A healthy diet is an important factor in the prevention of many acute and chronic diseases, including diabetes, cardiovascular disease and cancer. Probably the most important dietary contribution to cancer prevention is maintaining a healthy body weight. While dietary composition may play a role for cancer etiology, only a few foods and nutrients have been consistently linked to some specific cancers. The general lack of association between specific dietary components and cancer identified in epidemiologic studies is likely due to modest true associations in our study populations that we are unable to detect because of measurement error in diet assessment and to a lack of data on more extreme dietary choices.

In their commentary,1 Drs. Rudolf Kaaks and Elio Riboli agree with my observation elucidated in my recent review2 that energy balance may be the most important nutritional predictor for cancer development. They also agree with my assertion that the frequently cited estimation by Doll and Peto3 that 10–70% of cancer deaths in the United States might be attributable to diet may be overestimated; however, if we consider energy balance as part of the equation, Doll and Peto's evaluation may not be so far off. Kaaks and Riboli, both employees of the nutrition branch of the WHO's International Agency for Research on Cancer (IARC), cite support for a recent relative weakening of the evidence from their own institution regarding the potentially cancer-preventing properties of fruit and vegetable consumption.4 In their attempt to unearth remaining dietary candidates for cancer prevention, they focus largely on the same examples referenced in my review: red meat consumption and colorectal cancer and alcohol consumption and a number of cancers.

Kaaks and Riboli turn to descriptive studies and case-control studies in their hunt for dietary associations. These types of studies, however, do not lend themselves to causal inference about diet and cancer. International correlations are confounded by a variety of factors, including the gross national product, behavioral factors such as sedentary lifestyle and obesity, as well as levels of environmental toxins, to name a few. Migrant studies suffer from a similar potential for confounding by changes in many aspects of lifestyle and environmental exposures. Case-control studies of diet are susceptible to selection and recall bias that renders them largely uninterpretable.

In their defense of case-control studies of diet and cancer, Kaaks and Riboli argue that the results of prospective cohort studies and retrospective case-control studies are not always inconsistent, citing the example of fat intake and the risk of breast cancer. It is interesting that early case-control studies indicated a considerable positive link between high fat intake and the risk of breast cancer,5, 6 whereas a recent meta-analysis including 45 studies could not confirm these results7 and was largely compatible with the pooled analysis of cohort studies.8, 9 This variation in results from case-control studies on diet confirms the need to exercise caution in interpreting them.

So, is there a baby in the bath water? Kaaks and Riboli argue that I may not have done justice to the evidence on diet and cancer incidence from prospective cohort studies. They support their claim for an inverse association between fiber intake and colorectal cancer incidence with one study on fat intake and breast cancer incidence10 and another study on fiber intake and adenomas11—an entirely different pair of shoes. Any support for a link between fiber intake and colorectal cancer incidence and between fat intake and breast cancer incidence remains elusive despite what Kaaks and Riboli may want us to believe. Finally, the charge put forth by Kaaks and Riboli that prospective studies often employ less accurate diet assessment questionnaires than case-control studies has been disputed previously.12

If there are other important associations between dietary choices and cancer incidence, why are we not able to identify them? Kaaks and Riboli argue that relative risk estimates are “generally grossly underestimated” due to measurement error. Indeed, although the currently used diet assessment instruments are crude, they were able to identify associations between nutrition and cardiovascular disease and diabetes. While the role of diet may be even more important for the etiology of these diseases than quantified in observational research, our tools seem to be sufficient to identify important associations if they exist. In cancer epidemiology the yield has been meager.

The generally small or modest associations between diet and cancer that we do observe could be the result of confounding by other lifestyle factors. Although we might indeed underestimate associations obtained in observational studies due to nondifferential measurement error in the assessment of diet, confounding can create spurious associations and is a possible explanation for modest associations that we should not too readily dismiss. Individuals who consume an abundance of fruits and vegetables, choose whole grains and avoid animal products likely follow a healthy lifestyle in general and thus differ in many other aspects from individuals whose diet consists largely of meat, saturated and trans fats, refined carbohydrates and sugars. The suggestion put forth by Kaaks and Riboli that more heterogeneous study populations may provide more valid answers in nutritional cancer epidemiology is a mixed blessing: while we might increase variation in dietary choices, we also enhance the probability of confounding by other lifestyle factors. Small associations that are easily compatible with confounding or other biases do not lend themselves to public health interventions.

I am hardly disenchanted with the area of nutritional cancer epidemiology, as Kaaks and Riboli claim. In fact, I have pointed out a number of areas that remain little explored and that hold promise for further research. These include studies of more extreme nutrition regimens, such as diets consisting of largely raw foods and freshly ground whole grains, explorations of the effects of nutrition throughout the entire lifespan, and dietary modifications of gene expression. Finally, we may want to explore reasons beyond measurement error for our inability to confirm an important cancer-preventive effect of some foods that we previously assumed on the basis of their chemical or biologic properties. Currently used methods to assess dietary intake in observational studies allow only limited explorations of the role of diet to maintain or regain health. Diet assessment instruments such as the FFQ, the 7-day diary, or the 24-hour recall focus on the type of foods consumed, maybe the frequency of consumption and the serving size, but they generally do not capture whether foods were consumed raw or cooked, the preparation of foods, the freshness or storage condition of foods, the use of agricultural pesticides and fertilizers, how and where the food was grown, or the breeding conditions, medications and hormones given to animals providing meat, milk and eggs, all of which may have altered the nutritional value of many foods. Earlier observations, unfortunately largely forgotten today, have linked the consumption of refined carbohydrates and canned and packaged foods to diseases with high incidence in developed countries, including susceptibility to tuberculosis and dental decay,13 the consumption of cooked food and heated milk to bone degeneration and reproductive problems,14 the consumption of whole grains and whole foods to achieving optimal health15 and the consumption of raw and living foods to the successful treatment of previously hopeless cases of gastrointestinal ulcers, diabetes, arthritis, gout and multiple sclerosis.16 It is likely that we are unable to identify many aspects of diet that are associated with health and disease due to the limitations of current diet assessment methods. Refining these methods may improve our understanding of the role of diet and allow us to better define optimal nutrition to maintain health. Furthermore, evidence is accumulating that diet during early life may play an important role in the etiology of some cancers.17 Explorations of critical time periods of heightened susceptibility may provide valuable insights into the role of diet in carcinogenesis.


  1. Top of page
  2. References