A Case-Control Study of Colorectal Adenomatous Polyps and Consumption of Foods Containing Partially Hydrogenated Oils¹

Although there are plausible mechanisms for an effect of total dietary fat on CRC,³ consistent epidemiological evidence for this effect is lacking (1, 2, 3). The absence of associations between a high fat diet and CRC in individual level studies has led some researchers to focus attention on investigating the effects of specific fatty acids (4). There has been speculation about the effect of tFA consumption on carcinogenesis (5). Mechanisms for an effect could involve the disruption of the phospholipid cell membrane and associated enzymes and receptors or the disruption of eicosanoid synthesis.

tFAs are one of many fatty acid isomers produced during the partial hydrogenation of vegetable oils. Hydrogenation is a process that uses heat and a catalyst to add hydrogen atoms to the carbon⋕carbon double bonds in unsaturated fatty acids. It converts liquid oil to solid at room temperature and prolongs the shelf life of foods. The carbon⋕carbon double bonds present in unsaturated fatty acid normally exist in a kinked cis structure, whereas PHVOs contain fatty acids whose double bonds exist in a straightened trans structure. Dietary exposure to PHVOs occurs through consumption of margarine, some bottled oils, cake, pie, donuts, rolls, candy, cookies, mayonnaise, potato chips, french fries, and other foods.

The only estimates of hydrogenated oil consumption in the United States located during a Medline search using keywords “trans fatty acids” and “hydrogenated oil” date back to 1978 and 1983. Emken (6) estimated that hydrogenated oils contributed 14% of total energy intake in a diet containing 40% of energy from fat. Mounts (7) reported an estimate of about 10% of total energy intake in a diet containing 42% of energy from fat. The latter estimate relied on USDA food disappearance data. Dietary tFAs probably make up, on average, 5–10% of total fatty acids in the American diet (8) or an average daily consumption of 5–12 g (6, 8, 9). Some of these tFAs are supplied by dairy and meat as a result of biohydrogenation by micro-organisms in the stomach of ruminant animals. No published epidemiological studies conducted on an individual level have yet reported associations between CRC and tFAs or PHVOs.

It is difficult for observational epidemiological studies to separate a single nutrient’s effect from the effects of numerous other compounds present in whole foods because any one nutrient exposure is inextricably confounded by other constituents of the whole foods that supplied that nutrient. Consequently, we decided to use data on the consumption of whole foods that commonly contain PHVOs to address the question of whether or not tFAs might be a risk factor for CRC. This approach does not pretend to be able to distinguish the effects of tFAs from the effects of other fatty acid isomers produced during the hydrogenation of vegetable oils. Estimating the effects of whole foods has the advantage of providing direct information on how one should modify diet to reduce the risk of disease.

Polyps are widely viewed as intermediate stages in the development of CRC and are markers for an increased risk of disease (10). We, thus, examined data from a case-control study of colorectal adenomatous polyps diagnosed at a sigmoidoscopy screening clinic to estimate the effects for the major food sources of PHVOs.

A total of 519 cases of adenomatous polyps and 553 controls were ascertained from two Southern California Kaiser Permanente Medical Center sigmoidoscopy screening clinics between January 1991 and August 1993. Kaiser is a health care management organization that recommends routine screening using a 60-cm flexible sigmoidoscope for all of its members beginning at age 50. Study subjects were 50–74-year-old, English-speaking men and women residing in the metropolitan Los Angeles area. Average sigmoidoscope depth was 55 ± 11 cm (SD) among cases and 59 ± 5 cm (SD) among controls.

Medical eligibility was determined by reviewing medical records. Individuals were excluded for study if they had a history of a condition that could have induced dietary changes that were etiologically unrelated to polyp status. These included past polyps, history of any bowel surgery, inflammatory bowel disease, or cancer (except nonmelanoma skin cancer). Individuals who manifested genetic syndromes that confer exceptionally high risk of polyps (i.e., familial polyposis, as found in Gardner’s and Peutz-Jeghers syndromes) were also excluded.

Cases were “asymptomatic” (referred for routine screening, rectal bleeding, or as a follow-up for positive routine fecal occult blood testing), with a first-time diagnosis of at least one histologically confirmed adenomatous polyp. Approximately 16% of cases and 13% of controls were referred for screening by their physicians because of positive occult blood tests or rectal bleeding. Fifteen cases were excluded when invasive carcinoma was found at the time of sigmoidoscopy screening. Controls had no polyps of any type found at screening and were matched to cases on age (using predefined 5-year intervals), sex, Kaiser center, and date of sigmoidoscopy (using predefined 3-month intervals). If it was not possible to interview the initially matched control, up to three replacements were pursued. The response rate among cases was 83% of 628 initial contacts; control response rate was 80% of 689 initial contacts.

Dietary exposures were measured over the year before sigmoidoscopy using a self-administered FFQ containing 112 food items, 14 beverage items, and questions on the use of supplements and fats in cooking (11, 12). Because FFQs do not collect information on specific foods (i.e., brand and preparation) corresponding to each questionnaire item, it is customary to assign nutrient values for each item using common brands and food preparation methods. For this study, we chose nutrient values corresponding to the foods reported by adults aged 18 and over who participated in the 1988–1989 USDA Nationwide Food Consumption Survey (13) for the Southwest United States. Three cases and two controls whose total caloric intake fell below 500 kcal/day were excluded from our analyses.

Information was collected on physical activity, alcohol consumption, lifetime smoking history, use of NSAIDs, BMI, and demographics during an in-person interview that was administered an average of 5 months after the sigmoidoscopy. The interviewers were blinded to a subject’s disease status for 70% of the cases and 87% of the controls. For individuals who reported engaging in vigorous leisure-time activity at least three times per week, MET-hours per week were assigned by calculating the number of hours a subject engaged in a particular activity and multiplying this by the degree to which the activity increases the body’s metabolic rate (14). Smoking exposure was measured by pack-years.

All of the food items containing PHVOs were grouped according to principal ingredients and common modes of preparation (Table 1). Food group exposures were quantified using daily energy intake so that estimated effects could be interpreted in terms of the amount of whole food consumed. Groupings were defined as follows:

• (a) sweetened baked goods—principal ingredients are refined flour, sugar and fat;

• (b) candy bars—principal ingredients are sugar and fat;

• (c) oils and condiments—principal ingredient is fat; and

• (d) french fries and chips—principal ingredients are starch and fat.

Each item in these food groups contains PHVO as a principal source of its fat content. Brand information was available for the type of bottled oil used in cooking, so consumption of brands containing tFA (Crisco, Wesson, and generic soybean oil) was included in the oils and condiments grouping. The FFQ distinguished between home-baked and ready-made sweetened baked goods and collected information on fats used in baking, so home-baked items prepared using margarine, shortening, or tFA-containing bottled oils were grouped with the corresponding ready-made items that were assumed to contain PHVO. Total energy, red meat, and vegetable intake were considered potential dietary confounders. Red meat and vegetable consumption were measured by totaling the daily energy intakes of the food items in each category.

Main food group exposures were analyzed as categorical variables using 50-kcal daily increments in consumption. Because candy bar consumption was based on the subjects’ response to one food item only, it was analyzed according to the categorizations of intake on the FFQ (less than 1/month, 1–3/month, 1/week, 2–4/week, and so forth). Adjustment for matching variables and confounders was done using unconditional logistic regression. The following confounders were specified a priori based on a review of putative risk factors for adenomas: age, sex, smoking, BMI, alcohol consumption, vigorous leisure time physical activity, total energy intake, red meat consumption, vegetable consumption, and NSAID use. Main exposure variables, matching variables, smoking (current, former, never), and use of NSAIDs (yes/no for three or more tablets/week) were included in the models as categorical variables. BMI, alcohol consumption, physical activity (in MET-hours), red meat consumption, and vegetable consumption were included in the models as continuous variables. To assess whether adding the square or square root of the continuous variables improved model fit, a likelihood ratio test was used to compare the model with and without the additional term. Tests were performed for one covariate at a time, keeping all of the other covariates in the model. If the P of the likelihood ratio test was greater than 0.20 or if the point estimates and SEs for the main exposure changed less than 10%, then the additional square or square root term was omitted. Covariates were excluded from the model using the same P and change in estimate criteria.

Adjustment for total energy intake was done using a quadratic spline with the upper tail restricted (15). The use of a spline allowed for maximum flexibility in the relationship between total energy intake and adenomas. The spline’s ability to conform to the exposure-disease relationship in the data provides more complete control of confounding than conventional approaches (15). This attribute was considered potentially useful for modeling the relation between energy intake and adenomas, since a crude look at the data suggested the presence of a J-shape that would not be as closely approximated using a continuous variable even with the addition of a square root or square term.

Characteristics of the study groups with respect to potential confounders are presented in Table 2. Cases and controls were almost identical with respect to matching variables (age and sex) as well as ethnicity. The average age in both cases and controls was 61 years. Most (65%) of the cases and matched controls were male. Most subjects were non- or ex-smokers. Current smoking was more common among cases than among controls. There was a slightly higher proportion of drinkers among cases compared with controls. Most cases (76%) and most controls (68%) did not engage in vigorous physical activity at least three times per week. A large proportion of cases (71%) and fewer controls (60%) had a BMI above the upper bound of the ideal range (25 kg/m²). The distributions of daily red meat and total energy intake were higher in cases than in controls, whereas the distribution of daily vegetable intake was lower in cases than in controls. Details of the dietary associations have been reported elsewhere (16, 17).

There were positive associations between adenomas and the highest levels of sweetened baked goods consumption in both crude and adjusted data. A nearly 2-fold increase in prevalence odds was found in the three highest consumption levels starting with 250 kcal daily intake (Table 3). An example of a 250-kcal exposure could be an average piece of cake. The adjusted OR comparing 350 kcal daily with less than 50 was 2.1 (95% CI, 1.3–3.5). Associations were attenuated after adjustment for age, sex, smoking, BMI, physical activity, total energy, red meat, and vegetables. There were no clear associations between adenomas and the other three main exposure food groups. Hints of positive associations in the crude data for the highest levels of consumption were mostly eliminated upon adjustment for other covariates.

Most of the attenuation in effect estimates for main exposure food groups was due to adjustment for red meat consumption. The inclusion of NSAID use, alcohol use, date of exam, and clinic in the models did not alter estimates of effect for any of the main exposure food groups or their SEs; therefore, adjustment for them is omitted in the analyses presented here.

Daily dietary tFA was totaled for each subject. There was a monotonically increasing association between adenomas and 2-g increments of daily intake in the crude data (Table 4). A 2.3-fold increase in prevalence odds (95% CI, 1.4–3.7) was associated with the highest consumption level (6+ g/day) relative to the lowest level (<2 g/day). Upon adjustment for matching factors and confounding variables, the association was reduced. No single covariate accounted for the bulk of confounding. Additional adjustment for sweetened baked goods completely eliminated the positive association.

It would be most informative from a public health point of view to derive diet-related effect estimates for specific food items. However, most populations do not eat enough of a single food item for an epidemiologically detectable effect of consuming that item to occur. Obtaining effect estimates for food groups is an alternative. One caveat to the food group approach, however, is that it assumes that units of exposure from each item in a group confer the same risk of disease. For example, an estimate of effect for sweetened baked goods in our study assumes that 1 calorie from a cookie confers as great a risk as 1 calorie from pie.

Estimating effects of foods with similar overall nutrient profiles allows for the likely possibility that an observed effect is due to characteristics of the foods other than measured nutrients. It allows for the equally likely possibility that nutrients in combination are responsible for any observed association with disease. The same interpretations are possible in studies of nutrient effects when the majority of the nutrient is derived from similar foods. Observing consistent associations between an outcome and several different food groups that all supply the same nutrient may provide the strongest evidence for a nutrient effect. Hence, we chose to measure associations between adenomas and four very different food sources of PHVO (and tFAs).

Foods contributing to a group were weighted by total energy content. An alternative could have been to weight by PHVO or tFA content. The latter approach was not chosen because of the absence of a database for hydrogenated oils and the lack of a dependable database for tFA content in foods. Because FFQs do not typically collect information on food brands, there are inevitable inaccuracies in the energy content assumed for each food item as well. Nonetheless, weighting by total energy was chosen as the method that would be least likely to introduce dietary measurement error into the results. The error that exists is likely to be nondifferential, and, on the assumption that misclassification does not occur between category extremes, the resulting bias is likely to be toward attenuating or masking true associations.

The absence of positive associations in our data between adenomas and four very different food groups all characterized by their PHVO content does not support the hypothesis that PHVO increases adenoma risk. Similarly, the few studies that have looked for effects of tFA on induced colon cancer in laboratory animals have not found consistent positive associations (18, 19). The nearly 2-fold increase in risk observed in the highest category of baked goods consumption may be attributable to other characteristics of the food group. High concentrations of sugar and fat and low levels of vitamins, minerals, and fiber may be responsible for the observed association. Deleterious effects of consuming low fiber and high fat in combination have been discussed extensively in the CRC literature (20, 21, 22). Burkitt (23) postulated that the replacement of foods high in fiber with foods high in sugars and refined carbohydrates increases polyp and cancer risk through detrimental alteration of the intestinal bacterial flora. Several studies have reported associations between sugar consumption and CRC (24, 25, 26, 27, 28).

Separation of candy bar and chocolate consumption on our FFQ offered the opportunity to compare the effects of two similar food items differing primarily in source of fat (Table 5). Chocolate usually contains saturated cocoa butter as its primary source of fat, whereas an average traditional candy bar might contain PHVO and some cocoa butter. The two highest categories of chocolate consumption (starting at two servings per week) were associated with increased adenoma prevalence. Upon adjustment, the highest category (5+ servings per week) was still associated with a 2.4-fold increase in adenoma prevalence. The candy bar association was reduced upon adjustment for chocolate consumption. It may be worthwhile to explore the relationship between chocolate and adenomas in other datasets since methylxanthines and other compounds in chocolate affect many functions of the gastrointestinal tract (29).

It remains possible that positive associations between adenomas and the main exposure food groups—aside from sweetened baked goods—were obscured by an insufficient dietary range of exposure and “noise” originating from dietary measurement error. Detectable associations may exist between adenomas and the other food groups at higher levels of consumption than were observed in this study. Positive associations between adenomas and sweetened baked goods did not become apparent until levels of consumption reached 250 kcal daily. It was not possible to categorize consumption levels over 200 kcal daily for the other food groups because of an inadequate number of subjects in the higher consumption levels.

We analyzed the effect of total dietary tFA to address the possibility that exposure to each food group alone may not have reached the threshold value required for an effect to occur. In this analysis, red meat consumption was considered a potentially important confounder because it is a risk factor for adenomas and is also a source of tFA. However, adjustment for red meat alone only partly eliminated the positive association in the data. The bulk of the positive association between tFA and adenomas was eliminated upon adjustment for sweetened baked goods consumption.

An advantage of using adenomatous polyps as a marker for CRC is the reduction of bias that arises from differential recall of diet in case-control studies of life-threatening outcomes. Furthermore, asymptomatic polyps may be less likely than cancer to induce etiologically irrelevant dietary changes. On the other hand, adenomatous polyp outcomes focus on risk factors for presumed precursor lesions, only some of which may be associated with an increased risk of progression to CRC (30).

A disadvantage of using a sigmoidoscopy screening population lies in the potential for incomplete case ascertainment. Because subjects were not scoped throughout the entire length of the large intestine, both cases and controls may have had undetected polyps in the region beyond the reach of the sigmoidoscopy. The prevalence of polyps in subjects deemed polyp-free by sigmoidoscopy may be 15–17% (31, 32). The inclusion of diseased subjects in the control group would produce a bias toward the null in estimating the effect of PHVO consumption when the outcome of interest is defined as “any polyp.” On the other hand, different etiologies may exist for right- and left-sided colon cancer (33, 34, 35), so defining the outcome as “left-sided polyps only” is potentially informative. Our results are not subject to disease misclassification bias if the effect of PHVO consumption on left-sided polyps is of interest.

The associations we observed in our data may represent an effect of consuming sweetened baked goods on the occurrence of colorectal adenomatous polyps. Even if such an effect were confirmed, however, our results do not support the hypothesis that part of the effect is mediated through the PHVO or tFA component of the food. It may be more reasonable to infer that an effect is due to a combination of high sugar, high fat, and low fiber or correlated micronutrients. Finally, the greater-than-2-fold increase in adenoma prevalence odds associated with the highest consumption of chocolate may warrant further study.

We thank Corinne Aragaki for helpful comments.