiber approaching significance. Controlling for fiber intake had no effect on risks associated with dietary patterns, except that it heightened the risk-elevating influences of the traditional and snacks patterns. Scores on both of these patterns correlate positively with fiber intake at levels of 0.36 < r < 0.49 (Table 2). We infer, therefore, that the total risk associated with these two patterns may include not only the elevation of risk associated with high energy and fat intakes but also the protective influence often attributed to higher fiber composition. The high-fat pattern, on
270
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 4. Relative Risk of Colon Cancelr Among Males Associated With Dietary Patterns and Total Fat Intake, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.81 1.23
0.66-0.99 1.00-1.51
0.90
0.74-1.10
0.86 1.23
0.70-1.05 1.00-1.52
Healthful pattern Total fat
0.96
0.79-1.17
0.94 1.19
0.77-1.15 0.97-1.46
Traditional pattern Total fat
1.28
1.04-1.57
1.23 1.08
0.98-1.55 0.87-1.35
Snacks pattern Total fat
1.31
1.07-1.60
1.27 1.05
1.01-1.60 0.84-1.32
High-fat pattern Total fat
1.28
1.05-1.58
1.24 1.09
1.00-1.55 0.88-1.36
Whole-grain pattern Total fat
0.89
0.73-1.09
0.93 1.16
0.76-1.15 0.94-1.43
Dietary Variables
OR*
95% CI
Total fat intake
1.18
0.97-1.45
Salad pattern Total fat
0.84
Fruit pattern Total fat
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
the other hand, is unrelated to fiber intake, and its risk estimation is unaffected by control for fiber intake. Dietary Patterns and Colon Cancer Risk in Women Both nutrient intake and dietary patterns are important risk factors in women, although the patterns are better able to identify protective factors. Tables 6-8 present analyses of dietary patterns and colon cancer risk in women, controlling for intake of energy, total fat, and dietary fiber, respectively. The ability of dietary patterns to account for colon cancer risk appears to be different for men and women. Among women, both energy and total fat are found to elevate risk in the bivariate analyses. No dietary patterns are identified as risk elevating, but two patterns (salad and light) are found to be protective. Risk associated with high energy intake is increased by controlling for the effects of the salad, high-fat, and whole-grain patterns. Controlling for the effects of energy intake, on the other hand, increases the protection afforded by the salad and whole-grain patterns and yet has no effect on the protection offered by the light pattern. Similar results are obtained when the patterns and fat intake are entered jointly into logistic regression analyses. Fiber was unrelated to colon cancer risk in bivariate analyses, and controlling for fiber in logistic regression analyses has no effect on the patterns, except to enhance the protection afforded by the salad pattern. Interpreting the results for women is problematic. The light pattern is unrelated to energy or fat intake, and controlling for these two parameters has no influence on the protective effect of this pattern. This pattern is positively related (r = 0.20) to fiber intake, which itself is unrelated to colon cancer risk; controlling for fiber intake has no effect on the protection afforded by the light pattern.
Vol. 18, No. 3
271
Downloaded by [] at 10:54 04 February 2015
Table 5. Relative Risk of Colon Cancer Among Males Associated With Dietary Patterns and Dietary Fiber, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.83 1.04
0.66-1.03 0.84-1.29
0.90
0.74-1.10
0.90 1.02
0.71-1.13 0.81-1.29
Healthful pattern Dietary fiber
0.96
0.79-1.17
0.97 0.97
0.79-1.20 0.79-1.19
Traditional pattern Dietary fiber
1.28
1.04-1.57
1.42 0.81
1.12-1.81 0.64-1.02
Snacks pattern Dietary fiber
1.31
1.07-1.60
1.40 0.84
1.12-1.75 0.68-1.05
High-fat pattern Dietary fiber
1.28
1.05-1.58
1.29 0.95
1.05-1.58 0.78-1.15
Whole-grain pattern Dietary fiber
0.89
0.73-1.09
0.88 0.94
0.72-1.08 0.77-1.15
Dietary Variables
OR*
95% CI
Dietary fiber intake
0.96
0.79-1.17
Salad pattern Dietary fiber
0.84
Fruit pattern Dietary fiber
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
The salad pattern, on the other hand, is related positively to energy (r = 0.22) and total fat (r = 0.17) intakes. Controlling for the effect of the pattern exacerbates the risks associated with energy and fat, and controlling for the effects of energy and fat intakes enhances the protection offered by the salad pattern. Discussion Our work with dietary patterns is still in a developmental phase. We have adopted one statistical approach to capturing the multidimensional nature of diet, that of factor analysis. It was selected because of its ability to identify foods that are incorporated jointly into diets in a free-living population. Because the same nutrients are available from a variety of foods, when the focus of analysis is on habitual food use, the measure may be influenced more directly by behaviors than when the focus is on nutrient intake. Food choices are linked to such other health-related behaviors as smoking, alcohol use, and prudent consumption of fat (4) and also may be linked to additional behaviors such as occupation, stress management, intransigence regarding current health guidelines, and acceptance of risk in general; these traits have not yet been addressed in our analyses. Mechanisms of cancer development must be explained in biological terms. With dietary exposure to cancer risk expressed as intake of a specific nutrient, the investigator possesses a measure with a clear potential biologic effect. Understanding what it is intended to capture is relatively straightforward. However, such an approach may fail to incorporate the metabolic consequences that dietary factors consumed concurrently exert on total risk (1-3,9,10). To develop a more comprehensive measure of dietary exposure to risk, we have shifted the focus of measurement from nutrients to foods, and by so doing we are less able to define the nutrient
272
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 6. Relative Risk of Colon Cancer Among Females Associatec With Dietary Patterns and Energy, Western New York Diet Study, 1975-1984° ORC
95% CI
0.60-0.89
0.67 1.37
0.54-0.83 1.12-1.68
1.00
0.83-1.21
0.95 1.24
0.78-1.15 1.02-1.52
Low-cost pattern Energy
1.15
0.94-1.42
1.08 1.20
0.86-1.34 0.97-1.47
Fruit pattern Energy
1.08
0.90-1.31
1.01 1.22
0.83-1.24 1.00-1.50
High-fat pattern Energy
0.99
0.82-1.20
0.86 1.33
0.69-1.07 1.06-1.67
Light pattern Energy
0.77
0.63-0.93
0.77 1.23
0.63-0.94 1.01-1.49
Whole-grain pattern Energy
0.88
0.73-1.07
0.80 1.32
0.65-0.99 1.07-1.63
Dietary Variables
OR*
95% CI
Energy intake
1.23
1.01-1.49
Salad pattern Energy
0.73
Healthful pattern Energy
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
mechanisms that the measure might incorporate. However, we are closer to behaviors that can be modified than the investigator working with nutrient indexes. We argue that this focus on the way foods are grouped naturally is of particular value for describing the association of risk and behavior. Our previous work (2-4) suggests that multiple risk factors, both dietary and nondietary, coalesce within dietary patterns that may be shaped by life-styles. We assume that additional correlates of dietary patterns contribute to cancer risk but have yet to be examined. Despite these limitations, results of this study indicate that the measures of dietary patterns do provide additional information concerning the role of diet in cancer risk beyond that obtained with measures of intake of single nutrients, such as energy, total fat, or dietary fiber. Extensive preliminary investigations were completed (1-4) before this study, in which the patterns themselves and their associations with parameters implicated in cancer risk were examined using the full set of control subjects only. We followed that strategy to improve our understanding of such multidimensional measures of diet without adding potential bias in study design by identifying differences in case and control subjects before undertaking this case-control study of risk. These investigations substantiate additional distinctions among the patterns that may contribute to disease risk and to the aggregate risk for the various dietary patterns (3,4). These differences include behaviors affecting ingestion of saturated fat; variety of fruits and vegetables routinely eaten; total alcohol intake as well as type of alcoholic beverages consumed; intake of preserved and barbecued foods; and smoking, both current and lifelong exposures. Among men, dietary patterns account for more variability in colon cancer risk than do intakes of single nutrients, specifically total fat. We have focused attention on those three patterns correlated most strongly with high intakes of fat that would be likely to be expressed
Vol. 18, No. 3
273
Downloaded by [] at 10:54 04 February 2015
Table 7. Relative Risk of Colon Cancer Among Females Associated With Dietary Patterns and Total Fat, Western New York Diet Study, ] 975-1984" Dietary Variables
OR*
95% CI
OR'
95% CI
Total fat intake
1.28
1.05-1.55
Salad pattern Total fat
0.73
0.60-0.89
0.68 1.39
0.56-0.84 1.13-1.71
Healthful pattern Total fat
1.00
0.83-1.21
0.96 1.29
0.79-1.16 1.05-1.57
Low-cost pattern Total fat
1.15
0.94-1.42
1.06 1.25
0.85-1.32 1.01-1.54
Fruit pattern Total fat
1.08
0.90-1.31
1.03 1.27
0.85-1.25 1.04-1.55
High-fat pattern Total fat
0.99
0.82-1.20
0.84 1.41
0.67-1.05 1.11-1.78
Light pattern Total fat
0.77
0.63-0.93
0.77 1.27
0.63-0.94 1.04-1.55
Whole-grain pattern Total fat
0.88
0.73-1.07
0.82 1.34
0.67-1.00 1.09-1.65
a: Relative risk was assessed controlling for age and education, b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
by subjects categorized into the highest quartile of fat intake. The high-fat pattern in men is distinctive in incorporating a large number of additional risk-elevating factors (3,4), including cigarette and alcohol use. Accounting for the sources of the protection from colon cancer for women that are inherent in the salad and light patterns in women is difficult; this study leaves unresolved the source of this protection. Dietary fiber was found to be unrelated to risk definition. The previous methodological studies with the dietary patterns (3,4) indicate that there are no differences related to scores on these patterns for percentage of calories derived from fat or for the types of fat used, except that women with diets consistent with these patterns were more likely to cook with vegetable oil and to add salad oil to foods (4). The salad and light patterns were not distinguishable from other patterns on the bases of the use of fruits and vegetables or preserved or charbroiled foods. Alcohol consumption was unrelated to the salad pattern and actually is related positively to the light pattern (3,4). Lifelong exposure to cigarette smoke was unrelated to either pattern. Scores were higher in exsmokers than in current smokers on the light pattern, and there was no significant difference on this variable for the salad pattern. We suspect that underlying health behaviors tend to be less related to the dietary patterns in women than they are in men; women may adopt health-promoting behaviors into their lives regardless of their eating styles. This conclusion is consistent with earlier researchers who have documented that women are more likely than their spouses to adjust their diets to accommodate food choices of household members (11). Therefore, eating styles and personal health values may be less strongly associated for women. Men and women appear to differ in the aggregation of cancer risks incorporated into thendiets; they also differ in the health behaviors and patterns of ingestion associated with diet. The result of these two sources of difference may result in distinctive exposures to risk for
274
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 8. Relative Risk of Colon Cancer Among Females Associated With Dietary Patterns and Dietary Fiber, Western New York Diet Study, 1975-1984" ORC
95% CI
0.60-0.89
0.67 1.19
0.53-O.84 0.95-1.48
1.00
0.83-1.21
1.02 0.96
0.83-1.26 0.78-1.19
Low-cost pattern Dietary fiber
1.15
0.94-1.42
1.18 0.93
0.95-1.46 0.77-1.13
Fruit pattern Dietary fiber
1.08
0.90-1.31
1.16 0.89
0.92-1.47 0.70-1.12
High-fat pattern Dietary fiber
0.99
0.82-1.20
1.00 0.97
0.83-1.22 0.80-1.18
Light pattern Dietary fiber
0.77
0.63-0.93
0.76 1.01
0.63-0.93 0.84-1.23
Whole-grain pattern Dietary fiber
0.88
0.73-1.07
0.88 1.02
0.71-1.08 0.83-1.25
Dietary Variables
OR*
95% CI
Dietary fiber intake
0.97
0.81-1.17
Salad pattern Dietary fiber
0.73
Healthful pattern Dietary fiber
a: Relative risk was assessed controlling for age and education, b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
the two genders, not only for colon cancer but for other cancer sites as well as for other diet-related chronic diseases. These dietary patterns are not presented as final measures; whether they will be useful in other settings among other populations requires further study. These factors may need to be refined through further methodological research. One feature that may limit their ability to explain cancer risk is the low percentage of variance in food use explained by them; the seven patterns together account for only 20.3% and 21.7% of the total variance in food intake among males and females, respectively. This problem undoubtedly is a function of the large number of foods (95) included in the factoring process. Although dietary patterns were able to account for a greater amount of variance in the Bogalusa Heart Study (7) when a similar analytic approach was used, the food list in that study was considerably shorter. Because knowledge of dietary patterns is still in a developmental phase, there was no exclusion criterion for foods available to us except for infrequency of food use. The problem is reminiscent of earlier efforts of Block and co-workers (8) to develop a food frequency instrument of feasible size. The ability of a food frequency instrument to explain variance in nutrient intakes was increased by grouping highly specific items into broader composite foods. Multidimensional measures of diet, such as dietary patterns, may improve risk estimation by incorporating dietary risk more comprehensively. Should the strongest risk factors be substances other than the nutrients selected in a specific study, such as folate or phytochemicals, multidimensional measures may still capture that risk. Their superior performance may also be attributable to greater sensitivity to related nondietary health behaviors associated with life-style characteristics. Willett and associates (12) recently demonstrated the value of using meats and the ratio of intakes of meat to chicken and fish in accounting for risk in a prospective study of colon cancer in women. We suggest that epidemiological studies could
Vol. 18, No. 3
275
profit from evaluating factors such as these in addition to a priori indexes of fat, energy, fiber, and vitamins A, C, and E. They deserve further methodological development; in addition, their stability over region and time should be addressed. Acknowledgments and Notes This study was funded by National Cancer Institute Grants CA-11535 and RO-CA-51720-01 from the National Institutes of Health (Bethesda, MD). Address reprint requests to Dr. D. Elizabeth Randall, Nutrition Program, 301 Parker Hall, S.U.N.Y. at Buffalo, Buffalo, NY 14214. Submitted 26 May 1992; accepted in final form 5 August 1992.
Downloaded by [] at 10:54 04 February 2015
References 1. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "Frequency of Food Use Data and the Multidimensionality of Diet." J Am Diet Assoc 89, 1070-1075, 1989. 2. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "Patterns in Food Use and Their Associations With Nutrient Intakes." Am J Clin Nutr 52, 739-745, 1990. 3. Randall, E, Marshall, JR, Brasure, J, and Graham, S: "Patterns in Food Use and Compliance With NCI Dietary Guidelines." Nutr Cancer 15, 141-156, 1991. 4. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "High Risk Health Behaviors Associated With Various Dietary Patterns." Nutr Cancer 16, 135-151, 1991. 5. Graham, S, Marshall, J, Haughey, B, Mittelman, A, Swanson, M, et al.: "Dietary Epidemiology of Cancer of the Colon in Western New York." Am J Epidemiol 128, 490-503, 1988. 6. Randall, E, Nichaman, MZ, and Contant, CF, Jr.: "Diet Diversity and Nutrient Intake." J Am Diet Assoc 85, 830-836, 1985. 7. Nicklas, TA, Webber, LS, Thompson, B, and Berenson, GS: "A Multivariate Model for Assessing Eating Patterns and Their Relations to Cardiovascular Risk Factors: the Bogalusa Heart Study." Am J Clin Nutr 49, 320-327, 1989. 8. Block, G, Dresser, CM, Hartman, AM, and Carroll, MD: "Nutrient Sources in the American Diet: Quantitative Data From the NHANES II Survey. II. Macronutrients and Fats." Am J Epidemiol 122, 27-40, 1985. 9. Modan, B, Barell, V, Lubin, F, Modan, M, Greenberg, RA, et al.: "Low-Fiber Intake as an Etiologic Factor in Cancer of the Colon." JNCI 55, 15-18, 1975. 10. Freudenheim, JL, Graham, S, Marshall, JR, Haughey, BP, and Wilkinson, G.: "A Case-Control Study of Diet and Rectal Cancer in Western New York." Am J Epidemiol 131, 612-624, 1990. 11. Davis, MA, Randall, E, Forthofer, RN, Lee, ES, and Margen, S: "Living Arrangements and Dietary Patterns of Older Adults in the United States." J Gerontol 40, 434-442, 1985. 12. Willett, WC, Stampfer, MJ, Colditz, GA, Rosner, BA, and Speizer, FE.: "Relation of Meat, Fat, and Fiber Intake to the Risk of Colon Cancer in a Prospective Study Among Women." N Engl J Med 323, 1664-1672, 1990.
276
Nutrition and Cancer 1992
Nutrition and Cancer Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/hnuc20
Dietary patterns and colon cancer in western New York a
b
b
Elizabeth Randall , James R. Marshall , John Brasure & Saxon Graham
b
a
Nutrition Program , State University of New York , 301 Parker Hall, Buffalo, NY, 14214 b
Department of Social and Preventive Medicine , State University of New York , Buffalo, NY, 14214 Published online: 04 Aug 2009.
To cite this article: Elizabeth Randall , James R. Marshall , John Brasure & Saxon Graham (1992) Dietary patterns and colon cancer in western New York, Nutrition and Cancer, 18:3, 265-276, DOI: 10.1080/01635589209514227 To link to this article: http://dx.doi.org/10.1080/01635589209514227
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
Dietary Patterns and Colon Cancer in Western New York Elizabeth Randall, James R. Marshall, John Brasure, and Saxon Graham
Downloaded by [] at 10:54 04 February 2015
Abstract Seven dietary patterns were identified among control subjects in the Western New York Diet Study (1975-1986) by application of principal components analysis to data from a 95-item food frequency interview. The results of case-control analyses of colon cancer risk for these patterns are presented. Cases were matched with neighborhood controls on the bases of age and sex; 205 colon case-control male and 223 female pairs were obtained. The dietary patterns and intakes of energy, total fat, and dietary fiber were examined with logistic regression for their individual contributions to risk. In males, three of these dietary patterns were associated positively with fat and energy consumption; they elevated risk for colon cancer and accountedfor more risk than did the specific nutrients. Control for energy and fat intakes allowed the protective influences of additional dietary patterns to be expressed. No patterns elevated risk in women; two patterns were protective for colon cancer. Controlling for energy and fat intake enhanced the protection afforded by one of these patterns but had no influence on that of the other. Measures of foods rather than single nutrients may be more inclusive of dietary exposures to risk as well as being related more directly to underlying health behaviors. Therefore they may be better able to account for risk in diseases with multiple causation. (Nutr Cancer 18, 265-276, 1992)
Introduction
People eat meals consisting of a variety of foods, and as we have discovered in our previous inquiries, habitual food use can be expressed as dietary patterns (1-3). It is obvious that individuals do not ingest single dietary constituents in isolation but, rather, typically ingest several at the same time. Inasmuch as combinations of nutrients, as well as additional nonnutrient dietary constituents, may have different metabolic consequences depending on the combinations in which they are consumed, it is important to determine whether the patterns in food use are associated either positively or negatively with cancer. Previously we demonstrated that locating individuals within quartiles of nutrient intakes simultaneously arrays them on other dietary characteristics as well (1). We identified gender-specific dietary patterns (2,3) among the 1,475 male and 780 female normal healthy subjects in the complete control series of the Western New York Diet Study. These dietary patterns can be characterized by their degree of correlation with the various single nutrients linked to cancer risk (2) as well as by their associations with exposures to nonnutrient dietary factors and to such discretionary health behaviors as alcohol consumption and cigarette E. Randall is affiliated with the Nutrition Program and J. R. Marshall, J. Brasure, and S. Graham with the Department of Social and Preventive Medicine, State University of New York at Buffalo, Buffalo, NY 14214.
Copyright © 1992, Lawrence Erlbaum Associates, Inc.
smoking (4). Multiple risk factors coalesce within several of the dietary patterns, notably those associated with high fat intake, in ways that suggest that putative dietary risks may exceed those of single nutrients (3,4). In earlier analyses using this data set (5), energy, total fat, and dietary fiber were examined separately as risk factors for colon cancer. Among men, both energy and total fat intakes were found to elevate risk. Energy and fat were so collinear that their individual contributions to risk were indistinguishable. Total dietary fiber intake was unrelated to colon cancer risk. For women, total fat elevated risk, but less strongly than for men. The purpose of this study was to examine the role of the dietary patterns in the risk of colon cancer to determine whether the patterns identified were correlated with risk of this high-incidence cancer. We found that certain dietary patterns were more strongly associated with risk of colon cancer than any single nutrient, including fat. Methods
Downloaded by [] at 10:54 04 February 2015
Sample Characteristics
The Western New York Diet Study is a case-control study of diet and cancer of the lung, esophagus, larynx, mouth, stomach, bladder, rectum, and colon that was conducted in Erie, Niagara, and Monroe Counties of Western New York in 1975-1986. The procedures followed have been reported in detail elsewhere (5). Cases were identified by nurse interviewers using hospital records. Those cases whose participation was approved by their physicians and who agreed to be interviewed were recruited and matched to control subjects on the bases of age, gender, and neighborhood of residence. Each subject completed an extensive 2.5-hour interview, including a 128-item semiquantitative food frequency instrument supplemented by questions concerning food preparation and storage practices. Validity and reliability data for the interview are available elsewhere (5). The colon cancer case-control series was completed during 1975-1984 and consisted of 410 males and 446 females of whom equal numbers were case and control subjects. Dietary Data Subjects were queried regarding their frequency of use of 128 food items during the 12-month period preceding the interview for control subjects or the 12 months before cancer diagnosis for the cases. Nutrient intakes were calculated on the basis of the complete food list. Dietary patterns were identified in the full series of control subjects by use of principal components analysis with orthogonal rotation of the factors. This technique identifies factors in which cohesiveness of the constituent foods, expressed as factor weights, is maximized. To enhance their definition, factors were extracted from a subset of the 95 food items from the original instrument that had a mean frequency of consumption of >0.5 times a month, so that foods consumed only infrequently could not be influential in identifying patterns. Foods consumed rarely could be strongly cohesive within individual factors and, therefore, could be highly influential in pattern identification without making any significant contribution to the dietary status of the individual consuming them. In addition, only actual foods (rather than sauces, condiments, or spreads) were included in pattern definition. Full descriptions of these procedures are provided elsewhere (2,3). Although some items were excluded from the original list of 128 food items, associations between their use and pattern scores have been investigated separately (3,4). The seven patterns identified for males and for females are summarized in Table 1. They are listed in order of their contributions to variance in food use, and foods within patterns are listed in the order of the strength of their associations with the specific patterns. Names were assigned to patterns merely to describe the dominant foods within them. High-fat patterns
266
Nutrition and Cancer 1992
Table 1. Dietary Patterns Identified, Western New York Diet Study, 1975-1986 Dietary Pattern Salad Fruit Healthful Traditional Snacks High fat
Downloaded by [] at 10:54 04 February 2015
Whole grain
Salad Healthful
Low cost Fruit High fat Light Whole grain
Foods" Use Males (n = 1,475) Lettuce, celery, green pepper, cucumber, tomatoes, radishes, onions, carrots Peaches, plums, pears, apples, oranges or tangerines, cherries, grapes, melons, bananas, berries Broccoli, spinach, cauliflower, rice, poultry, mushrooms, green beans, summer squash, asparagus Wax beans, green beans, potatoes, peas, beets, cabbage, roast beef, cakes, pies, corn Cookies, candy, crackers, donuts or pastries, hamburger, ice cream, baked beans Eggs, bacon, sausage, steak, beer, salami or pepperoni, distilled alcohol Whole wheat bread, cooked cereal, oatmeal
Females (n = 780) Lettuce, onion, cucumber, tomato, celery, green pepper, radishes, carrot, apples, broccoli Green beans, cauliflower, spinach, beets, summer squash, pineapple, liver, wax beans, broccoli, winter squash, cottage cheese, mushrooms, peas, cabbage Macaroni, spaghetti, bologna, hot dogs, french fries, noodles, hamburger, processed cheese, salami or pepperoni, canned fish, coffee Pears, peaches, plums, oranges or tangerines, apples, bananas, grapes, grapefruit Pies, candy, cake, potatoes, roast beef, donut or pastries, bacon, rolls, potato chips Mushrooms, lemons or limes, hard cheese, fin fish, yogurt, melon, broccoli Cooked cereal, cookies, dried fruit, unsweetened cereal, crackers, oat products, whole wheat bread, nuts, yogurt
Nonuse* None identified Beer Potatoes, cookies, white bread None identified None identified Unsweetened cereal, winter squash, summer squash, cookies White bread, spaghetti, macaroni, coffee, french fries, hamburger, noodles, cola None identified None identified
Asparagus Distilled alcohol, beer, wine Poultry, yogurt Corn, white bread, peas, potatoes Coffee, white bread
a: Listed in order of factor weight. b: Use of these foods is negatively related to the pattern.
were so named because of the fat density of the constituent foods. The traditional pattern includes meats, several common vegetables, and baked desserts; the low-cost pattern includes foods that are relatively inexpensive. The healthful patterns appear to be consistent with the current trend toward the consumption of lean and fiber-rich foods. Data Analysis Preliminary analyses in which pattern scores were calculated using all 95 food items were highly similar in their abilities to estimate colon cancer risk to patterns composed only of their strongest elements. Therefore, we deviated from the standard procedures of factor analysis by including in the patterns only those foods that load most heavily on each factor.
Vol. 18, No. 3
267
Downloaded by [] at 10:54 04 February 2015
To do otherwise would have required including all 95 food items in each pattern and would have resulted in data unwieldy -both conceptually and economically. As a consequence, however, some dietary patterns are intercorrelated. This effect is expressed most strongly in men, with positive correlations among the salad, fruit, healthful, and traditional patterns (0.18 < r < 0.38) and negative correlations between the whole-grain and the snacks patterns (r = -0.22) as well as with the high-fat pattern (r = -0.18). For women, positive intercorrelations were strongest among the salad, healthful, fruit, and light patterns (0.18 < r < 0.42), whereas the only strong negative association was between the high-fat and light patterns (r = -0.24). Such relatedness of the patterns is consistent with the propensity of human beings to change food behaviors in response to changes in living situations; individuals would be likely to exhibit more than one dietary pattern. Pattern scores were calculated in the colon series, for both case and control subjects, by multiplying the standardized reported frequency of intake of individual foods by the factor weight associated with that food among the full control series and summing over all foods included in the specific patterns. Each subject was assigned scores indicative of the degree to which his/her diet adhered to each of the seven sex-specific dietary patterns. A previous study of the colon cancer series (5) documented no differences in findings for matched and unmatched analyses. We therefore opted for unmatched analyses to maximize the number of observations. In these earlier analyses, energy, total fat, and dietary fiber were examined separately as risk factors for colon cancer. The authors reported that they had obtained essentially the same values whether or not they controlled for age and education in their analyses and, thus, presented crude odds ratios. Using logistic regression, we have analyzed the same data and have controlled for these two additional variables. Therefore, any discrepancies, between these data and those presented previously are minor and result from these procedural changes. We estimated colon cancer risk using standard procedures for logistic regression analysis, with dietary patterns as well as energy, total fat, and dietary fiber intakes entered as independent variables of interest. Results
Table 2 describes the associations in the full set of control subjects between each of the patterns and intakes of nutrients commonly examined in studies of diet and cancer (energy, total fat, dietary fiber, and vitamins A and Q and for measures of diversity in fruit and vegetable use. The design of the original food frequency instrument is particularly strong in its ability to examine variety in use of fruits and vegetables as advocated in national dietary guidelines because of their potential abilities to lower cancer risk. We have defined diversity as the numbers of different fruits or vegetables included on the food frequency instrument that were consumed at least monthly (3). Among men, of those patterns associated most strongly (r 10.4) with fat intake (traditional, snacks, high-fat), the traditional pattern is unique in the strength of its associations with each of the other nutrients. The snacks pattern, associated positively with fiber intake, is unrelated to intakes of vitamins A and C, and the high-fat pattern is weakly associated with intakes of these three nutrients. The high-fat pattern is noteworthy for its lack of association with diversity in either vegetable or fruit consumption, because previous work with diet diversity suggests that this measure is higher among persons with higher nutrient intakes, including fat (6). The traditional pattern, on the other hand, is strongly positively related to diversity in vegetable consumption. The observations regarding diet diversity are consistent with the observed differences in intakes of dietary fiber and vitamins A and C. Among women, the patterns associated most strongly with high fat intakes (low-cost and high-fat) are less readily distinguishable. Both relate moderately to fiber intake and are either unrelated or only slightly related to intakes of vitamins A and C. The low-cost pattern is
268
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 2. Coefficients for Dietary Pattern Scores Correlated With Measures of Nutrient Intake and Diet Diversity Among Control Subjects, Western New York Diet Study, 1975-1986"* Nutrient Intake Measure
Dietary Pattern
Energy
Fat
Salad Fruit Healthful Traditional Snacks High fat Whole grain
0.19 0.32 0.06* 0.45 0.51 0.27 -0.32
0.16 0.18 0.04* 0.40 0.41 0.41 -0.31
Salad Healthful Low cost Fruit High fat Light Whole grain
0.22 0.29 0.43 0.29 0.51 0.06* 0.26
0.17 0.25 0.43 0.16 0.49 0.07* 0.16
Dietary fiber
Vitamin A
Diversity Measure Vitamin C
Vegetable
Fruit
Males (n = 1,475) 0.38 0.47 0.57 0.37 0.40 0.27 0.49 0.36 0.36 0.08* 0.14 0.03* -0.06* -0.02*
0.51 0.42 0.41 0.28 0.06* 0.09* 0.01*
0.51 0.31 0.57 0.50 0.13 0.03* -0.08*
0.32 0.73 0.32 0.24 0.16 -0.06* 0.09
Females (it = 780) 0.54 0.48 0.45 0.61 0.20 0.12 0.41 0.53 0.25 0.09* 0.20 0.31 0.37 0.17
0.59 0.47 0.14 0.46 0.02* 0.37 0.11*
0.57 0.66 0.23 0.29 0.18 0.25 0.11*
0.39 0.34 0.04* 0.63 0.12 0.34 0.21
a: Values are Pearson product-moment correlation coefficients. b: Statistical s ignificance is as follows *, not statistically significant at p < 0.001 (2 tailed).
unrelated and the high-fat pattern is only weakly related to diversity in fruit consumption; both patterns are only modestly related to vegetable diversity. Dietary Patterns and Colon Cancer Risk in Men Data in Tables 3-5 provide a ready comparison of individual nutrients and dietary patterns as risk factors for colon cancer in men. The odds ratios in the bivariate analyses indicate the level of risk associated with the individual variables alone. With both pattern score and nutrient intake in the logistic regression analyses, risk associated with each of the dietary patterns is estimated, controlling for the effects of energy, total fat, and dietary fiber, respectively. Likewise, risks associated with intakes of energy, fat, and dietary fiber are estimated, controlling for differences in dietary patterns. In the bivariate analyses, with dietary parameters treated as continuous variables, high-energy and total fat intakes are not identified as risk elevating; high-fat intake does approach statistical significance, however. The levels of risk estimated for dietary patterns tend to be stronger than those for nutrient intakes. The traditional, snacks, and high-fat patterns elevate risk significantly and the salad pattern approaches statistical significance as a protective factor. The three dietary patterns in men that are related to high intakes of energy and fat (traditional, snacks, and high-fat) elevate risk for colon cancer, but they contribute more to this risk than does either energy or total fat alone. In fact, controlling for the effects of these dietary patterns substantially lessens the colon cancer risk previously associated with energy intake. On the other hand, controlling for the effects of energy intake has virtually no effect on risk associated with these dietary patterns, except that it increases the apparent strength of the protective effects of the salad pattern. This latter observation supports our previous suggestion (4) that the salad pattern may not be indicative of a prudent eating style but may be associated with life-styles as different as health spas and steak houses.
Vol. 18, No. 3
269
Downloaded by [] at 10:54 04 February 2015
Table 3. Relative Risk of Colon Cancer Among Males Associated With Dietary Patterns and Energy, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.81 1.19
0.66-0.99 0.97-1.46
0.90
0.74-1.10
0.84 1.21
0.68-1.05 0.98-1.51
Healthful pattern Energy
0.96
0.79-1.17
0.94 1.15
0.77-1.15 0.94-1.41
Traditional pattern Energy
1.28
1.04-1.57
1.26 1.03
1.00-1.58 0.83-1.29
Snacks pattern Energy
1.31
1.07-1.60
1.32 0.98
1.04-1.69 0.77-1.24
High-fat pattern Energy
1.28
1.05-1.58
1.26 1.08
1.02-1.56 0.88-1.33
Whole-grain pattern Energy
0.89
0.73-1.09
0.93 1.11
0.75-1.15 0.90-1.38
Dietary Variables
OR*
95% CI
Energy intake
1.14
0.94-1.02
Salad pattern Energy
0.84
Fruit pattern Energy
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
When the effects of the traditional, snacks, and high-fat patterns are controlled, risk associated with high fat intake is diminished. Again, controlling for the effects of high fat intake allows the protective influence of the salad pattern to be expressed more clearly. It also diminishes the risk associated with the three patterns linked to high intakes of energy and fat. Although the elevations of risk associated with the traditional and high-fat patterns approach significance, only the snacks pattern continues to be statistically significant in elevating risk. We infer from these observations that in men several dietary patterns are better able to account for colon cancer risk than is either high-energy or total fat intake alone. There are risk factors incorporated into the patterns, in addition to the greater energy and fat content, and their influence may be most profound in the snacks pattern. The salad pattern includes protective factors that may be masked in persons consuming diets high in energy and total fat. Dietary fiber was not found to be protective for colon cancer for either gender, a finding consistent with previous findings by Graham and co-workers (5). In the current analyses we observed that controlling for the effects of the traditional and snacks patterns among men resulted in the protective effects of dietary fiber approaching significance. Controlling for fiber intake had no effect on risks associated with dietary patterns, except that it heightened the risk-elevating influences of the traditional and snacks patterns. Scores on both of these patterns correlate positively with fiber intake at levels of 0.36 < r < 0.49 (Table 2). We infer, therefore, that the total risk associated with these two patterns may include not only the elevation of risk associated with high energy and fat intakes but also the protective influence often attributed to higher fiber composition. The high-fat pattern, on
270
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 4. Relative Risk of Colon Cancelr Among Males Associated With Dietary Patterns and Total Fat Intake, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.81 1.23
0.66-0.99 1.00-1.51
0.90
0.74-1.10
0.86 1.23
0.70-1.05 1.00-1.52
Healthful pattern Total fat
0.96
0.79-1.17
0.94 1.19
0.77-1.15 0.97-1.46
Traditional pattern Total fat
1.28
1.04-1.57
1.23 1.08
0.98-1.55 0.87-1.35
Snacks pattern Total fat
1.31
1.07-1.60
1.27 1.05
1.01-1.60 0.84-1.32
High-fat pattern Total fat
1.28
1.05-1.58
1.24 1.09
1.00-1.55 0.88-1.36
Whole-grain pattern Total fat
0.89
0.73-1.09
0.93 1.16
0.76-1.15 0.94-1.43
Dietary Variables
OR*
95% CI
Total fat intake
1.18
0.97-1.45
Salad pattern Total fat
0.84
Fruit pattern Total fat
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
the other hand, is unrelated to fiber intake, and its risk estimation is unaffected by control for fiber intake. Dietary Patterns and Colon Cancer Risk in Women Both nutrient intake and dietary patterns are important risk factors in women, although the patterns are better able to identify protective factors. Tables 6-8 present analyses of dietary patterns and colon cancer risk in women, controlling for intake of energy, total fat, and dietary fiber, respectively. The ability of dietary patterns to account for colon cancer risk appears to be different for men and women. Among women, both energy and total fat are found to elevate risk in the bivariate analyses. No dietary patterns are identified as risk elevating, but two patterns (salad and light) are found to be protective. Risk associated with high energy intake is increased by controlling for the effects of the salad, high-fat, and whole-grain patterns. Controlling for the effects of energy intake, on the other hand, increases the protection afforded by the salad and whole-grain patterns and yet has no effect on the protection offered by the light pattern. Similar results are obtained when the patterns and fat intake are entered jointly into logistic regression analyses. Fiber was unrelated to colon cancer risk in bivariate analyses, and controlling for fiber in logistic regression analyses has no effect on the patterns, except to enhance the protection afforded by the salad pattern. Interpreting the results for women is problematic. The light pattern is unrelated to energy or fat intake, and controlling for these two parameters has no influence on the protective effect of this pattern. This pattern is positively related (r = 0.20) to fiber intake, which itself is unrelated to colon cancer risk; controlling for fiber intake has no effect on the protection afforded by the light pattern.
Vol. 18, No. 3
271
Downloaded by [] at 10:54 04 February 2015
Table 5. Relative Risk of Colon Cancer Among Males Associated With Dietary Patterns and Dietary Fiber, Western New York Diet Study, 1975-1984" ORC
95% CI
0.69-1.02
0.83 1.04
0.66-1.03 0.84-1.29
0.90
0.74-1.10
0.90 1.02
0.71-1.13 0.81-1.29
Healthful pattern Dietary fiber
0.96
0.79-1.17
0.97 0.97
0.79-1.20 0.79-1.19
Traditional pattern Dietary fiber
1.28
1.04-1.57
1.42 0.81
1.12-1.81 0.64-1.02
Snacks pattern Dietary fiber
1.31
1.07-1.60
1.40 0.84
1.12-1.75 0.68-1.05
High-fat pattern Dietary fiber
1.28
1.05-1.58
1.29 0.95
1.05-1.58 0.78-1.15
Whole-grain pattern Dietary fiber
0.89
0.73-1.09
0.88 0.94
0.72-1.08 0.77-1.15
Dietary Variables
OR*
95% CI
Dietary fiber intake
0.96
0.79-1.17
Salad pattern Dietary fiber
0.84
Fruit pattern Dietary fiber
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
The salad pattern, on the other hand, is related positively to energy (r = 0.22) and total fat (r = 0.17) intakes. Controlling for the effect of the pattern exacerbates the risks associated with energy and fat, and controlling for the effects of energy and fat intakes enhances the protection offered by the salad pattern. Discussion Our work with dietary patterns is still in a developmental phase. We have adopted one statistical approach to capturing the multidimensional nature of diet, that of factor analysis. It was selected because of its ability to identify foods that are incorporated jointly into diets in a free-living population. Because the same nutrients are available from a variety of foods, when the focus of analysis is on habitual food use, the measure may be influenced more directly by behaviors than when the focus is on nutrient intake. Food choices are linked to such other health-related behaviors as smoking, alcohol use, and prudent consumption of fat (4) and also may be linked to additional behaviors such as occupation, stress management, intransigence regarding current health guidelines, and acceptance of risk in general; these traits have not yet been addressed in our analyses. Mechanisms of cancer development must be explained in biological terms. With dietary exposure to cancer risk expressed as intake of a specific nutrient, the investigator possesses a measure with a clear potential biologic effect. Understanding what it is intended to capture is relatively straightforward. However, such an approach may fail to incorporate the metabolic consequences that dietary factors consumed concurrently exert on total risk (1-3,9,10). To develop a more comprehensive measure of dietary exposure to risk, we have shifted the focus of measurement from nutrients to foods, and by so doing we are less able to define the nutrient
272
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 6. Relative Risk of Colon Cancer Among Females Associatec With Dietary Patterns and Energy, Western New York Diet Study, 1975-1984° ORC
95% CI
0.60-0.89
0.67 1.37
0.54-0.83 1.12-1.68
1.00
0.83-1.21
0.95 1.24
0.78-1.15 1.02-1.52
Low-cost pattern Energy
1.15
0.94-1.42
1.08 1.20
0.86-1.34 0.97-1.47
Fruit pattern Energy
1.08
0.90-1.31
1.01 1.22
0.83-1.24 1.00-1.50
High-fat pattern Energy
0.99
0.82-1.20
0.86 1.33
0.69-1.07 1.06-1.67
Light pattern Energy
0.77
0.63-0.93
0.77 1.23
0.63-0.94 1.01-1.49
Whole-grain pattern Energy
0.88
0.73-1.07
0.80 1.32
0.65-0.99 1.07-1.63
Dietary Variables
OR*
95% CI
Energy intake
1.23
1.01-1.49
Salad pattern Energy
0.73
Healthful pattern Energy
a: Relative risk was assessed controlling for age and education. b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
mechanisms that the measure might incorporate. However, we are closer to behaviors that can be modified than the investigator working with nutrient indexes. We argue that this focus on the way foods are grouped naturally is of particular value for describing the association of risk and behavior. Our previous work (2-4) suggests that multiple risk factors, both dietary and nondietary, coalesce within dietary patterns that may be shaped by life-styles. We assume that additional correlates of dietary patterns contribute to cancer risk but have yet to be examined. Despite these limitations, results of this study indicate that the measures of dietary patterns do provide additional information concerning the role of diet in cancer risk beyond that obtained with measures of intake of single nutrients, such as energy, total fat, or dietary fiber. Extensive preliminary investigations were completed (1-4) before this study, in which the patterns themselves and their associations with parameters implicated in cancer risk were examined using the full set of control subjects only. We followed that strategy to improve our understanding of such multidimensional measures of diet without adding potential bias in study design by identifying differences in case and control subjects before undertaking this case-control study of risk. These investigations substantiate additional distinctions among the patterns that may contribute to disease risk and to the aggregate risk for the various dietary patterns (3,4). These differences include behaviors affecting ingestion of saturated fat; variety of fruits and vegetables routinely eaten; total alcohol intake as well as type of alcoholic beverages consumed; intake of preserved and barbecued foods; and smoking, both current and lifelong exposures. Among men, dietary patterns account for more variability in colon cancer risk than do intakes of single nutrients, specifically total fat. We have focused attention on those three patterns correlated most strongly with high intakes of fat that would be likely to be expressed
Vol. 18, No. 3
273
Downloaded by [] at 10:54 04 February 2015
Table 7. Relative Risk of Colon Cancer Among Females Associated With Dietary Patterns and Total Fat, Western New York Diet Study, ] 975-1984" Dietary Variables
OR*
95% CI
OR'
95% CI
Total fat intake
1.28
1.05-1.55
Salad pattern Total fat
0.73
0.60-0.89
0.68 1.39
0.56-0.84 1.13-1.71
Healthful pattern Total fat
1.00
0.83-1.21
0.96 1.29
0.79-1.16 1.05-1.57
Low-cost pattern Total fat
1.15
0.94-1.42
1.06 1.25
0.85-1.32 1.01-1.54
Fruit pattern Total fat
1.08
0.90-1.31
1.03 1.27
0.85-1.25 1.04-1.55
High-fat pattern Total fat
0.99
0.82-1.20
0.84 1.41
0.67-1.05 1.11-1.78
Light pattern Total fat
0.77
0.63-0.93
0.77 1.27
0.63-0.94 1.04-1.55
Whole-grain pattern Total fat
0.88
0.73-1.07
0.82 1.34
0.67-1.00 1.09-1.65
a: Relative risk was assessed controlling for age and education, b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
by subjects categorized into the highest quartile of fat intake. The high-fat pattern in men is distinctive in incorporating a large number of additional risk-elevating factors (3,4), including cigarette and alcohol use. Accounting for the sources of the protection from colon cancer for women that are inherent in the salad and light patterns in women is difficult; this study leaves unresolved the source of this protection. Dietary fiber was found to be unrelated to risk definition. The previous methodological studies with the dietary patterns (3,4) indicate that there are no differences related to scores on these patterns for percentage of calories derived from fat or for the types of fat used, except that women with diets consistent with these patterns were more likely to cook with vegetable oil and to add salad oil to foods (4). The salad and light patterns were not distinguishable from other patterns on the bases of the use of fruits and vegetables or preserved or charbroiled foods. Alcohol consumption was unrelated to the salad pattern and actually is related positively to the light pattern (3,4). Lifelong exposure to cigarette smoke was unrelated to either pattern. Scores were higher in exsmokers than in current smokers on the light pattern, and there was no significant difference on this variable for the salad pattern. We suspect that underlying health behaviors tend to be less related to the dietary patterns in women than they are in men; women may adopt health-promoting behaviors into their lives regardless of their eating styles. This conclusion is consistent with earlier researchers who have documented that women are more likely than their spouses to adjust their diets to accommodate food choices of household members (11). Therefore, eating styles and personal health values may be less strongly associated for women. Men and women appear to differ in the aggregation of cancer risks incorporated into thendiets; they also differ in the health behaviors and patterns of ingestion associated with diet. The result of these two sources of difference may result in distinctive exposures to risk for
274
Nutrition and Cancer 1992
Downloaded by [] at 10:54 04 February 2015
Table 8. Relative Risk of Colon Cancer Among Females Associated With Dietary Patterns and Dietary Fiber, Western New York Diet Study, 1975-1984" ORC
95% CI
0.60-0.89
0.67 1.19
0.53-O.84 0.95-1.48
1.00
0.83-1.21
1.02 0.96
0.83-1.26 0.78-1.19
Low-cost pattern Dietary fiber
1.15
0.94-1.42
1.18 0.93
0.95-1.46 0.77-1.13
Fruit pattern Dietary fiber
1.08
0.90-1.31
1.16 0.89
0.92-1.47 0.70-1.12
High-fat pattern Dietary fiber
0.99
0.82-1.20
1.00 0.97
0.83-1.22 0.80-1.18
Light pattern Dietary fiber
0.77
0.63-0.93
0.76 1.01
0.63-0.93 0.84-1.23
Whole-grain pattern Dietary fiber
0.88
0.73-1.07
0.88 1.02
0.71-1.08 0.83-1.25
Dietary Variables
OR*
95% CI
Dietary fiber intake
0.97
0.81-1.17
Salad pattern Dietary fiber
0.73
Healthful pattern Dietary fiber
a: Relative risk was assessed controlling for age and education, b: OR, bivariate logistic regression analysis: OR associated with an increase of 1 SD of range of exposure. c: OR, logistic regression, adjusted for the other factor.
the two genders, not only for colon cancer but for other cancer sites as well as for other diet-related chronic diseases. These dietary patterns are not presented as final measures; whether they will be useful in other settings among other populations requires further study. These factors may need to be refined through further methodological research. One feature that may limit their ability to explain cancer risk is the low percentage of variance in food use explained by them; the seven patterns together account for only 20.3% and 21.7% of the total variance in food intake among males and females, respectively. This problem undoubtedly is a function of the large number of foods (95) included in the factoring process. Although dietary patterns were able to account for a greater amount of variance in the Bogalusa Heart Study (7) when a similar analytic approach was used, the food list in that study was considerably shorter. Because knowledge of dietary patterns is still in a developmental phase, there was no exclusion criterion for foods available to us except for infrequency of food use. The problem is reminiscent of earlier efforts of Block and co-workers (8) to develop a food frequency instrument of feasible size. The ability of a food frequency instrument to explain variance in nutrient intakes was increased by grouping highly specific items into broader composite foods. Multidimensional measures of diet, such as dietary patterns, may improve risk estimation by incorporating dietary risk more comprehensively. Should the strongest risk factors be substances other than the nutrients selected in a specific study, such as folate or phytochemicals, multidimensional measures may still capture that risk. Their superior performance may also be attributable to greater sensitivity to related nondietary health behaviors associated with life-style characteristics. Willett and associates (12) recently demonstrated the value of using meats and the ratio of intakes of meat to chicken and fish in accounting for risk in a prospective study of colon cancer in women. We suggest that epidemiological studies could
Vol. 18, No. 3
275
profit from evaluating factors such as these in addition to a priori indexes of fat, energy, fiber, and vitamins A, C, and E. They deserve further methodological development; in addition, their stability over region and time should be addressed. Acknowledgments and Notes This study was funded by National Cancer Institute Grants CA-11535 and RO-CA-51720-01 from the National Institutes of Health (Bethesda, MD). Address reprint requests to Dr. D. Elizabeth Randall, Nutrition Program, 301 Parker Hall, S.U.N.Y. at Buffalo, Buffalo, NY 14214. Submitted 26 May 1992; accepted in final form 5 August 1992.
Downloaded by [] at 10:54 04 February 2015
References 1. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "Frequency of Food Use Data and the Multidimensionality of Diet." J Am Diet Assoc 89, 1070-1075, 1989. 2. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "Patterns in Food Use and Their Associations With Nutrient Intakes." Am J Clin Nutr 52, 739-745, 1990. 3. Randall, E, Marshall, JR, Brasure, J, and Graham, S: "Patterns in Food Use and Compliance With NCI Dietary Guidelines." Nutr Cancer 15, 141-156, 1991. 4. Randall, E, Marshall, JR, Graham, S, and Brasure, J: "High Risk Health Behaviors Associated With Various Dietary Patterns." Nutr Cancer 16, 135-151, 1991. 5. Graham, S, Marshall, J, Haughey, B, Mittelman, A, Swanson, M, et al.: "Dietary Epidemiology of Cancer of the Colon in Western New York." Am J Epidemiol 128, 490-503, 1988. 6. Randall, E, Nichaman, MZ, and Contant, CF, Jr.: "Diet Diversity and Nutrient Intake." J Am Diet Assoc 85, 830-836, 1985. 7. Nicklas, TA, Webber, LS, Thompson, B, and Berenson, GS: "A Multivariate Model for Assessing Eating Patterns and Their Relations to Cardiovascular Risk Factors: the Bogalusa Heart Study." Am J Clin Nutr 49, 320-327, 1989. 8. Block, G, Dresser, CM, Hartman, AM, and Carroll, MD: "Nutrient Sources in the American Diet: Quantitative Data From the NHANES II Survey. II. Macronutrients and Fats." Am J Epidemiol 122, 27-40, 1985. 9. Modan, B, Barell, V, Lubin, F, Modan, M, Greenberg, RA, et al.: "Low-Fiber Intake as an Etiologic Factor in Cancer of the Colon." JNCI 55, 15-18, 1975. 10. Freudenheim, JL, Graham, S, Marshall, JR, Haughey, BP, and Wilkinson, G.: "A Case-Control Study of Diet and Rectal Cancer in Western New York." Am J Epidemiol 131, 612-624, 1990. 11. Davis, MA, Randall, E, Forthofer, RN, Lee, ES, and Margen, S: "Living Arrangements and Dietary Patterns of Older Adults in the United States." J Gerontol 40, 434-442, 1985. 12. Willett, WC, Stampfer, MJ, Colditz, GA, Rosner, BA, and Speizer, FE.: "Relation of Meat, Fat, and Fiber Intake to the Risk of Colon Cancer in a Prospective Study Among Women." N Engl J Med 323, 1664-1672, 1990.
276
Nutrition and Cancer 1992
