Epidemiology of Diet and Cancer Risk1

Epidemiology of Diet and Cancer Risk1

Walter C. Willett

Edward Giovannucci


Following cardiovascular disease, cancer is the second leading cause of death in most affluent countries. It also contributes significantly to mortality rates among adults in developing countries (1, 2). In the United States, about one in three persons will be diagnosed with cancer during their lifetime and about 60% of those diagnosed will die of cancer (3). Because rates of cardiovascular death have been declining rapidly and overall cancer mortality has not substantially changed, cancer will likely become the most important cause of death in the United States (2, 4). Although overall cancer rates among adults vary only modestly around the world, the types of cancers are dramatically different (1, 2). In most affluent countries, cancers of the lung, colon, breast, and prostate contribute most to incidence and mortality. In poorer regions and the Far East, cancers of the stomach, liver, oral cavity, esophagus, and uterine cervix are most important. However, cancer incidence rates are very dynamic; many areas of the world are experiencing a transition from the cancer incidence patterns of poorer to those of affluent areas (1). Rates of breast and colon cancer have been increasing in almost all countries.

Although genetics play an important role in the development of cancer, inherited mutations cannot account for the dramatic differences in cancer rates seen around the world. Populations that move from countries with low rates of specific cancers to areas with high rates, or the reverse, almost invariably achieve the rates characteristic of the new homeland (5, 6, 7). The time required to attain the new rate can vary, however, from several decades in the case of colon cancer, to about three generations for breast cancer (7, 8, 9, 10). The dramatic changes in cancer rates within a country provide further evidence for the importance of noninherited factors. For example, in Japan rates of colon cancer mortality increased about 2.5-fold between 1950 and 1985 (11).

The dramatic variations in cancer rates around the world and changes over time imply that these malignancies are potentially avoidable if we were able to identify and then avoid the causal factors. For a few cancers, the primary causes are well known, such as smoking in the case of lung cancer; but for most others, the etiologic factors are less well established. However, strong reasons exist to suspect that dietary and nutritional factors may account for many of these variations in cancer rates. First, a role of diet has been suggested by observations that national rates of specific cancers are strongly correlated with aspects of diet such as per capita consumption of fat (12). Also, numerous studies in animals, including a series of detailed experiments conducted during the 1930s (13), clearly demonstrated that dietary manipulations could influence tumorigenesis dramatically.

A multitude of steps in the pathogenesis of cancer have been identified in which dietary factors could plausibly act either to increase or decrease the probability that the clinical cancer will develop. For example, carcinogens in food that can directly damage DNA are discussed elsewhere in this volume, and other dietary factors may block the endogenous synthesis of carcinogens or induce enzymes involved in the activation or deactivation of exogenous carcinogenic substances (14). Oxidative damage to DNA is likely to be an important cause of mutations and potentially can be enhanced by some dietary factors, such as polyunsaturated fats and iron, or reduced by dietary antioxidants or nutrients that are cofactors for antioxidant enzymes, such as selenium or copper (15). Inadequate intake of dietary factors
needed for DNA synthesis, repair, and methylation, such as folic acid, also could influence the risk of mutation or gene expression. The rate of cell division influences whether DNA lesions are replicated and is thus likely to influence the probability of cancer developing (15). Thus, energy balance and growth rates, which can be influenced by a variety of essential nutrients, could affect cancer rates. Dietary factors can influence endogenous hormone levels, including estrogens and various growth factors, which can influence cell cycling, and thus potentially cancer incidence. Estrogenic substances found in some plant foods also can interact with estrogen receptors, and thus could either mimic or block the effects of endogenous estrogens (14). Many other aspects of diet can alter cell proliferation or differentiation either by direct hormonal effects, such as by vitamins A or D, or indirectly by influencing inflammatory or irritative processes, such as specific fatty acids that are precursors of prostaglandins or that inhibit their synthesis. Many other examples can be given by which dietary factors could plausibly influence the development of cancer (14, 15).


The strong suggestions from international comparisons, animal studies, and mechanistic investigations that various aspects of diet may importantly influence risk of cancer raises two critical sets of questions: Which dietary factors are actually important determinants of human cancer? What is the nature of the dose-response relationships? The nature of the dose-response relationships is particularly important because a substance could be carcinogenic to humans, but there could be no important risk within the range of intakes actually consumed by humans. Alternatively, another factor could be critical for protection against cancer, but all persons in a population already may be consuming sufficient amounts to receive the maximal benefit. In either case, no potential exists for reduction in cancer rates by altering current intakes. The important factors to identify are those for which at least some part of the population is either consuming a toxic level or is not eating a sufficient amount for optimal health. Because cancer is a multistage process, the temporal relationship also is critical to identify.

Various epidemiologic approaches can be used to investigate diet and human cancer relationships (Table 86.1). Relationships between diet and nutrition and cancer incidence in epidemiologic studies can be evaluated by collecting data on dietary intake, using biochemical indicators of dietary factors, or measuring body size and composition. Food frequency questionnaires have been used to assess diet in most epidemiologic studies because they provide information on usual diet over an extended
period of time and are sufficiently adequate to be used in large populations. Food frequency questionnaires have been shown to be sufficiently valid to detect important diet-disease relationships in comparisons with more detailed assessments of diet and biochemical indicators (16). Biochemical indicators of diet can be useful in some situations, but for many dietary factors of interest, such as total fat, fiber, and sodium, no useful indicators exist. DNA specimens have been collected from participants in many studies and allow the examination of gene-diet interactions. Until recently, most information on diet and cancer has been obtained from case-control studies. However, many large prospective cohort studies of diet and cancer in various countries are now ongoing and are generating data that are transforming our understanding of nutrition and cancer. Because the number of studies is now becoming large, systematic reviews and metaanalyses that summarize the findings statistically are becoming increasingly important. However, these summaries are limited by the possibility of selective publication of positive results, the difficulty of combining data on diet, which can be expressed in many different ways, and variations in the control of covariates. Analyses that combine the primary data from the original studies, often called pooled analyses, overcome most of these limitations and are a more reliable way to summarize that data, but such analyses are laborious and not always available.





Prospective Cohort



Cancer rates in populations having different diets are compared by assessing average intake of specific nutrients and determining cancer incidence or mortality.

Earlier diets reported by patients with a particular type of cancer are compared with diets reported by matched controls without cancer.

Incidence of cancer is compared in people whose diets and other potentially relevant factors are determined before follow-up begins.

Incidence of cancer in two groups randomized to specific interventions or to no interventions is compared.


Diet is but one of many variables that distinguish populations. Even crude data on average nutrient intake is difficult to gather. These studies are probably best used to generate hypothesis.

Selection bias may occur if controls do not accurately represent the population from which cases arose. Recall bias can result when patients differ systematically from controls in ability to recall diets. Memory of dietary habits can be faulty among patients and controls.*

In rapidly fatal cancers, researchers must often rely on recall of proxy respondents such as spouses.

Selection bias and recall bias should not occur, but thousands or even tens of thousands of people must be enrolled and their health be monitored for many years before statistical power can be achieved.

Adherence to substantial dietary changes is difficult for many people. Participants cannot be easily blinded to their status. Optimal dosages (e.g. of supplemental nutrients) and dose-response relationships can be difficult to ascertain.

Duration of intervention required is generally unknown but may be decades.

* Measurements of vitamins in blood are sometimes substituted for dietary recall questionnaires in case-control and cohort studies. However, this strategy is not universally applicable because levels in blood do not always accurately reflect intake. For example, β-carotene blood levels are a good index of dietary intake whereas retinol levels in blood are only weakly related to intake. Blood levels must be interpreted with caution in case-control studies because cancer can change the level of a vitamin in the plasma.

Epidemiologic investigations should be viewed as complementary to animal studies, in vitro investigations, and metabolic studies of diet in relation to intermediate end points, such as hormone levels. Although conditions can be controlled to a much greater degree in laboratory studies than in free-living human populations, the relevance of findings to humans always will be uncertain, particularly in regard to dose-response and temporal relationships. Ultimately, our knowledge is best based on a synthesis of epidemiologic, metabolic, animal, and mechanistic studies.


Diet is a complex composite of various nutrients and nonnutritive food constituents, and many types of human cancer exist, each with its own pathogenetic mechanisms. Thus, the combinations of specific dietary factors and cancers are almost limitless. This brief overview focuses primarily on the major cancers of affluent populations and aspects of diet for which strong hypotheses and substantial epidemiologic data exist. Several aspects of diet for which hypothesized preventive roles exist are discussed in further detail in the chapter on chemoprevention.

Energy Balance, Growth Rates, and Body Size

Studies by Tannenbaum and colleagues (13, 17) during the first half of the twentieth century indicated that energy restriction could reduce the development of mammary tumors in animals profoundly. This finding has been consistently replicated in a wide variety of mammary tumor models and for a wide variety of other tumors (18, 19, 20, 21, 22). For example, restriction in energy intake by approximately 30% can reduce mammary tumors by as much as 90% (23). The possibility that this relationship, which is the most consistent and strongest effect of diet in animal studies, also may apply to humans received relatively little attention until recently.

In evaluating the effect of energy restriction on cancer rates in humans, it may be tempting to examine the association between energy intake and incidence of cancer. However, such an approach is likely to be completely misleading because in free-living populations, variation in energy intake is determined largely by energy expenditure in the form of physical activity (24). Thus, for example, energy intake is inversely associated with risk of coronary heart disease owing to the protective effect of exercise against this disease (16). The most sensitive indicators of the balance between energy intake and expenditure are growth rates and body size, which can be measured well in epidemiologic investigations, although they also reflect genetic and other nonnutritional factors. Thus, adult height can provide an indirect indicator of pre-adult nutrition, and adult weight gain and obesity reflect positive energy balance later in life. In populations that were traditionally short, such as the Japanese, rapid gains in height during the last several decades (25) have corresponded with increases in breast and colon cancer rates. Further support for an important role of growth rates comes from epidemiologic studies of age at menarche. An early menarche is a well-established risk factor for breast cancer. The difference in the late age of menarche in China—until recently approximately 17 years (26), compared with 12 and 13 years of age in the United States (27)—contributes importantly to differences in breast cancer rates between these populations. Body mass index, height, and weight have been consistently strong determinants or correlates of age at menstruation (28, 29, 30), but the composition of diet appears to have little if any effect. Collectively, these studies provide strong evidence, consistent with animal experiments, that rapid growth rates before puberty play an important role in determining future risk of breast and probably other cancers. Whether the epidemiologic findings result from only restriction of energy intake in relation to requirements for maximal growth, or whether the limitation of other nutrients, such as essential amino acids, may also play a role cannot be determined from available data.

A positive energy balance during adult life and the resultant accumulation of body fat also contributes significantly to several human cancers. The best-established relationships are with cancers of the colon, kidney, pancreas, esophagus (adenocarcinoma), endometrium, and gall bladder (31, 32, 33, 34, 35, 36, 37, 38). The relation between body fatness and breast cancer is more complex. Before menopause, women with greater body fat have reduced risk of breast
cancer (39, 40), and after menopause a positive but weak association with adiposity is seen. These findings may be the result of anovulatory menstrual cycles in fatter women before menopause (41), which should reduce risk, and the synthesis of endogenous estrogen by adipose tissue in postmenopausal women (42), which is presumed to increase the risk of breast cancer. A complex association between body fat and prostate cancer also may exist (43).

In animal models, reduction in insulinlike growth factor-I (IGF-I) mediates at least part of the effect of energy reduction (44). Insulin is known to be a powerful modulator of bioavailable IGF-I (45). In human studies, increasing evidence exists that high circulating levels of IGF-I and insulin are associated with an increased risk of some cancers that occur in affluent populations, particularly colon cancer (45, 46). In the Physicians’ Health Study of male physicians (47), there was 2.5-fold increased risk of colorectal cancer with increasing levels of plasma C-peptide (a marker of insulin secretion) when extreme quintiles (highest versus lowest) were compared. Increasing waist-to-hip ratio is also associated with an increased risk of colon cancer, independently of body mass index (48). In fact, indirect evidence (49) suggests that factors related to energy balance and dietary patterns that stimulate insulin and IGF-I secretion throughout the life span could account largely for the approximately one third of cancers in affluent countries that are believed to be influenced by nutrition (50).

Dietary Fat and Macronutrients

In the landmark 1982 National Academy of Sciences review of diet (51), reduction in fat intake to 30% of calories was the primary recommendation. This objective has been echoed in subsequent dietary recommendations as well (52, 53). Two lines of evidence stimulated this interest in dietary fat as a cause of cancer.

In the first half of the twentieth century, studies by Tannenbaum and colleagues (13, 17) indicated that diets high in fat could promote tumor growth in animal models. A vast literature on dietary fat and cancer in animals has accumulated subsequently (reviewed elsewhere) (22, 51, 54, 55, 56). However, although dietary fat has an effect on tumor incidence in most models (57, 58), the influence of fat has not been definitely established to be independent of the effect of energy intake (22, 23, 54, 55, 59). Second, a possible relation of dietary fat intake to cancer incidence also has been hypothesized because the large international differences in rates of cancers of the breast, colon, prostate, and endometrium are strongly correlated with apparent per capita animal fat consumption (12, 60, 61, 62).

Fat and Breast Cancer

Although a major rationale for the dietary fat hypothesis has been the international correlation between fat consumption and national breast cancer mortality (12), a study of 65 Chinese counties (63), in which per capita fat intake varied from 6% to 25% of energy, showed only a weak positive association between fat intake and breast cancer mortality. Notably, five counties consumed approximately 25% of energy from fat yet experienced rates of breast cancer far below those of US women with similar fat intake (64), thus providing strong evidence that factors other than fat intake account for the large international differences. Breast cancer incidence rates increased substantially in the United States during the twentieth century, as have the estimates of per capita fat consumption, based on food disappearance data. However, surveys based on reports of individual intake, rather than food disappearance, indicate that consumption of energy from fat, either as absolute intake or as a percentage of energy, actually has declined in the last half of the twentieth century (65, 66), a time during which breast cancer incidence has increased (67).

Many case-control studies have been conducted to investigate the dietary fat effect on breast cancer. In one large study (68), animal fat and total fat intake were not associated with breast cancer. The results from 12 smaller case-control studies were summarized in a metaanalysis by Howe et al (69), which included 4312 cases and 5978 controls. The pooled relative risk (RR) was 1.35 (p < .0001) for a 100-g increase in daily total fat intake, although the risk was somewhat stronger for postmenopausal women (RR = 1.48; p < .001). This magnitude of association, however, potentially could be compatible with biases resulting from recall of diet or selection of controls (70).

A substantial body of data from cohort studies is now available to assess the relation between dietary fat intake and breast cancer in developed countries. In a pooled analysis of prospective studies that included 4980 incident cases of breast cancer (71), no overall association was seen for overall fat intake over the range of less than 20% to greater than 45% of energy from fat. A similar lack of association was seen among postmenopausal women only and for specific types of fat. Only among the small number of women consuming less than 15% of energy from fat was a significant association seen; breast cancer risk was elevated twofold in this group. The lack of any suggestion of an increase in risk with higher total fat intake was confirmed in an update of the pooled analysis with 7329 incident cases (72, 73) and a more recent large prospective study from Europe with 7119 cases (74). In a large cohort of older American women (3501 cases), a weak and marginally significant positive association was seen with total fat (for highest versus lowest quintile RR 1.11 [95% confidence interval = 1.00 to 1.24]) (75). In the Nurses’ Health Study, analyses have been conducted with 14 years of follow-up (2956 cases) (73)— 20 years for postmenopausal women (76)—and with up to six assessments of fat intake, which improves the measurement of long-term intake. No indication of an increased risk associated with high fat intake was found.

These studies included mostly postmenopausal women. A study conducted among 90,655 premenopausal women
26 to 46 years of age at baseline found a statistically significant positive association between animal fat, mainly from red meat and high fat dairy sources, and risk of premenopausal breast cancer (77). In the same population, intakes of red meat and total fat (which were not possible to distinguish) during adolescence were associated with greater risk of premenopausal breast cancer (78). Overall, the prospective studies provide strong evidence against any major association between intake of total fat during midlife and breast cancer incidence. The suggestion that intake of animal fat or red meat during adolescence or premenopausal years may increase risk in premenopausal women requires confirmation. It is possible that diet later in life may have little influence on postmenopausal breast cancer, whereas diet earlier in life may impact premenopausal breast cancer. The effect of early life diet on postmenopausal breast cancer also needs to be examined.

The effect of reducing fat intake on risk of breast cancer has been assessed in two large randomized trials. In the Women’s Health Initiative trial, 48,000 women were randomized to a low-fat diet that tended to be higher in fruits, vegetables, and whole grains than their usual intake (79). After an average of about 7 years, a nonsignificant 9% lower risk of breast cancer was seen in the intervention group (80). However, no differences in plasma concentrations of triglycerides or high-density lipoprotein (HDL) cholesterol between the groups were seen at any time in the trial. This provides clear evidence that there was little difference in fat intake, because a true difference in fat intake does affect these lipid fractions (81). Even the small and nonsignificant reduction in breast cancer incidence could have been due to the modest difference in weights between groups that is compatible with a nonspecific effect of diet counseling. In the second trial, conducted in Canada among women with higher risk of breast cancer determined by mammograms, a nonsignificant 19% higher risk of breast cancer was seen among those randomized to a low-fat diet (82). In this study, the expected changes in plasma HDL cholesterol and triglycerides were seen, confirming that the hypothesis of fat reduction was actually tested.

Although total fat intake has been unrelated to breast cancer risk in prospective epidemiologic studies, and the results of two randomized trials have not supported a benefit of reducing fat intake in midlife or later, some evidence suggests that the type of fat may be important. In animal mammary tumor models, the tumor-promoting effect of fat intake has been observed primarily for polyunsaturated fats when fed in the presence of diets containing approximately 45% of energy from fat (83, 84). However, in prospective studies, polyunsaturated fat generally has not been associated with higher risk of breast cancer within the much lower range seen in human diets (72, 73). The relatively low rates of breast cancer in southern European countries have suggested that the use of olive oil as the primary fat may reduce risk of breast cancer. In case-control studies in Spain and Greece, women who used more olive oil had lower risks of breast cancer (85, 86). Furthermore, olive oil has been shown to be protective relative to other sources of fats in some animal studies (54). More evidence should emerge from prospective studies being conducted in southern Europe.

Fat and Colon Cancer

In comparisons among countries, rates of colon cancer are strongly correlated with national per capita disappearance of animal fat and meat, with correlation coefficients ranging between 0.8% and 0.9% (12, 62). Based on these epidemiologic investigations and animal studies, a hypothesis has developed that dietary fat increases excretion of bile acids, which can be converted to carcinogens or promoters (87). However, evidence from many studies that higher body weight increases risk and higher levels of physical activity reduce risk of colon cancer indicates that at least part of the high rates in affluent countries previously attributed to fat intake may result from sedentary lifestyle and excess energy intakes.

With some exceptions (88, 89, 90, 91), case-control studies generally have shown an association between risk of colon cancer and intake of fat (92, 93, 94, 95, 96, 97, 98, 99) or red meat (100, 101, 102, 103, 104, 105). However, in many of these studies, a positive association between total energy intake and risk of colon cancer also has been observed (92, 93, 94, 95, 96, 98, 99). A metaanalysis by Howe et al (106) of 13 case-control studies found a significant association between total energy and colon cancer, but saturated, monounsaturated, and polyunsaturated fat were not associated with colon cancer independently of total energy.

The relation between diet and colon cancer has been examined in several large prospective studies. These have not confirmed the positive association with total energy intake in case-control studies (107, 108, 109, 110, 111), suggesting that the case-control studies were distorted by reporting bias. Most of the studies did not support an association between fat intake and colon cancer risk independent of energy intake. One exception was the Nurses’ Health Study, which showed about a twofold higher risk of colon cancer among women in the highest compared with those in the lowest quintile of animal fat intake (107). However, in a multivariate analysis of these data, which included red meat and animal fat intakes in the same model, red meat intake remained significantly predictive of risk of colon cancer, whereas the association with animal fat was eliminated. A metaanalysis of 13 prospective cohort studies found no appreciable association between total, animal, or plant fat intake and risk of colorectal cancer (112). In a randomized trial of a low-fat dietary pattern, no effect on colorectal cancer incidence was observed (113).

Fat and Prostate Cancer

Associations between fat intake and prostate cancer risk have been seen in many case-control studies (114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124) but sometimes only in subgroups. In a large case-control
study among various ethnic groups within the United States (125), consistent associations with prostate cancer risk were seen for saturated fat but not with other types of fat. Some of these studies found stronger associations for fat intake and risk of advanced or fatal disease than for total prostate cancer (121, 125, 126).

The association between fat intake and prostate cancer risk has been assessed in several cohort studies. In a cohort of 8000 Japanese men living in Hawaii, no association was seen between intake of total or unsaturated fat (127). However, diet was assessed with a single 24-hour recall in this study so the lack of association may not be informative. In a study of 14,000 Seventh-Day Adventist men living in California, a positive association between the percentage of calories from animal fat and prostate cancer risk was seen, but this was not statistically significant (128). In the Health Professionals Follow-up Study of 51,000 men, a positive association was seen with intake of red meat, total fat, and animal fat, which was largely limited to aggressive prostate cancers (129). No association was seen with vegetable fats. In another cohort from Hawaii, increased risks of prostate cancer were seen with consumption of beef and animal fat (130). Two small studies of men with prostate cancer suggest that high intake of saturated fat at the time of diagnosis is associated with an increased risk of biochemical failure (131) and prostate cancer-specific death (132). The stronger findings for advanced disease and progression, if confirmed, suggest that dietary fat may influence late stages of carcinogenesis. However, the European Prospective Investigation into Cancer and Nutrition (EPIC) study, a large European cohort study, did not find any association between total, saturated, or monounsaturated fat intake and advancedstage prostate cancer (133).

A somewhat puzzling observation has been that intake or blood levels of α-linolenic, a fatty acid comprising only about 1% of total energy intake, has been associated with an increased risk of prostate cancer (especially advanced cancers) in two prospective studies (129, 134) and five case-control studies in diverse populations: Uruguay (135), Spain (136), Norway (137), China (138), and the United States (139). However, other studies have not supported this (140, 141, 142, 143, 144). Whether or not this association is causal needs to be determined, especially because this fatty acid is beneficial in regard to cardiovascular disease (145, 146). Although further data are desirable, the evidence from international correlations, case-control, and cohort studies is reasonably consistent in support of an association between consumption of fat-containing animal products and prostate cancer incidence, particularly with advanced prostate cancer.

Only gold members can continue reading. Log In or Register to continue

Stay updated, free articles. Join our Telegram channel

Jul 27, 2016 | Posted by in PUBLIC HEALTH AND EPIDEMIOLOGY | Comments Off on Epidemiology of Diet and Cancer Risk1

Full access? Get Clinical Tree

Get Clinical Tree app for offline access