Int. J. Cancer: 51,858-861 (1992) 0 1992 Wiley-Liss, Inc.
Publication of the International Union Against Cancer Publication de I'Union Internationale Contre le Cancer
COLON CANCER IN ARGENTINA. 11: RISK FROM FIBRE, FAT AND NUTRIENTS J.M. ISCOVICH',7,9,K.A. L ' h B E 2 . ' , R. cASTELLET03, A. CALZONA", A. BERNED04,N.A. cHOPITA4, A.C. JMELNITZSKY", ~ , ~ G.R. H O W E ~ J. K A L D O R and 'Depaument of Epidemiology Studies, Ministly of Health, La Plata, Argentina; 21ntemationalAgency for Research on Cancer, Lyon, France; 3Departmentof Pathology, La Plata National University, La Plata; "Unit of Gastroenterology, San Martin Hospital, La Plata, Argentina; Sunitof HIV Epidemiology and Clinical Research, University of South Wales, Sydney, Australia; and 6NationalCancer Institute of Canada, Epidemiology Unit, Toronto, Ontario, Canada. A case-control study has been conducted to investigate the relationship between total energy intake, fibre and nutrients and colon cancer in Argentina. Cases are I I0 newly diagnosed patients from I 0 hospitals between I985 and 1986. Two neighbourhood controls per case were matched on age, sex and place of residence. The intake of calories, fibre and nutrients was estimated from the information collected on food consumption during the 5-year period up to 6 months prior to interview of subjects, based on a pilot study and standard portion sizes in Argentina. In conditional regression models, dietary fibre is highly protective (odds ratio (OR) = 0.07 per 19.02 g/day; 95% confidence interval (CI): 0.02 to 0.25) and total energy intake increases risk (RR = 1.82 per 1000 kcal/day; 95% CI: 1.20 to 2.79, each with adjustment for the other. Analysis of caloric components (fat, protein and carbohydrates) revealsthat carbohydrates are the most important factor driving the total energy effect. Other nutrients make little apparent contribution to risk.
o 1992 Wiley-Liss,Inc. One hypothesis concerning the effect of dietary nutrients in cancer is that fibre decreases the risk of colon cancer. Another hypothesis is that fat intake, particularly animal fat, increases the risk. Case-control studies have yielded inconsistent results in this regard, while a recent large prospective study of women reported increased risk associated with intake of total fat, due to animal fat (Willett et al., 1990). This study also reported that dietary fibre was associated with decreased risk, but not to a statistically significant degree. Total energy intake and nutrient values were estimated from the frequency of consumption of food items, as reported in Part I of this study. Part I1 presents the risk of colon cancer in this population according to intake of nutritional factors. METHODS
The study subjects and methods of data collection are the same as in Part I of this report. Using the frequency data for the dietary items reported in Part I, the intake of calories (total energy) and a number of specific nutrients was estimated. Standard portion sizes were estimated for each of the food items, using data collected from a pilot study carried out in Rosario, Argentina, between 1980 and 1982. In this pilot study, a full diet history, including quantification of portion sizes using food models, was collected for 20 breast-cancer cases and 20 controls. The dietary methods used have been described by Morgan et al. (1983). Using the estimated standard portion sizes together with the frequency data, the weight of each food item consumed daily was estimated for the subjects in the present study. These weights were then converted to estimated daily nutrient intake, using tables of nutrient intake per 100 g for Argentinean foods (Instituto Nacional de la Nutricion, 1945). Conditional logistic regression, maintaining the matching of subjects, was used for the analysis. Two types of statistical models were used to examine the effects of dietary factors. In the first, data were divided into quartiles based on the distribution of intake by all study subjects. Chi-square tests for
association were conducted with 3 degrees of freedom (i.e., number of quartiles minus one), and z-tests for trend were performed (one degree of freedom). The appendix shows the cutpoints corresponding to the bounds for the quartiles. In the second approach, each dietary factor was represented by a continuous variable, with ORs expressed in scaled units corresponding to the difference in the mean of the highest and lowest quartiles of intake. All models included terms to represent the effect of total energy intake (Willett and Stampfer, 1986), and in view of the high correlation between caloric components, the method proposed by Howe (1989) was also used, where estimates for fat, protein and carbohydrates are adjusted for other sources of calories. Models in which the energy intake was represented by its components (fat, protein and carbohydrates) gave results essentially identical to those presented. All p values reported are 2-sided. Part I of this report describes characteristics of cases and controls and the odds ratios for the frequency of consumption of specific foods. RESULTS
Table I shows the odds ratio and 95% confidence interval (CI) for quartiles of total energy intake, total fat, protein, carbohydrates, dietary fibre, beta-carotene, retinol and vitamin C. The p value for the test for trend (one degree of freedom) included in Table I is that based on treating the 4 categories as a single variable with scores 1,2,3 and 4. There are 2 dietary factors associated with risk for colon cancer: total energy intake, and dietary fibre. Therefore, the odds ratios presented for one are adjusted for the other. Total energy intake shows a statistically significant increasing trend in risk of colon cancer as intake increases (p = 0.009). Cases and controls have approximately the same mean level of total energy intake, at 2649 kcaldday and 2612 kcals/day respectively (t-test p value > 0.60). In contrast, there is a monotonically decreasing dose-response relationship for the intake of dietary fibre (p < 0.001). Those in the highest quartile of intake (mean, 25.5 g/day) have less than one tenth the risk of those in the lowest quartile (mean, 6.5 g/day). Odds ratios for the 3 macronutrients that contribute to total energy intake are also adjusted for fibre. These macronutrients, i.e., fat, protein and carbohydrates, do not exhibit strong associations with risk. Odds ratios for each of the 3 macronutri-
'Current address: Israel Cancer Registry, Ministry of Health, 107 Derech Hebron, 93480 Jerusalem, Israel. RCurrent address: Preventive Medicine and Biostatistics, Faculty of Medicine, University of Toronto, Ontario, Canada. 'To whom correspondence and reprint requests should be addressed.
Received: December 12,1991 and in revised form March 13,1992.
859
COLON CANCER IN ARGENTINA. 11: FIBRE, FAT AND NUTRIENTS
TABLE 1- MATCHED ORs FOR COLON CANCER ACCORDING TO QUARTILE OF TOTAL ENERGY INTAKE, ITS COMPONENTS, FIBRE AND NUTRIENTS
1
2
3
4
Test for association (p value)
1.o
1.79 (0.84-3.81) 1.10 (0.54-2.26) 0.94 (0.46-1.90) 1.47 (0.65-3.32) 0.87 (0.41-1.82) 2.26* (1.06-4.83) 0.77 (0.39-1.53) 0.83 (0.38-1.18)
1.53 (0.62-3.76) 0.74 (0.32-1.17) 0.63 (0.26-1.49) 1.25 (0.51-3.10) 0.23* (0.09-0.61) 2.24 (0.78-6.46) 0.65 (0.26-1.63) 0.65 (0.23-1.18)
4.76* (1.32-14.02) 1.18 (0.42-3.33) 1.03 (0.33-3.21) 4.46* (1.45-13.71) 0.07* (0.02-0.26) 1.26 (0.30-5.35) 1.28 (0.60-2.73) 0.49 (1.12-1.97)
11.144 (0.01) 4.71 (0.19) 2.52 (0.47) 7.04 (0.07) 27.19 ( < 0.0001) 6.96 (0.07) 3.4 (0.33) 1.15 (0.77)
Odds ratio by quartile of consumption and 95% CI
Variable
Total energy intake’ (kcal /day) Total fat’.2 (&?/day) Protein’J (dday)
1.0 1.0 1.0
Test for trend (p value)
6.93 (0.009) 0.41 (0.52) 0.36 (0.55) 2.62 0.11 22.08 ( < 0.001) 0.53 (0.47) 0.58 (0.45) 1.12 (0.29)
(g/day) Dietary fibre3 1.0 (81day) Beta- carotene'^^ 1.o (IU/day) Retin~ll.~ 1.0 (IU/day) Vitamin 0 3 1.0 (mdday) ‘Adjusted for fibre at 19.02 g per day.JAdjusted for other sources of energy intake.-3Adjusted for total energy intake of 1000kcals per day by regression analysis.-*p < 0.05 (95% CI does not include 1.0). ents are inconsistent across quartiles, and tests for association and trend are not statistically significant. The 3 remaining nutrients, which do not contribute to total caloric intake, namely beta-carotene, retinol and vitamin C, are adjusted both for total energy intake (at 1000 kcals/day) and for dietary fibre (at 19.02 g/day). These nutrients do not exhibit strong inverse trends in risk as intake increases (lowest p value is for vitamin C at 0.29). Odds ratios across quartiles are inconsistent both for beta-carotene and for retinol, and confidence intervals are wide, indicating low power in this study to detect even strong associations with these nutrients. It can be stated, however, that any effect of these variables is certainly weaker than that of dietary fibre. The strong inverse association seen with fibre intake was unaffected by adjustment for height, weight, body mass index or education. There was also no evidence of confounding by these latter factors for any of the other dietary components. Interaction effects between total energy intake, fibre and nutrients were examined in a number of models, and there were no statistically significant interactions. The 2-sided p values for the trends in odds ratios with increasing total caloric intake and dietary fibre, derived by treating these variables as continuous, are both highly significant statistically. On the basis of this model, the OR for total energy intake per 1000 kcals unadjusted for any other factor is 1.07 (95% CI: 0.78 to 1.48), and that for fibre per 19.02 grams/day, unadjusted for other factors, is 0.20 (95% CI: 0.08 to 0.50). When adjusted for each other, these ORs are 1.82 (95% CI: 1.20 to 2.77) and 0.07 (95% CI: 0.02 to 0.25) respectively, as seen in Table 11. Substantial confounding is thus evident and the ORs presented are adjusted for each other. They represent the difference in total energy intake and dietary fibre between the means for the highest and lowest quartiles, each adjusted for the effect of the other. Thus, the ORs represent smoothed estimates of the approximate risks between extreme quartiles. Since total energy intake is associated with risk of colon cancer, it is necessary to differentiate between the individual effects of the components of calories and an effect of total energy intake itself. A further analysis is shown in Table 111, where the individual effects of the components of calories, i.e., total fat, protein and carbohydrates, arc sorted out. The methods both of Howe (1989) and of Willett and Stampfer (1986) are shown in Table 111. Total fat, protein and carbohydrates are treated as continuous variables, and in the first method (Howe, 1989) each nutrient, together with calories from other sources, is included in the regression model, each
TABLE I1 MATCHED ORs FOR COLON CANCER FOR TOTAL ENERGY INTAKE AND DIETARY FIBRE, EACH ADJUSTED FOR THE OTHER ~
Dietarv factor
OR
95% CI
Total energy intake per 1000 kcals/day Dietary fibre (per 19.02 g/day)
1.82
1.20 to 2.77
0.07
0.02 to 0.25
with adjustment for dietary fibre at 19.02 g/day. The same units are used for the 3 components of total energy intake, i e . , 1000 kcals/day, in order to make the odds ratios directly comparable. This approach follows the conventional method of adjusting for confounders, and the corresponding risk estimate is that for the main effect of the nutrient, independent of other sources of calories. In the second method (Willett and Stampfer, 1986), represented by the last column of Table 111, the “difference” in effect between the nutrient and other sources of calories is estimated. Results are shown in terms of the log OR and its standard error, since these provide the most direct comparison. The units of nutrient intake used are arbitrary, since they do not affect the statistical significance of any observed association. The data in the first column of Table I11 clearly show that the only component of total energy having a positive association with colon-cancer risk is carbohydrates. Therefore, carbohydrates are driving the harmful total energy intake effect, but this association is not completely independent of the other sources of calories, as seen by thep value of 0.58 given in the second column of Table 111. The difference in effect between carbohydrates and other sources of calories is not statistically significant, as shown by thep value for the difference in the last column of Table 111. The analysis for total fat and protein shows no evidence for any association with risk, in accordance with the analysis in Table I. DISCUSSION
The present study provides some support for the existence of a positive association between total caloric intake and increased risk of colon cancer, while there is strong evidence of monotonically decreasing risk with increasing intake of dietary fibre. Part I of this report showed that increasing vegetable consumption was associated with decreasing risk, and the association with dietary fibre reflects the higher consumption of vegetables by control subjects. Another marker of fruit and vegetable intake, namely vitamin C, also exhibits an inverse association, but the effect is weaker and may be the result of
860
ISCOVICH ETAL.
TABLE 111- LOG OR” FOR COLON CANCER PER 1000KCALSIDAY FOR TOTAL FAT, PROTEIN AND CARBOHYDRATE INTAKE Other calories
Nutrient Dietary factor
Total fat Protein Carbohydrates
h e OR
SE
D
h e OR
SE
I)
-0.26 -0.50 1.13
0.52 1.77 0.43
0.61 0.78 0.008
1.22 0.78 0.19
0.41 0.36 0.35
0.003 0.03 0.58
Difference
SE
P
1.48 1.28 0.94
0.82 2.04 0.64
0.07 0.53 0.14
*Estimatesfrom a continuous variable model.-“Other calories” model: E(D) = d + d (T - N) + d N (Howe et ab, 1986;Howe, 1989) and “Difference”model: E(D = c + c T + c N (Willett and Stampfer, 1986), where T = total calories, N = nutrient which contributes to calories. and c and d are coe cients.
A
measurement error. Fibre intake and vitamin C are highly correlated (r2 = 0.89), and the effect of imperfect measurement of dietary fibre is that adjustment for the confounding effect of fibre on vitamin C will be incomplete, which could explain the weak association seen for this vitamin. Although in Part I it was shown that increased risk was associated with food containing high levels of fat, such as red meat, eggs and cheese, the findings in Part I1 suggest that the effect of total caloric intake is not due to total fat, but perhaps to carbohydrates. For policy recommendations, it is helpful to know that reduction in total calories may reduce risk, while within total intake of food items a proportional increase in foods containing fibre, particularly vegetables, may contribute to decreased risk for colon cancer. Some methodological problems in case-control studies of diet have been well described. Downward bias in the OR results from the imprecise nature of questionnaire-based dietary information (Tzonou et al., 1986), relative to the small variation in food consumption habits within populations (Wynder, 1975; Morgan et al., 1983). The questionnaire used for this study was only semi-quantitative, and was not intended to provide detailed information on micro- or macro-nutrients, such as the type of fat. The availability of an earlier dietary survey nevertheless permitted a rough estimation of nutrient levels. Biases which inflate ORs may be produced through differential recall between cases and controls (Marshall et al., 1981; Block, 1982), or through the influence of pre-clinical disease on diet (Byers et al., 1987), particularly if this state is long-standing (Lillienfeld et al., 1979). To avoid this latter problem, the questionnaire on food consumption asked about the 5-year period up to 6 months before the interview. The positive association between total energy intake and risk of colon cancer found in this study is in contrast to the results of a prospective study among 88,751 women, including 150 cases of colon cancer, where total fat, mainly from animal sources, was associated with increased risk, while total energy intake was not (Willett et al., 1990). While total energy intake has been associated with significantly increased risk in various studies (Jain et al., 1980; Bristol et al., 1985; Potter and McMichael, 1986; Kune and Kune, 1987; Lyon et al., 1987; Graham et al., 1988; Slattery et al., 1988; Whittemore et al., 1989), it has not been in others (Berta et al., 1985; MacquartMoulin et al., 1986; Tuyns et al., 1987). Analysis is complicated,
however, by the fact that components of total energy intake are correlated, and it is necessary to sort out whether positive associations with total energy intake are due to total energy intake itself or one of its components, independent of the total intake. As shown in this study, methods proposed by Howe (1989) and by Willett and Stampfer (1986) provide means by which this can be accomplished. Several case-control studies have reported inverse associations with fibre and the risk of colon cancer (Modan et al., 1975; Dales et al., 1979; Willett and Stampfer, 1986; Lyon et al., 1987; Graham et al., 1988; Slattery et al., 1989; Whittemore et al., 1990), but this has not been reported in other studies (Jain et al., 1980; Berta et al., 1985; Bristol et al., 1985; MacquartMoulin et al., 1986; Potter and McMichael, 1986). When sources of fibre have been examined, it has been shown that fibre from fruits and vegetables, and not cereals, is consistently associated with decreased risk of colon cancer (Willett et al., 1990). Several hypotheses attempt to explain the mechanism by which dietary fibre may protect against colon cancer. Potential mechanisms include the physical capacity of fibre to increase stool bulk and transit time, the binding by fibre of bile acids (which may promote cancer), and the possibility that fibre acts as a substrate for bacterial fermentation, which can increase stool bulk and have some anti-carcinogenic effects. Fibre may, however, be a marker for anti-carcinogens found in plants, and it is therefore important to emphasize that the primary source of fibre in this study was plants and not supplementation. Recommendations based on the findings of the present study should therefore be made on the basis of increasing the consumption of foods rich in fibre, within a diet limited to a certain level of total energy intake.
ACKNOWLEDGEMENTS
We thank the clinics, hospitals and pathologists of La Plata city. We also thank Ms. R. Iscovich and Ms. F. Lubin for their valuable comments and advice. The analysis was undertaken in part while Dr. L‘Abbe was the recipient of a post-doctoral fellowship from the International Agency for Research on Cancer, and in part through her National Health Research Scholar Award (Canada).
REFERENCES
BERTA,J.L., COSTE,T., RAUTUREAU, J., GUILLOUD BATAILLE,M. and PEQUIGNOT, G., Diet and rectocolonic cancers: results of a case study. Gastroenterol.din. Biol., 9,348-353 (1985). BLOCK,G., Review of validations of dietary assessment methods. Amer. J. Epidemiol., 115,492-50s (1982). BRISTOL,J.B., EMMENT,P.M., HEATON,K.W. and WILLIAMSON, R.C.N., Sugar and fat, and the risk of colorectal cancer. Brit. med. J., 291,1467-1470 (1985). BYERS,T.E., MARSHALL, J.R., ANTHONY, E., FIEDLER, F. and ZIELEZNY, M., The reliability of dietary history from the distant past. Amer. J. Epidemiol., 107,510-521 (1987).
DALES, L.C., FREIDMAN,G.D., URY, H.K., GROSSMAN,S. and WILLIAMS, S.R., A case-controlstudy of relationships of diet and other traits to colorectal cancer in American blacks.Amer.J. Epiderniol., 109,
132-144 (1979). GRAHAM, S., MARSHALL,J., HAUGHEY, B., MITTELMAN, A., SWANSON, M., ZIELEZNEY, M., BYERS,T., WILKINSON, G. and WEST,D., Dietary epidemiology of cancer of the colon in western New York. Amer. J. Epidemiol., 128,490-SO3 (1988). HOWE,G.R., Letter to the editor re: “Total energy intake: implications for epidemiologic analyses” by Willett and Stampfer. Amer. J. Epidemiol., 129,1314-1315 (1989).
861
COLON CANCER IN ARGENTINA. 11: FIBRE, FAT AND NUTRIENTS
comparison of dietary methods in epidemiologic studies. Amer. J. Epidemivl., 107,488-498 (1983). POTTER,J.D. and MCMICHAEL, A.J., Diet and cancer of the colon and rectum: a case-control study. J. nut. Cancer Inst., 76,557-569 (1986). SLATTERY, M.L., SORENSON, A.W., MUHONEY, A.W., FRENCH,T.K., D. and STREET,J.C., Diet and colon cancer: assessment KRITCHEVSKY, of risk factor by fiber type and food source. J. nat Cancer Inst., 80, 1474-1480 (1988). [Erratum,J. nat Cancer Insf.,81,1042 (1989)l. TUYNS,A.J., HAELTERMAN, M. and KAAS, R., Colorectal cancer and the intake of nutrients: oligosaccharides are a risk factor, fats are not: a case-control study in Belgium. Nufr.Cancer, 10,181-196 (1987). TZONOU,A,, KALDOR,J.M., DAY, N.E. and TRICHOPOULOS, D., Misclassification in case-control studies with two dichotomous risk factors. Rev. Epidimivl. Sanfe‘Publ.,34,lO-17 (1986). WHITTEMORE, A.S., WU-WILIJAMS,A.H., LEE, M., SHU, Z., GALLAGHER, R.P., DENG-Ao,J., LUN,Z., XIANGSUI, W., KUN,C., JUNG, L., YAO,K.J., PAFFENBARGER, JR., R.S. and D., THE,C.Z., CHENGDE, HENDERSON, B.E., Diet, physical activity, and colorectal cancer among Chinese in North America and China. J. nut. Cancer Insf., 82,915-923 (1989). WILLETT, W.C. and STAMPFER, M.J., Total energy intake: implications for epidemiologic analyses. Amer. J. Epidemivl., 124,307-309 (1986). WILLETT,W.C., STAMPFER, M.J., COLDITZ,G.A., ROSNER, B.A. and F.E., A prospective study of diet and colon cancer in women. SPEIZER, New Engl. J. Med., 323,1664-1672 (1990). WYNDER, E.L., The epidemiology of large-bowel cancer. Cancer Res., 35,3388-3394 (1975).
HOWE,G.R., MILLER,A.B. and JAIN,M., Letter to the editor re: “Total energy intake: implications for epidemiologic analyses.” Amer. J. Epidemivl., 124,157-159 (1986). INSTITUTO NACIONAL DE LA NUTRICION, Tablas de las cvmpvsicivn quimica de lvs alimenfvs.Buenos Aires (1945). JAIN,M., COOK,G.M., DAVIS,F.G., GRACE,M.G., HOWE,G.R. and MILLER,A.B., A case-control study of diet and colo-rectal cancer. Int. J. Cancer, 26,757-768 (1980). KUNE,G.A. and KUNE,S., The nutritional causes of colorectal cancer: an introduction to the Melbourne Study. Nutr. Cancer, 9,1-4 (1987). LILLIENFELD, D.E., GARGAGLIANO, C.F. and LILLIENFELD, A.M., Gastric cancer (letter). J. nut. Cancer Insf., 57,9 (1979). LYON,J.L., MAHONEY, A.W., WEST,D.W., GARDNER, J.W., SMITH, K.R., SORENSON, A.W. and STANISH, W., Energy intake: its relationship to colon-cancer risk.J. nut. Cancer Insf., 78,853-861 (1987). MACQUART-MOULIN, G., RIBOLI, E., CORNEA,J., CHARNAY, B., BERTHEZ~NE, P. and DAY,N., Case-control study on colorectal cancer and diet in Marseilles. Znf.J. Cancer, 38,183-191 (1986). MARSHALL, J.R., PRIORE,R., GRAHAM,S . and BRASURE, J., On the distortion of risk estimates in multiple exposure level case-control studies. Amer. J. Epidemivl., 113,464-473 (1981). MODAN,B., BAREIL,V., LUBIN,F., MODAN,M., GREENBERG, R.A. and GRAHAM,S., Low fiber intake as an etiology factor in cancer of the co1on.J. nut. Cancer Inst., 55,15-18 (1975). MORGAN,R.W., JAIN,M., MILLER,A.B., CHOI,N.W., MATHEWS, V., MUNAN, L., BURCH,J.D., FEATHER,J., HOWE,G.R. and KELLY,A., A
APPENDIX TABLE I - CUTPOINTS FOR QUARTILE AND DAILY INTAKE FOR CASES AND CONTROLS
Total energy (kcaliday) Total fat (g/day) Protein (giday) Carbohydrates (g/day) Dietary fibre (g/day) B-carotene (IU/day) Vitamin C (mg/day)
1
2
3
1915.09 95.04 89.15 167.41 8.87 44 18.21 73.19
2452.52 123.31 119.10 233.50 14.31 8021.68 126.02
3195.65 149.20 140.50 309.70 21.71 13044.8 184.65
Mean intake
Cases
Controls
2649 124.40 115.50 265.40 14.08 8194 118.10
2612 126.00 116.80 253.20 16.21 9085 134.50
Publication of the International Union Against Cancer Publication de I'Union Internationale Contre le Cancer
COLON CANCER IN ARGENTINA. 11: RISK FROM FIBRE, FAT AND NUTRIENTS J.M. ISCOVICH',7,9,K.A. L ' h B E 2 . ' , R. cASTELLET03, A. CALZONA", A. BERNED04,N.A. cHOPITA4, A.C. JMELNITZSKY", ~ , ~ G.R. H O W E ~ J. K A L D O R and 'Depaument of Epidemiology Studies, Ministly of Health, La Plata, Argentina; 21ntemationalAgency for Research on Cancer, Lyon, France; 3Departmentof Pathology, La Plata National University, La Plata; "Unit of Gastroenterology, San Martin Hospital, La Plata, Argentina; Sunitof HIV Epidemiology and Clinical Research, University of South Wales, Sydney, Australia; and 6NationalCancer Institute of Canada, Epidemiology Unit, Toronto, Ontario, Canada. A case-control study has been conducted to investigate the relationship between total energy intake, fibre and nutrients and colon cancer in Argentina. Cases are I I0 newly diagnosed patients from I 0 hospitals between I985 and 1986. Two neighbourhood controls per case were matched on age, sex and place of residence. The intake of calories, fibre and nutrients was estimated from the information collected on food consumption during the 5-year period up to 6 months prior to interview of subjects, based on a pilot study and standard portion sizes in Argentina. In conditional regression models, dietary fibre is highly protective (odds ratio (OR) = 0.07 per 19.02 g/day; 95% confidence interval (CI): 0.02 to 0.25) and total energy intake increases risk (RR = 1.82 per 1000 kcal/day; 95% CI: 1.20 to 2.79, each with adjustment for the other. Analysis of caloric components (fat, protein and carbohydrates) revealsthat carbohydrates are the most important factor driving the total energy effect. Other nutrients make little apparent contribution to risk.
o 1992 Wiley-Liss,Inc. One hypothesis concerning the effect of dietary nutrients in cancer is that fibre decreases the risk of colon cancer. Another hypothesis is that fat intake, particularly animal fat, increases the risk. Case-control studies have yielded inconsistent results in this regard, while a recent large prospective study of women reported increased risk associated with intake of total fat, due to animal fat (Willett et al., 1990). This study also reported that dietary fibre was associated with decreased risk, but not to a statistically significant degree. Total energy intake and nutrient values were estimated from the frequency of consumption of food items, as reported in Part I of this study. Part I1 presents the risk of colon cancer in this population according to intake of nutritional factors. METHODS
The study subjects and methods of data collection are the same as in Part I of this report. Using the frequency data for the dietary items reported in Part I, the intake of calories (total energy) and a number of specific nutrients was estimated. Standard portion sizes were estimated for each of the food items, using data collected from a pilot study carried out in Rosario, Argentina, between 1980 and 1982. In this pilot study, a full diet history, including quantification of portion sizes using food models, was collected for 20 breast-cancer cases and 20 controls. The dietary methods used have been described by Morgan et al. (1983). Using the estimated standard portion sizes together with the frequency data, the weight of each food item consumed daily was estimated for the subjects in the present study. These weights were then converted to estimated daily nutrient intake, using tables of nutrient intake per 100 g for Argentinean foods (Instituto Nacional de la Nutricion, 1945). Conditional logistic regression, maintaining the matching of subjects, was used for the analysis. Two types of statistical models were used to examine the effects of dietary factors. In the first, data were divided into quartiles based on the distribution of intake by all study subjects. Chi-square tests for
association were conducted with 3 degrees of freedom (i.e., number of quartiles minus one), and z-tests for trend were performed (one degree of freedom). The appendix shows the cutpoints corresponding to the bounds for the quartiles. In the second approach, each dietary factor was represented by a continuous variable, with ORs expressed in scaled units corresponding to the difference in the mean of the highest and lowest quartiles of intake. All models included terms to represent the effect of total energy intake (Willett and Stampfer, 1986), and in view of the high correlation between caloric components, the method proposed by Howe (1989) was also used, where estimates for fat, protein and carbohydrates are adjusted for other sources of calories. Models in which the energy intake was represented by its components (fat, protein and carbohydrates) gave results essentially identical to those presented. All p values reported are 2-sided. Part I of this report describes characteristics of cases and controls and the odds ratios for the frequency of consumption of specific foods. RESULTS
Table I shows the odds ratio and 95% confidence interval (CI) for quartiles of total energy intake, total fat, protein, carbohydrates, dietary fibre, beta-carotene, retinol and vitamin C. The p value for the test for trend (one degree of freedom) included in Table I is that based on treating the 4 categories as a single variable with scores 1,2,3 and 4. There are 2 dietary factors associated with risk for colon cancer: total energy intake, and dietary fibre. Therefore, the odds ratios presented for one are adjusted for the other. Total energy intake shows a statistically significant increasing trend in risk of colon cancer as intake increases (p = 0.009). Cases and controls have approximately the same mean level of total energy intake, at 2649 kcaldday and 2612 kcals/day respectively (t-test p value > 0.60). In contrast, there is a monotonically decreasing dose-response relationship for the intake of dietary fibre (p < 0.001). Those in the highest quartile of intake (mean, 25.5 g/day) have less than one tenth the risk of those in the lowest quartile (mean, 6.5 g/day). Odds ratios for the 3 macronutrients that contribute to total energy intake are also adjusted for fibre. These macronutrients, i.e., fat, protein and carbohydrates, do not exhibit strong associations with risk. Odds ratios for each of the 3 macronutri-
'Current address: Israel Cancer Registry, Ministry of Health, 107 Derech Hebron, 93480 Jerusalem, Israel. RCurrent address: Preventive Medicine and Biostatistics, Faculty of Medicine, University of Toronto, Ontario, Canada. 'To whom correspondence and reprint requests should be addressed.
Received: December 12,1991 and in revised form March 13,1992.
859
COLON CANCER IN ARGENTINA. 11: FIBRE, FAT AND NUTRIENTS
TABLE 1- MATCHED ORs FOR COLON CANCER ACCORDING TO QUARTILE OF TOTAL ENERGY INTAKE, ITS COMPONENTS, FIBRE AND NUTRIENTS
1
2
3
4
Test for association (p value)
1.o
1.79 (0.84-3.81) 1.10 (0.54-2.26) 0.94 (0.46-1.90) 1.47 (0.65-3.32) 0.87 (0.41-1.82) 2.26* (1.06-4.83) 0.77 (0.39-1.53) 0.83 (0.38-1.18)
1.53 (0.62-3.76) 0.74 (0.32-1.17) 0.63 (0.26-1.49) 1.25 (0.51-3.10) 0.23* (0.09-0.61) 2.24 (0.78-6.46) 0.65 (0.26-1.63) 0.65 (0.23-1.18)
4.76* (1.32-14.02) 1.18 (0.42-3.33) 1.03 (0.33-3.21) 4.46* (1.45-13.71) 0.07* (0.02-0.26) 1.26 (0.30-5.35) 1.28 (0.60-2.73) 0.49 (1.12-1.97)
11.144 (0.01) 4.71 (0.19) 2.52 (0.47) 7.04 (0.07) 27.19 ( < 0.0001) 6.96 (0.07) 3.4 (0.33) 1.15 (0.77)
Odds ratio by quartile of consumption and 95% CI
Variable
Total energy intake’ (kcal /day) Total fat’.2 (&?/day) Protein’J (dday)
1.0 1.0 1.0
Test for trend (p value)
6.93 (0.009) 0.41 (0.52) 0.36 (0.55) 2.62 0.11 22.08 ( < 0.001) 0.53 (0.47) 0.58 (0.45) 1.12 (0.29)
(g/day) Dietary fibre3 1.0 (81day) Beta- carotene'^^ 1.o (IU/day) Retin~ll.~ 1.0 (IU/day) Vitamin 0 3 1.0 (mdday) ‘Adjusted for fibre at 19.02 g per day.JAdjusted for other sources of energy intake.-3Adjusted for total energy intake of 1000kcals per day by regression analysis.-*p < 0.05 (95% CI does not include 1.0). ents are inconsistent across quartiles, and tests for association and trend are not statistically significant. The 3 remaining nutrients, which do not contribute to total caloric intake, namely beta-carotene, retinol and vitamin C, are adjusted both for total energy intake (at 1000 kcals/day) and for dietary fibre (at 19.02 g/day). These nutrients do not exhibit strong inverse trends in risk as intake increases (lowest p value is for vitamin C at 0.29). Odds ratios across quartiles are inconsistent both for beta-carotene and for retinol, and confidence intervals are wide, indicating low power in this study to detect even strong associations with these nutrients. It can be stated, however, that any effect of these variables is certainly weaker than that of dietary fibre. The strong inverse association seen with fibre intake was unaffected by adjustment for height, weight, body mass index or education. There was also no evidence of confounding by these latter factors for any of the other dietary components. Interaction effects between total energy intake, fibre and nutrients were examined in a number of models, and there were no statistically significant interactions. The 2-sided p values for the trends in odds ratios with increasing total caloric intake and dietary fibre, derived by treating these variables as continuous, are both highly significant statistically. On the basis of this model, the OR for total energy intake per 1000 kcals unadjusted for any other factor is 1.07 (95% CI: 0.78 to 1.48), and that for fibre per 19.02 grams/day, unadjusted for other factors, is 0.20 (95% CI: 0.08 to 0.50). When adjusted for each other, these ORs are 1.82 (95% CI: 1.20 to 2.77) and 0.07 (95% CI: 0.02 to 0.25) respectively, as seen in Table 11. Substantial confounding is thus evident and the ORs presented are adjusted for each other. They represent the difference in total energy intake and dietary fibre between the means for the highest and lowest quartiles, each adjusted for the effect of the other. Thus, the ORs represent smoothed estimates of the approximate risks between extreme quartiles. Since total energy intake is associated with risk of colon cancer, it is necessary to differentiate between the individual effects of the components of calories and an effect of total energy intake itself. A further analysis is shown in Table 111, where the individual effects of the components of calories, i.e., total fat, protein and carbohydrates, arc sorted out. The methods both of Howe (1989) and of Willett and Stampfer (1986) are shown in Table 111. Total fat, protein and carbohydrates are treated as continuous variables, and in the first method (Howe, 1989) each nutrient, together with calories from other sources, is included in the regression model, each
TABLE I1 MATCHED ORs FOR COLON CANCER FOR TOTAL ENERGY INTAKE AND DIETARY FIBRE, EACH ADJUSTED FOR THE OTHER ~
Dietarv factor
OR
95% CI
Total energy intake per 1000 kcals/day Dietary fibre (per 19.02 g/day)
1.82
1.20 to 2.77
0.07
0.02 to 0.25
with adjustment for dietary fibre at 19.02 g/day. The same units are used for the 3 components of total energy intake, i e . , 1000 kcals/day, in order to make the odds ratios directly comparable. This approach follows the conventional method of adjusting for confounders, and the corresponding risk estimate is that for the main effect of the nutrient, independent of other sources of calories. In the second method (Willett and Stampfer, 1986), represented by the last column of Table 111, the “difference” in effect between the nutrient and other sources of calories is estimated. Results are shown in terms of the log OR and its standard error, since these provide the most direct comparison. The units of nutrient intake used are arbitrary, since they do not affect the statistical significance of any observed association. The data in the first column of Table I11 clearly show that the only component of total energy having a positive association with colon-cancer risk is carbohydrates. Therefore, carbohydrates are driving the harmful total energy intake effect, but this association is not completely independent of the other sources of calories, as seen by thep value of 0.58 given in the second column of Table 111. The difference in effect between carbohydrates and other sources of calories is not statistically significant, as shown by thep value for the difference in the last column of Table 111. The analysis for total fat and protein shows no evidence for any association with risk, in accordance with the analysis in Table I. DISCUSSION
The present study provides some support for the existence of a positive association between total caloric intake and increased risk of colon cancer, while there is strong evidence of monotonically decreasing risk with increasing intake of dietary fibre. Part I of this report showed that increasing vegetable consumption was associated with decreasing risk, and the association with dietary fibre reflects the higher consumption of vegetables by control subjects. Another marker of fruit and vegetable intake, namely vitamin C, also exhibits an inverse association, but the effect is weaker and may be the result of
860
ISCOVICH ETAL.
TABLE 111- LOG OR” FOR COLON CANCER PER 1000KCALSIDAY FOR TOTAL FAT, PROTEIN AND CARBOHYDRATE INTAKE Other calories
Nutrient Dietary factor
Total fat Protein Carbohydrates
h e OR
SE
D
h e OR
SE
I)
-0.26 -0.50 1.13
0.52 1.77 0.43
0.61 0.78 0.008
1.22 0.78 0.19
0.41 0.36 0.35
0.003 0.03 0.58
Difference
SE
P
1.48 1.28 0.94
0.82 2.04 0.64
0.07 0.53 0.14
*Estimatesfrom a continuous variable model.-“Other calories” model: E(D) = d + d (T - N) + d N (Howe et ab, 1986;Howe, 1989) and “Difference”model: E(D = c + c T + c N (Willett and Stampfer, 1986), where T = total calories, N = nutrient which contributes to calories. and c and d are coe cients.
A
measurement error. Fibre intake and vitamin C are highly correlated (r2 = 0.89), and the effect of imperfect measurement of dietary fibre is that adjustment for the confounding effect of fibre on vitamin C will be incomplete, which could explain the weak association seen for this vitamin. Although in Part I it was shown that increased risk was associated with food containing high levels of fat, such as red meat, eggs and cheese, the findings in Part I1 suggest that the effect of total caloric intake is not due to total fat, but perhaps to carbohydrates. For policy recommendations, it is helpful to know that reduction in total calories may reduce risk, while within total intake of food items a proportional increase in foods containing fibre, particularly vegetables, may contribute to decreased risk for colon cancer. Some methodological problems in case-control studies of diet have been well described. Downward bias in the OR results from the imprecise nature of questionnaire-based dietary information (Tzonou et al., 1986), relative to the small variation in food consumption habits within populations (Wynder, 1975; Morgan et al., 1983). The questionnaire used for this study was only semi-quantitative, and was not intended to provide detailed information on micro- or macro-nutrients, such as the type of fat. The availability of an earlier dietary survey nevertheless permitted a rough estimation of nutrient levels. Biases which inflate ORs may be produced through differential recall between cases and controls (Marshall et al., 1981; Block, 1982), or through the influence of pre-clinical disease on diet (Byers et al., 1987), particularly if this state is long-standing (Lillienfeld et al., 1979). To avoid this latter problem, the questionnaire on food consumption asked about the 5-year period up to 6 months before the interview. The positive association between total energy intake and risk of colon cancer found in this study is in contrast to the results of a prospective study among 88,751 women, including 150 cases of colon cancer, where total fat, mainly from animal sources, was associated with increased risk, while total energy intake was not (Willett et al., 1990). While total energy intake has been associated with significantly increased risk in various studies (Jain et al., 1980; Bristol et al., 1985; Potter and McMichael, 1986; Kune and Kune, 1987; Lyon et al., 1987; Graham et al., 1988; Slattery et al., 1988; Whittemore et al., 1989), it has not been in others (Berta et al., 1985; MacquartMoulin et al., 1986; Tuyns et al., 1987). Analysis is complicated,
however, by the fact that components of total energy intake are correlated, and it is necessary to sort out whether positive associations with total energy intake are due to total energy intake itself or one of its components, independent of the total intake. As shown in this study, methods proposed by Howe (1989) and by Willett and Stampfer (1986) provide means by which this can be accomplished. Several case-control studies have reported inverse associations with fibre and the risk of colon cancer (Modan et al., 1975; Dales et al., 1979; Willett and Stampfer, 1986; Lyon et al., 1987; Graham et al., 1988; Slattery et al., 1989; Whittemore et al., 1990), but this has not been reported in other studies (Jain et al., 1980; Berta et al., 1985; Bristol et al., 1985; MacquartMoulin et al., 1986; Potter and McMichael, 1986). When sources of fibre have been examined, it has been shown that fibre from fruits and vegetables, and not cereals, is consistently associated with decreased risk of colon cancer (Willett et al., 1990). Several hypotheses attempt to explain the mechanism by which dietary fibre may protect against colon cancer. Potential mechanisms include the physical capacity of fibre to increase stool bulk and transit time, the binding by fibre of bile acids (which may promote cancer), and the possibility that fibre acts as a substrate for bacterial fermentation, which can increase stool bulk and have some anti-carcinogenic effects. Fibre may, however, be a marker for anti-carcinogens found in plants, and it is therefore important to emphasize that the primary source of fibre in this study was plants and not supplementation. Recommendations based on the findings of the present study should therefore be made on the basis of increasing the consumption of foods rich in fibre, within a diet limited to a certain level of total energy intake.
ACKNOWLEDGEMENTS
We thank the clinics, hospitals and pathologists of La Plata city. We also thank Ms. R. Iscovich and Ms. F. Lubin for their valuable comments and advice. The analysis was undertaken in part while Dr. L‘Abbe was the recipient of a post-doctoral fellowship from the International Agency for Research on Cancer, and in part through her National Health Research Scholar Award (Canada).
REFERENCES
BERTA,J.L., COSTE,T., RAUTUREAU, J., GUILLOUD BATAILLE,M. and PEQUIGNOT, G., Diet and rectocolonic cancers: results of a case study. Gastroenterol.din. Biol., 9,348-353 (1985). BLOCK,G., Review of validations of dietary assessment methods. Amer. J. Epidemiol., 115,492-50s (1982). BRISTOL,J.B., EMMENT,P.M., HEATON,K.W. and WILLIAMSON, R.C.N., Sugar and fat, and the risk of colorectal cancer. Brit. med. J., 291,1467-1470 (1985). BYERS,T.E., MARSHALL, J.R., ANTHONY, E., FIEDLER, F. and ZIELEZNY, M., The reliability of dietary history from the distant past. Amer. J. Epidemiol., 107,510-521 (1987).
DALES, L.C., FREIDMAN,G.D., URY, H.K., GROSSMAN,S. and WILLIAMS, S.R., A case-controlstudy of relationships of diet and other traits to colorectal cancer in American blacks.Amer.J. Epiderniol., 109,
132-144 (1979). GRAHAM, S., MARSHALL,J., HAUGHEY, B., MITTELMAN, A., SWANSON, M., ZIELEZNEY, M., BYERS,T., WILKINSON, G. and WEST,D., Dietary epidemiology of cancer of the colon in western New York. Amer. J. Epidemiol., 128,490-SO3 (1988). HOWE,G.R., Letter to the editor re: “Total energy intake: implications for epidemiologic analyses” by Willett and Stampfer. Amer. J. Epidemiol., 129,1314-1315 (1989).
861
COLON CANCER IN ARGENTINA. 11: FIBRE, FAT AND NUTRIENTS
comparison of dietary methods in epidemiologic studies. Amer. J. Epidemivl., 107,488-498 (1983). POTTER,J.D. and MCMICHAEL, A.J., Diet and cancer of the colon and rectum: a case-control study. J. nut. Cancer Inst., 76,557-569 (1986). SLATTERY, M.L., SORENSON, A.W., MUHONEY, A.W., FRENCH,T.K., D. and STREET,J.C., Diet and colon cancer: assessment KRITCHEVSKY, of risk factor by fiber type and food source. J. nat Cancer Inst., 80, 1474-1480 (1988). [Erratum,J. nat Cancer Insf.,81,1042 (1989)l. TUYNS,A.J., HAELTERMAN, M. and KAAS, R., Colorectal cancer and the intake of nutrients: oligosaccharides are a risk factor, fats are not: a case-control study in Belgium. Nufr.Cancer, 10,181-196 (1987). TZONOU,A,, KALDOR,J.M., DAY, N.E. and TRICHOPOULOS, D., Misclassification in case-control studies with two dichotomous risk factors. Rev. Epidimivl. Sanfe‘Publ.,34,lO-17 (1986). WHITTEMORE, A.S., WU-WILIJAMS,A.H., LEE, M., SHU, Z., GALLAGHER, R.P., DENG-Ao,J., LUN,Z., XIANGSUI, W., KUN,C., JUNG, L., YAO,K.J., PAFFENBARGER, JR., R.S. and D., THE,C.Z., CHENGDE, HENDERSON, B.E., Diet, physical activity, and colorectal cancer among Chinese in North America and China. J. nut. Cancer Insf., 82,915-923 (1989). WILLETT, W.C. and STAMPFER, M.J., Total energy intake: implications for epidemiologic analyses. Amer. J. Epidemivl., 124,307-309 (1986). WILLETT,W.C., STAMPFER, M.J., COLDITZ,G.A., ROSNER, B.A. and F.E., A prospective study of diet and colon cancer in women. SPEIZER, New Engl. J. Med., 323,1664-1672 (1990). WYNDER, E.L., The epidemiology of large-bowel cancer. Cancer Res., 35,3388-3394 (1975).
HOWE,G.R., MILLER,A.B. and JAIN,M., Letter to the editor re: “Total energy intake: implications for epidemiologic analyses.” Amer. J. Epidemivl., 124,157-159 (1986). INSTITUTO NACIONAL DE LA NUTRICION, Tablas de las cvmpvsicivn quimica de lvs alimenfvs.Buenos Aires (1945). JAIN,M., COOK,G.M., DAVIS,F.G., GRACE,M.G., HOWE,G.R. and MILLER,A.B., A case-control study of diet and colo-rectal cancer. Int. J. Cancer, 26,757-768 (1980). KUNE,G.A. and KUNE,S., The nutritional causes of colorectal cancer: an introduction to the Melbourne Study. Nutr. Cancer, 9,1-4 (1987). LILLIENFELD, D.E., GARGAGLIANO, C.F. and LILLIENFELD, A.M., Gastric cancer (letter). J. nut. Cancer Insf., 57,9 (1979). LYON,J.L., MAHONEY, A.W., WEST,D.W., GARDNER, J.W., SMITH, K.R., SORENSON, A.W. and STANISH, W., Energy intake: its relationship to colon-cancer risk.J. nut. Cancer Insf., 78,853-861 (1987). MACQUART-MOULIN, G., RIBOLI, E., CORNEA,J., CHARNAY, B., BERTHEZ~NE, P. and DAY,N., Case-control study on colorectal cancer and diet in Marseilles. Znf.J. Cancer, 38,183-191 (1986). MARSHALL, J.R., PRIORE,R., GRAHAM,S . and BRASURE, J., On the distortion of risk estimates in multiple exposure level case-control studies. Amer. J. Epidemivl., 113,464-473 (1981). MODAN,B., BAREIL,V., LUBIN,F., MODAN,M., GREENBERG, R.A. and GRAHAM,S., Low fiber intake as an etiology factor in cancer of the co1on.J. nut. Cancer Inst., 55,15-18 (1975). MORGAN,R.W., JAIN,M., MILLER,A.B., CHOI,N.W., MATHEWS, V., MUNAN, L., BURCH,J.D., FEATHER,J., HOWE,G.R. and KELLY,A., A
APPENDIX TABLE I - CUTPOINTS FOR QUARTILE AND DAILY INTAKE FOR CASES AND CONTROLS
Total energy (kcaliday) Total fat (g/day) Protein (giday) Carbohydrates (g/day) Dietary fibre (g/day) B-carotene (IU/day) Vitamin C (mg/day)
1
2
3
1915.09 95.04 89.15 167.41 8.87 44 18.21 73.19
2452.52 123.31 119.10 233.50 14.31 8021.68 126.02
3195.65 149.20 140.50 309.70 21.71 13044.8 184.65
Mean intake
Cases
Controls
2649 124.40 115.50 265.40 14.08 8194 118.10
2612 126.00 116.80 253.20 16.21 9085 134.50
