The association between diet and the risk of cancer has been the subject of a number of epidemiologic studies. The most prevalent designs are the case-control study, the cohort study, and the randomized clinical trial. When the results from epidemiologic studies are interpreted, the potential for confounding must be considered. Individuals who maintain a healthy diet are likely to exhibit other indicators of a healthy lifestyle, including regular physical activity, lower body weight, use of multivitamin supplements, lower smoking rates, and lower alcohol consumption. Even if the influence of these confounding variables is analytically controlled, residual confounding remains possible.

In ecologic studies or international correlation studies, variation in food disappearance data and the prevalence of a certain disease are correlated, generally across different countries. A linear association may provide preliminary data to inform future research but, due to the high probability of confounding, cannot provide strong evidence for a causal link. Food disappearance data also may not provide a good estimate for human consumption. The gross national product is correlated with many dietary factors such as fat intake.² Many other differences besides dietary fat exist between the countries with low fat consumption (less affluent) and high fat consumption (more affluent); reproductive behaviors, physical activity level, and body fatness are particularly notable and are strongly associated with specific cancers.

Studies of populations migrating from areas with low incidence of disease to areas with high incidence of disease (or vice versa) can help sort out the role of environmental factors versus genetics in the etiology of a cancer, depending on whether the migrating group adopts the cancer rates of the new environment. Specific dietary components linked to disease are difficult to identify in a migrant study.

Case-control studies of diet may be affected by recall bias, control selection bias, and confounding. In a case-control study, participants affected by the disease under study (cases) and healthy controls are asked to recall their past dietary habits. Cases may overestimate their consumption of foods that are commonly considered “unhealthy” and underestimate their consumption of foods considered “healthy.” Giovannucci et al.³ have documented differential reporting of fat intake before and after disease occurrence. Thus, the possibility of recall bias in a case-control study poses a real threat to the validity of the observed associations. Even more importantly, in contemporary case-control studies using a population sample of controls, the participation rate of controls is usually far from complete, often 50% to 70%. Unfortunately, health-conscious individuals may be more likely to participate as controls and will thus be less overweight, will consume fruits and vegetables more frequently, and will consume less fat and red meat, which can substantially distort associations observed.

Prospective cohort studies of the effects of diet are likely to have a much higher validity than retrospective case-control studies because diet is recorded by participants before disease occurrence. Cohort studies are still affected by measurement error because diet consists of a large number of foods eaten in complex combinations. Confounding by other unmeasured or imperfectly measured lifestyle factors can remain a problem in cohort studies.

Now that the results of a substantial number of cohort studies have become available, their findings can be compared with those of case-control studies that have examined the same relations. In
many cases, the findings of the case-control studies have not been confirmed; for example, the consistent finding of lower risk of many cancers with higher intake of fruits and vegetables in case-control studies has generally not been seen in cohort studies.⁴ These findings suggest that the concerns about biases in case-control studies of diet, and probably many other lifestyle factors, are justified, and findings from such studies must be interpreted cautiously.

The gold standard in medical research is the randomized clinical trial (RCT). In an RCT on nutrition, participants are randomly assigned to one of two or more diets; hence, the association between diet and the cancer of interest should not be confounded by other factors. The problem with RCTs of diet is that maintaining the assigned diet strictly over many years, as would be necessary for diet to have an impact on cancer incidence, is difficult. For example, in the dietary fat reduction trial of the Women’s Health Initiative (WHI), participants randomized to the intervention arm reduced their fat intake much less than planned.⁵ The remaining limited contrast between the two groups left the lack of difference in disease outcomes difficult to interpret. Furthermore, the relevant time window for intervention and the necessary duration of intervention are unclear, especially with cancer outcomes. Hence, randomized trials are rarely used to examine the effect of diet on cancer but have better promise for the study of diet and outcomes that require a considerably shorter follow-up time (e.g., adenoma recurrence). Also, the randomized design may lend itself better to the study of the effects of dietary supplements such as multivitamin or fiber supplements, although the control group may adopt the intervention behavior because nutritional supplements are widely available. For example, in the WHI trial of calcium and vitamin D supplementation, two-thirds of the study population used vitamin D or calcium supplements that they obtained outside of the trial, again rendering the lack of effect in the trial uninterpretable.

Observational studies depend on a reasonably valid assessment of dietary intake. Although, for some nutrients, biochemical measurements can be used to assess intake, for most dietary constituents, a useful biochemical indicator does not exist. In population-based studies, diet is generally assessed with a self-administered instrument. Since 1980, considerable effort has been directed at the development of standardized questionnaires for measuring diet, and numerous studies have been conducted to assess the validity of these methods. The most widely used diet assessment instruments are the food frequency questionnaire, the 7-day diet record, and the 24-hour recall. Although the 7-day diet record may provide the most accurate documentation of intake during the week the participant keeps a diet diary, the burden of computerizing the information and extracting foods and nutrients has prohibited the use of the 7-day diet record in most large-scale studies. The 24-hour recall provides only a snapshot of diet on one day, which may or may not be representative of the participant’s usual diet and is thus affected by both personal variation and seasonal variation. The food frequency questionnaire, the most widely used instrument in large population-based studies, asks participants to report their average intake of a large number of foods during the previous year. Participants tend to substantially overreport their fruit and vegetable consumption on the food frequency questionnaire.⁶ This tendency may reflect social desirability bias, which leads to overreporting healthy foods and underreporting less healthy foods. Studies of validity using biomarkers or detailed measurements of diet as comparisons have suggested that carefully designed questionnaires can have sufficient validity to detect moderate to strong associations. Validity can be enhanced by using the average of repeated assessments over time.⁷

The most important impact of diet on the risk of cancer is mediated through body weight. Overweight, obesity, and inactivity are major contributors to cancer risk. (A more detailed discussion is provided in Chapter 10.) In the large American Cancer Society Cohort, obese individuals had substantially higher mortality from all cancers and, in particular, from colorectal cancer, postmenopausal breast cancer, uterine cancer, cervical cancer, pancreatic cancer, and gallbladder cancer than their normal-weight counterparts.⁸ Adiposity and, in particular, waist circumference are predictors of colon cancer incidence among women and men.^9,10 A weight gain of 10 kg or more is associated with a significant increase in postmenopausal breast cancer incidence among women who never used hormone replacement therapy, whereas a weight loss of comparable magnitude after menopause substantially decreases breast cancer risk.¹¹ Regular physical activity contributes to a lower prevalence of being overweight and obesity and consequently reduces the burden of cancer through this pathway.

The mechanisms whereby adiposity increases the risk of various cancers are probably multiple. Being overweight is strongly associated with endogenous estrogen levels, which likely contribute to the excess risks of endometrial and postmenopausal breast cancers. The reasons for the association with other cancers are less clear, but excess body fat is also related to higher circulating levels of insulin, insulin-like growth factor (IGF)-1, and C-peptide (a marker of insulin secretion), lower levels of binding proteins for sex hormones and IGF-1, and higher levels of various inflammatory factors, all of which have been hypothesized to be related to risks of various cancers.

Energy restriction is one of the most effective measures to prevent cancer in the animal model. While energy restriction is more difficult to study in humans, voluntary starvation among anorectics and situations of food rationing during famines provide related models. Breast cancer rates were substantially reduced among women with a history of severe anorexia.¹² Although breast cancer incidence was higher among women exposed to the Dutch famine during childhood or adolescence, such short-term involuntary food rationing for 9 months or less was often followed by overnutrition.¹³ A more prolonged deficit in food availability during World War II in Norway was associated with a reduction in adult risk of breast cancer if it occurred during early adolescence.¹⁴

Aside from body weight, alcohol consumption is the best established dietary risk factor for cancer. Alcohol is classified as a carcinogen by the International Agency for Research on Cancer. The consumption of alcohol increases the risk of numerous cancers, including those of the liver, esophagus, pharynx, oral cavity, larynx, breast, and colorectum in a dose-dependent fashion.¹⁵ Evidence is convincing that excessive alcohol consumption increases the risk of primary liver cancer, probably through cirrhosis and alcoholic hepatitis. At least in the developed world, about 75% of cancers of the esophagus, pharynx, oral cavity, and larynx are attributable to alcohol and tobacco, with a marked increase in risk among drinkers who also smoke, suggesting a multiplicative effect. Mechanisms may include direct damage to the cells in the upper gastrointestinal tract; modulation of DNA methylation, which affects susceptibility to DNA mutations; and an increase in acetaldehyde, the main metabolite of alcohol, which enhances the proliferation of epithelial cells, forms DNA adducts, and is a recognized carcinogen. The association between alcohol consumption and breast cancer is notable because a small but significant risk has been found even
with one drink per day. Mechanisms may include an interaction with folate, an increase in endogenous estrogen levels, and an elevation of acetaldehyde. Some evidence suggests that the excess risk is mitigated by adequate folate intake possibly through an effect on DNA methylation.¹⁶ Notably, for most cancer sites, no important difference in associations was found with the type of alcoholic beverage, suggesting a critical role of ethanol in carcinogenesis.

In recent years, reducing dietary fat has been at the center of cancer prevention efforts. In the landmark 1982 National Academy of Sciences review of diet, nutrition, and cancer, a reduction in fat intake to 30% of calories was the primary recommendation.

Interest in dietary fat as a cause of cancer began in the first half of the 20th century, when studies by Tannenbaum¹⁷ indicated that diets high in fat could promote tumor growth in animal models. Dietary fat has a clear effect on tumor incidence in many models, although not in all; however, a central issue has been whether this is independent of the effect of energy intake. In the 1970s, the possible relation of dietary fat intake to cancer incidence gained greater attention as the large international differences in rates of many cancers were noted to be strongly correlated with apparent per capita fat consumption in ecologic studies.² Particularly strong associations were seen with cancers of the breast, colon, prostate, and endometrium, which include the most important cancers not due to smoking in affluent countries. These correlations were observed to be limited to animal, not vegetable, fat.