In a 6-year prospective study, the authors examined the relation between diet and incident colon cancer
among 32,051 non-Hispanic white cohort members of the Adventist Health Study (California, 1976-1982) who,
at baseline, had no documented or reported history of cancer. The risk of colon cancer was determined from
proportional hazards regression with adjustment for age and other covariates.

The authors found a positive
association with total meat intake (risk ratio (RR) for ≥1 time/week vs. no meat intake =  1.85, 95% confidence
interval (Cl) 1.19-2.87; p for trend =  0.01) and, among subjects who favored specific types of meat, positive
associations with red meat intake (RR for >1 time/week vs. no red meat intake =  1.90,  95% Cl 1.16-3.11; p
for trend =  0.02)  and white meat intake (RR for >1 time/week vs.  no white meat intake =  3.29,  95%  Cl
1.60-6.75;  p for trend =  0.006). An inverse association with legume intake (RR for >2 times/week vs.  <1
time/week =  0.53, 95% Cl 0.33-0.86; p for trend =  0.03) was observed. Among men, a positive association
with body mass index was observed (relative to the RR for fertile III (>25.6 kg/m2) vs. fertile I (<22.5 kg/m2) = 2.63,
95%  Cl 1.12-6.13; p for trend =  0.05).

A complex relation was identified whereby subjects exhibiting a high
red meat intake, a low legume intake, and a high body mass experienced a more than threefold elevation in
risk relative to all other patterns based on these variables. This pattern of putative risk factors would likely
contribute to increases in both insulin resistance (high body mass, high red meat intake)  and glycemic load
(low legume intake), a synergism that, if causal, implicates hyperinsulinemic exposure in colon carcinogenesis.

The overall findings from this cohort identify both red meat intake and white meat intake as important dietary
risk factors for colon cancer and raise the possibility that the risk due to red meat intake reflects a more
complex etiology. Am J Epidemiol 1998;148:761-74.

Colon cancer is one of the most commonly diagnosed malignancies in the United States and is expected to produce 47,700 cancer deaths in 1998 (1).

Although hereditary syndromes are an established risk factor for this disease,  current evidence suggests that less than 20 percent of the variation in colon cancer incidence is explained by known hereditary syndromes (2,  3). 

International correlation studies (4,  5)  have shown that the highest incidences of colon cancer occur in North America,  Great Britain,  and parts of Europe and that the lowest incidences occur in Asia, Latin America,  and Africa.  Migrant studies (6,  7) show an elevation in risk of colon cancer in populations that have moved from low-incidence (Japan, China)  to high-incidence (United States)  areas.  These findings suggest that the variation in colon cancer incidence is strongly influenced by environmental factors.

Numerous prospective and case-control studies (8) have shown associations between diet and colon cancer.  A relation commonly found   in epidemiologic studies is an increase in risk associated with a high-fat, low-fiber diet pattern.  In some (9-12)  but not all studies (13,  14),  antioxidant vitamins,  calcium,  and vitamin D have shown protective effects against colon cancer, colorectal adenomas, or colonic epithelial cell proliferation.  Specific   foods   associated with a decreased risk of colon cancer include cruciferous vegetables,  fruits,  and legumes (8,  15-17). 

The association   with   these   foods   has been   attributed   to   the putative anticarcinogenic effects of certain compounds (e.g.,  carotenoids,  ascorbic   acid,  tocopherols,  flavonoids, indoles, folate,  protease inhibitors, plant sterols,  selenium,  diallyl sulfide)  found in high concentrations in vegetables and fruits (8,  18).  Meat intake has shown a positive association with colon cancer risk in a number of populations in the United States (19-21)  and other nations (22-28). 

In a few of these studies, specific components of total meat intake, such as meat fat,  meat protein, red meat,  and certain methods of cooking and processing meat, have been identified as contributors to the elevation in risk (19, 20, 25,  29).  Some recent reports (30-34)  indicate that other exogenous factors (physical activity,  obesity, aspirin use,  cigarette smoking)  may contribute to the variation in colon cancer risk and should be considered when investigating the independent effect of diet.

In this study,  we investigated the relation between diet and colon cancer among cohort members of the Adventist Health Study (35). 

Surveillance data reported in an earlier study of a cohort of California Seventh-day Adventists have shown about 60 percent of the mortality rate from colorectal cancer that was found among comparable members of the American Cancer Society cohort and that the lower risk among Adventists persists even after restricting the comparison to members of both cohorts who never smoked (36). By church proscription, the Adventist population is characterized not only by very little tobacco use and alcohol consumption but also by the large proportion   of   the population   that   adheres to   a church-recommended vegetarian diet pattern (35). 

Therefore, a reasonable hypothesis would be that the lower risk of colon cancer among Adventists is attributable to a lower intake of animal products. Some support for this hypothesis can be found in clinical data showing lower rates of colonic epithelial cell proliferation   among vegetarian cohort members of the Adventist Health Study as compared with either nonvegetarian cohort members or a nonvegetarian general population sample (37).  In the current prospective investigation of 32,051 cohort members of the Adventist Health Study, we tested such hypotheses by examining the association between incident colon cancer and intake of selected foods and food groups in the cohort.

MATERIALS AND METHODS
Study population
The Adventist Health Study is a prospective investigation of 34,198 non-Hispanic white California Seventh-day Adventists and others living in Adventist households. 

Between 1974 and 1976, an attempt was made to identify all California Seventh-day Adventists by using a questionnaire that was mailed to households listed on the membership rosters of all   California Seventh-day Adventist churches.  The details of this census taken to identify California Seventh-day Adventists have been described elsewhere (35). In 1976, a lifestyle questionnaire was sent to 59,081 persons identified by this census as being age 25 years or older.  There was a 75.1 percent response rate to the lifestyle   questionnaire among non-Hispanic whites.

This respondent group (n =  34,198) became the incidence population for our cohort study.

During the follow-up period,  incident cancer cases in the study population were ascertained by using two methods. The first involved mailing annual questionnaires to all participants, in which we requested information on any hospitalization during the previous 12 months.  Permission to review any relevant medical records was also obtained,  and pertinent portions of these hospital records were microfilmed by study field representatives to enable confirmation of the diagnosis by Adventist Health Study physicians. This confirmation required a histology report of primary adenocarcinoma of the colon. Follow-up using this method was completed for 97 percent of the participants.

The second method involved linking computerized records (37) with two population-based tumor registries operating in California in 1976 (the Cancer Surveillance Program in Los Angeles County and the Resource for Cancer Epidemiology   Program   in   San Francisco).

From chart review (20,702 medical charts) and tumor registry record linkage,  1,406 incident cancer cases were identified during the 6-year period. Of these, we selected the 166 cases who were diagnosed as primary adenocarcinomas of the colon (International Classification of Diseases, Ninth Revision (ICD-9) code 153) (38)  through the rectosigmoid junction (ICD-9 code 154.0)  (38)  as the endpoint used in this study.

For this analysis, we excluded subjects who reported on their questionnaires that they had a previous history of cancer and subjects whose medical charts indicated a previous history of cancer (n =  2,147).  Hence,  the analytic population at baseline consisted of 32,051 subjects,  with 157 cases (colon,  135 cases;  rectosigmoid junction, 22 cases) diagnosed during the follow-up.

Lifestyle questionnaire
Subjects completed a mailed lifestyle questionnaire that included questions on demographics,  diet,  physical activity,  psychosocial factors,  socioeconomic factors, and medical history. The dietary section consisted of 55 semiquantitative food frequency questions. Most dietary questions had eight frequency categories ranging from “never” to “more than once each day.”

The meat index was determined from responses to six questions on the current frequency of consumption of specific meats (beef (hamburger,  steak,  other),  pork, poultry (chicken,  turkey),  and fish)  and one question on the current frequency of consumption of any meat.

A physical activity index was calculated from subjects’  responses to questions about their participation in vigorous leisure-time or occupational activities and was considered “high” for frequent (≥15 minutes per session,  >three times per week)  participation,  “moderate”  for less frequent   (<15 minutes per session, <three times per week)  participation,  or “none/low” for “rarely or never” participation in vigorous activity.

Body mass index (kg/m2) was determined from weight and height information reported on the questionnaire.

In a random sample of 168 cohort members (39), the correlation was 0.95 (Pearson’s r) between the weight reported on the 1976 questionnaire and the weight measured during an in-person interview up to 1 year after the questionnaire was returned. Other variables considered in the analysis were obtained from responses to questions on smoking (current, past, never), alcohol consumption (beer, wine, liquor combined), age at first pregnancy (age <24 years, age >24 years), hormone replacement therapy (recent or prior use,  never used),  aspirin use (≥1 time/week, <1 time/week), diabetes (any type, no history),  and parental history of colon cancer (one or two parents, neither parent).

Validity study
The validity of the dietary data was tested in a random sample of 147 cohort members who participated in a detailed dietary substudy that has been described in other reports (40-41).

The substudy participants completed a food frequency   questionnaire similar to the one used in the present study,  and they also provided five 24-hour recalls on random days during a 3-month period. Using the averaged 24-hour recall data as a standard,  we calculated corrected correlation coefficients (42-43) between the frequency of consumption and use of foods and food groups reported on the questionnaire and the estimated gram weight of the corresponding items reported on the 24-hour recalls.  For pertinent foods examined in this study, these correlations were similar in magnitude to those documented in other populations (44-45)  and were as follows:  meat index,  0.83;  beef index,  0.49; poultry, 0.57; and legumes, 0.31. All were statistically significant at a =  0.02. 

Among those subjects in the higher intake categories of meat ( s i time/week)  and legumes (>2 times/week)  as measured by 24-hour recalls,  corresponding questionnaire categories correctly classified 93 percent of the meat intake and 57 percent of the legume intake.  In a random sample of cohort members for whom stool samples were collected up to 1 year after baseline (46),  a significant positive correlation was found between the intake of legumes as reported on the baseline questionnaire and the total fiber content of the stool (r =  0.30);  similar positive correlations were found with the stool content of water-insoluble fibers (cellulose, lignin,  hemicelluloses) but not of water-soluble fiber (pectin).

Statistical analysis
For the analysis,  food and food group data were each divided into three frequency levels.  Specifically, for total meat intake, the index described above contained the following three levels to enable investigation of the major meat intake patterns in this cohort: 1) strict vegetarian (no meat intake), 2)  occasional meat intake ( > 0 -

< l time/week), and 3) nonvegetarian (meat intake S:l time/week). Criteria for categorization of the food variables were established before the analysis began. Subjects with missing values for dietary and nondietary variables were retained in all analyses by using methods described by Woodward et al. (47).

The association between dietary variables and the risk of colon cancer was investigated by using Cox proportional hazards models that included covariates for age, sex, body mass index, parental history of colon cancer, physical activity level, current smoking, past smoking, alcohol consumption, and aspirin use.

The time variable for each subject was the duration of follow-up as measured from the date on which the questionnaire was returned (1976-1977).

Cases were assigned follow-up time from the date that they returned the questionnaire to the date of their colon cancer diagnosis; noncases were censored at the end of the follow-up period (January 1, 1983), the date of death (if mortality was ascertained during follow-up), or the date of last contact (loss to follow-up affected less than 3 percent of the subjects).

Two statistical tests were performed for each dietary variable. To assess the overall significance of the individual food variables, we performed a log-likelihood ratio test of the indicator food variables. A multivariate test for linear trend across food intake levels was performed by replacing the indicator food variables in each multivariate model with a single variable representing the median frequency of consumption for a given intake level and by using the Wald x1 value computed for the regression coefficient of this variable to test the null hypothesis of no linear trend component in colon cancer risk across levels of intake.

Variables for the trend test included eight intake levels taken directly from the questionnaire for the nondairy food variables (never, > 0 - < l time/month, 1-2 times/month, 1-2 times/week, 3-4 times/week, 5-6 times/ week, 1 time/day, > 1 time/day); seven levels from the questionnaire for the dairy variables (never, > 0 - < l time/week, 1-6 times/week, 1 time/day, 2-3 times/day, 4-5 times/day, >5 times/day); and three levels for the meat indices (never, > 0 - < l time/week, >1 time/week) computed from questionnaire variables.

RESULTS
During 178,544 person-years of follow-up (1977-1982),  157 colon cancer cases were identified in the analytic population.  Baseline characteristics of the population are presented in table 1. Results are shown by level of meat intake,  as this was an exposure of particular interest.  Subjects in each of the meat intake groups   were   compared,  and   the   significance   of between-group differences was assessed by using x2, analysis of variance, or Kruskal-Wallis tests, as appropriate.  We found no significant differences between the meat intake groups based on age,  sex,  parental history of colon cancer,  or physical activity level.

Nonvegetanans had a significantly higher body mass index than the other two groups. This group also used more aspirin,  drank more alcohol,  and were more likely to be current or past smokers. Among nonvegetarians,  the mean intake level of all meats was five times per week,  and beef was consumed most fre-
quently.

The risk estimates for certain nondietary factors are presented in table 2, with adjustment for age, sex, and parental history of colon cancer. Positive associations were found for a parental history of colon cancer, alcohol consumption,  body mass index,  and diabetes.

When analyses were stratified by sex (men, 65 cases; women,  92 cases),  strong positive associations for body mass index and diabetes were evident among men but not among women.

The risk estimates for 12 food variables are presented in table 3, with adjustment for age, sex, parental history of colon cancer,  and other covariates.  The following food variables were also tested but showed little evidence of important associations:  eggs,  tomatoes,  white rice,  brown rice,  a fruit index,  total milk products, vegetarian protein products, vitamin supplements (A, C, E), and coffee.

The strongest risk factor association among the food variables listed in table 3 was found for total meat intake. An elevated risk was also apparent for red meat and white meat (poultry   +  fish),  but the overall strength of these associations in the total population was of a lower magnitude when compared with total meat intake. A high intake of legumes (beans, lentils, and split peas)  showed the strongest protective association among the foods shown in table 3,  and,  in further analyses, we found that legume intake was not strongly correlated with body mass index (r =  -0.08) or total meat (r =  -0.27). A significant inverse relation was also evident for cottage cheese,  and similar although nonsignificant (p ≤  0.10) trends were shown for salad and green vegetables.

We found that in the total study population, red meat intake and white meat intake were highly correlated (r =  0.77). This finding raises the possibility that risk ratios for specific meats listed in table 3 reflect confounding by other meat types. 

Therefore,  to more closely examine the independent contribution of red meat and white meat to the risk identified for total meat intake,  we used stratified analyses to estimate the risk among subjects who consumed a particular type of meat more frequently than other types.

Strong positive trends were shown for red meat intake   among   subjects   who consumed   low   levels ( 0 - < l time/week)  of white meat and for white meat intake   among subjects   who consumed   low   levels ( 0 - < l   time/week)  of red meat. 

These associations remained evident after further categorization of the red meat (relative to no red meat intake: risk ratio (RR) for > 0 - < l   time/week   =  1.38,  95 percent   confidence interval (CI)  0.86-2.20;  RR for 1-4 times/week   =1.77,  95 percent CI 1.05-2.99; and RR for >4 times/week =  1.98,  95 percent CI 1.00-3.89)  and white meat (relative to no white meat intake: RR for > 0 - < 1 time/week =  1.55,  95 percent CI 0.97-2.50;  RR for 1-4 times/week =  3.37, 95 percent CI 1.60-7.11; and RR for >4 times/week =  2.74,  95 percent CI 0.37-20.19)  variables to higher intake levels.  Taken together, these data suggest that both red meat and white meat are important contributors to the overall risk observed for total meat intake.

The associations with total meat intake and with legume intake that were identified in single food models were examined further in a model containing both food groups along with the usual potential confounders. 

In this model,  high intakes of total meat   ( s i time/week)  and legumes (>2 times/week)  continued to have significant associations with colon cancer risk, the overall significance of each food variable was borderline,  and strong estimated trends remained.  Adding other foods (cottage cheese,  salad)  to this model diminished the magnitude of their estimated associations without,  however,  indicating major confounding of the risk estimates for total meat and legumes.

To investigate a possible complex relation between these food variables and the risk of colon cancer,  we formally tested for a multiplicative interaction using a model in which the product term for meat X legumes attained significance (p   =  0.03). 

Next,  we further evaluated the interaction from a model that provided risk ratios (relative to vegetarians with a legume intake of < 1 time/week) for nine categories of total meat (nonvegetarian,  occasional meat intake, vegetarian)  by legume intake (>2 times/week,  1-2 times/week,  <1 time/week)  and adjusted for age, sex, and parental history of colon cancer (refer to the Appendix for specific values).  This model identified an especially potent increase in risk for nonvegetarians who had a low legume intake (RR =  2.54, 95 percent Cl 1.20-5.37).

To examine whether the apparent complex relation between meat intake and legume intake was restricted to specific meats,  we tested separate models for red meat by legume intake and white meat by legume intake. 

As noted above,  to investigate the contribution of specific meats, we restricted the analysis of red meat to those subjects who consumed low levels (< 1 time/week) of white meat (112 cases) and the analysis of white meat to those who consumed low levels (< 1 time/week)  of red meat (82 cases).  The models in figure 2 show that the apparent modification in risk for meat intake by legume intake was evident only for red meat intake (RR for a red meat intake of s 1 time/week +  a legume intake of < 1 time/week =2.28 , 95 percent Cl 1.28-4.05).

Full article

—
Pramil N. Singh and Gary E. Fraser

REFERENCES
1.  Landis SH, Murray T, Bolden S, et al. Cancer statistics, 1998. CA Cancer J Clin 1998;48:6-29.
2.  Toribara NW, Sleisenger MH. Screening for colorectal cancer. N Engl J Med 1995;332:861-7.
3.  Lynch HT,  Smyrck TC,  Watson P,  et al.  Genetics,  natural history, tumor spectrum,  and pathology of hereditary nonpolyposis colorectal cancer: an updated review. Gastroenterology 1993;104:1535-49.
4.  Doll R, Peto R. The causes of cancer: quantitative estimates of avoidable risks of cancer in the United States today.  J Natl Cancer Inst 1981;66:1191-1308.
5.  Vogel V,  McPherson RS.  Dietary epidemiology   of colon cancer.  Hematol Oncol Clin North Am 1989;3:35-63.
6.  Whittemore AS. Colorectal cancer incidence among Chinese in North America and the People’s Republic of China:  variation with age, sex, and anatomical site. Int J Epidemiol 1989; 18:563-8.
7.  Haenszel W,  Ber JW,  Segi M,  et al.  Large bowel cancer in Hawaiian Japanese. J Natl Cancer Inst 1973;51:1765-79.
8.  Potter JD.  Nutrition and colorectal cancer.  Cancer Causes Control 1996;7:127-46.
9.  Lipkin M, Newmark HL, Kelloff G. Calcium, vitamin D, and prevention of colon cancer. Boca Raton, FL: CRC Press, 1991.
10.  Garland C,  Shekelle RB,  Barrett-Connor E,  et al.  Dietary vitamin D and calcium and risk of colorectal cancer: a 19-year prospective study in men.  Lancet 1985;l:307-9.
11.  Bostick RM, Potter JD, Sellers TA, et al. Relation of calcium, vitamin D, and dairy food intake to incidence of colon cancer among older women. The Iowa Women’s Health Study. Am J Epidemiol 1993;137:1302-17.