Epidemiologic studies of nutrition and cancer: Let us not throw out the baby with the bath water
Version of Record online: 22 APR 2005
Copyright © 2005 Wiley-Liss, Inc.
International Journal of Cancer
Volume 116, Issue 5, pages 662–664, 20 September 2005
How to Cite
Kaaks, R. and Riboli, E. (2005), Epidemiologic studies of nutrition and cancer: Let us not throw out the baby with the bath water. Int. J. Cancer, 116: 662–664. doi: 10.1002/ijc.21027
- Issue online: 20 JUL 2005
- Version of Record online: 22 APR 2005
Many cancers, including those of the large bowel, pancreas and gall bladder, breast, ovary and endometrium, have up to 10-fold higher incidence rates in industrially developed, affluent societies compared to less affluent parts of the world. Studies on migrants from low- to high-income countries and studies on temporal trends in incidence rates within the affluent countries have clearly shown that these increased incidence rates are due to behavioral and environmental (i.e., nongenetic) factors, which most likely include nutrition, but also use of tobacco and alcohol, childbearing patterns and viral infections. Descriptive studies have shown a strong relationship between economical development and the average composition of the diet, which, with increasing wealth, becomes increasingly rich in total and saturated (animal) fats, animal protein, sugar and other refined and rapidly digestible carbohydrates, and increasingly deplete of whole-grain cereals and dietary fiber. In international comparisons, these various changes in diet were all found to be strongly correlated with age-standardized incidence and/or mortality rates of cancers of the breast, colon, pancreas, endometrium and other tumors frequent in Western countries. These and other observations have led to speculations that up to 70% of cancer in Western societies might be preventable by adopting a healthier nutritional lifestyle, as well as to an avalanche of epidemiologic research in which it was attempted to relate cancer risk to specific aspects of diet and nutrition at the level of individuals. In a mini review in this issue of journal, Dr. Karin Michels draws up a balance of this research so far and also briefly discusses some major methodologic difficulties (possible biases in studies with retrospective design, errors in dietary intake assessments, possible confounding by other lifestyle variables) that complicate epidemiologic studies of diet and cancer.1 Taking the evidence from (some selected) prospective cohort studies and the randomized trials (the ATBC study) as final touchstones of evidence of diet-cancer relationships (or the lack of such of relationships), Michels concludes that, based on the most recent evidence, “very few foods or nutrients remain candidates for cancer prediction or prevention,” although she acknowledges that for a few dietary risk factors (e.g., red meat vs. colorectal cancer; alcohol vs. cancers of the gastrointestinal tract, liver and breast), as well as for excess body weight and lack of physical activity, there remains fairly consistent evidence for an association with cancer risk.
Michels' critical analysis expresses a certain degree of disenchantment with the diet and cancer research area. To some extent, this disenchantment is understandable. Still only 7 years ago, in 1997, the World Cancer Research Fund (WCRF) published an extensive and thorough review prepared by an international panel of experts on “food, nutrition and the prevention of cancer.”2 One of the strongest conclusions of this review was that the consumption of fruits and vegetables protected against a wide variety of cancers, including mouth and pharynx, esophagus, stomach, large bowel, lung and possibly even breast and pancreas. Other major conclusions were that alcohol consumption increased risk for a number of cancers (upper aerodigestive tract, breast), that weight control and regular physical activity were important for the prevention of several frequent cancers (colon, breast, endometrium, kidney) and that the use of refrigeration as a means to preserve foods had most likely contributed to the decrease in gastric cancer incidence in many industrially developed countries. More recent evaluations, for example by international review panels that convened at the International Agency for Research on Cancer, confirmed the importance of weight control and physical activity in cancer prevention,3 but concluded that there was only limited evidence for a cancer-preventive effect of fruit and vegetable consumption against cancers of the mouth and pharynx, esophagus, stomach, colon-rectum, larynx, lung, ovary, bladder and kidney, and inadequate evidence for all other cancer sites.4 This relative weakening of the evidence regarding the potentially preventive effects of fruit and vegetable consumption, compared to the earlier report by WCRF, was related to the fact that more recent epidemiologic evidence from prospective cohort studies did not all confirm earlier observations, mostly from retrospective case-control studies, in which generally a strong inverse relationship of fruit and vegetable consumption with cancer risk had been observed. Apart from fruits and vegetables, another aspect of diet for which there has been conflicting evidence between case-control and prospective cohort studies is that relating total dietary fat intake to breast cancer risk,5 although a recent meta-analysis showed quite a similar summary relative risk estimates for relationships of total and saturated fat intakes with breast cancer risk in both case-control and prospective studies.6
Because of contrasting findings between case-control and prospective cohort studies, observations from retrospective case-control studies on diet and cancer are increasingly being dismissed for being potentially biased by differential errors between patients with disease and control subjects in recalls of past diet. This concern has been one of the major arguments to start large prospective cohort studies.7 However, differences between the findings from prospective cohort studies and retrospective case-control studies (e.g., with regard to fruit and vegetable consumption and cancer risk) do not by themselves prove that the results from case-control studies must have all been wrong. First of all, there are also examples of diet-disease relationships (e.g., breast cancer in relation to alcohol consumption or saturated fat intakes) where case-control and prospective cohort studies have led to very similar findings.6, 8 Second, comparisons between retrospective and prospective analyses within a study cohort, such as the one cited by Michels9 have been very few and some have not shown meaningful differences between the 2 types of study design.10 Finally, prospective studies have often used relatively short and self-administered dietary assessment questionnaires that may have been somewhat less accurate than the more extensive and/or interview-administered questionnaires that have been often used in retrospective case-control settings.
Besides expressing reservations against case-control studies, which we only partially share, Michels concludes that prospective cohort studies have also mostly failed to identify important dietary risk factors for cancer. We do not agree with the latter conclusion. In fact, findings from prospective cohort studies on diet and cancer have not all been as negative as Michels' review might suggest, and we feel that she bases her argument on a rather selective citation of previous study results. For example, while some prospective cohort studies are cited in which there was a lack of inverse relationship of colorectal cancer risk with intake levels of dietary fiber, other and no less important studies that did show such inverse relationship were ignored.11, 12 The other negative examples given (about absence of an inverse relationship of breast cancer risk with vegetable consumption, or of lung cancer risk with intakes of fats) have in fact never been the strongest dietary hypotheses with regard to these 2 cancer sites. By contrast, no mention was made of an extension of a large pooled analysis of prospective cohort studies on dietary fats and breast cancer risk, including a total of about 7,000 incident cases of breast cancer, which showed an increase in breast cancer risk among women consuming increased amounts of saturated fats as a percent of total energy intake (an old hypothesis!).8 Preliminary analyses within the European Prospective Investigation into Cancer (EPIC),7 also with close to 7,000 incident cases of breast cancer, entirely confirm these findings (unpublished observations). For lung cancer, recent meta-analyses of epidemiologic studies of both case-control and prospective cohort design showed a modest reduction in risk of about 15% among men or women who had comparatively elevated consumption levels of fruits,13 and this, too, was confirmed in the EPIC study.14 Various prospective cohort studies have also shown inverse relationships of lung cancer risk with dietary intakes or serum levels of carotenoids (markers of fruit and vegetable intake).15 The fact, cited by Michels, that in the β-carotene intervention arm of the randomized ATBC trial lung cancer risk was actually found to be increased among male smokers, contrary to the expectations of a reduced risk, does not invalidate previous studies showing an inverse relationship with estimated intakes or plasma levels of β-carotene, but merely indicates the dangers of overinterpretation and extrapolation of these findings, i.e., attributing the protection against lung cancer to β-carotene itself, and assuming that high pharmacologic doses of β-carotene (as those given in the ATBC trial) will be potentially even more beneficial than the amounts normally consumed from diet. Other important observations from prospective cohort studies include a protective effect of milk and dairy products against cancers of the large bowel,16 and increased risks of large bowel cancer at elevated intakes of alcohol,17 red meat and processed meats.18
The several diet-cancer relationships observed in prospective studies were not generally very strong (relative risk estimates generally did not exceed 2.0), and even taken together certainly do not explain the 10–70% of cancer incidence originally postulated by Doll and Peto19 to be due to nutritional factors. However, there is little doubt that relative risk estimates are generally grossly underestimated due to random dietary measurement errors. Furthermore, even the true relative risks between the extreme quartiles or quintiles of dietary intake distributions will generally be only of a modest magnitude when studies are performed within relatively homogeneous populations with regard to diet and lifestyles. It is likely that, by combining data from multiple populations with more heterogeneous dietary exposures and cancer incidence rates, stronger associations of diet with cancer risk will be found.7 In addition, there are indications that the use of more accurate dietary assessment methods, e.g., using multiple days of food consumption records instead of a questionnaire, may lead to stronger associations of diet with cancer risk.11 Finally, the term “nutrition” comprises more than just the foods we consume and refers to a metabolic state that integrates also the effects of physical activity and other lifestyle factors such as psychologic stress (e.g., effects on insulin sensitivity), as well as genetically determined responses to these lifestyle factors. Prospective cohort studies in which biologic samples (e.g., blood, urine, saliva) have been collected from cohort participants will allow more in-depth studies on nutritional-metabolic risk factors for cancer, which certainly do include only epigenetics. This type of study has already shown, for example, a 3- to 4-fold increase in the risk of colon cancer in women or men who have elevated plasma insulin levels,20, 21 a metabolic risk factor that is strongly determined by nutrition, including composition of diet, physical activity and energy balance.
Dr. Michels' charge against the area of nutritional cancer epidemiology—depicting the area as close to hopeless in the light of insurmountable methodologic problems and discarding many valid study results—in our view does no justice to the much more important progress that has been made and carries the risk of throwing out the baby with the bath water. Although we certainly recognize that many methodologic difficulties may limit our estimations of nutritional effects on cancer risk, we believe that these can be largely overcome by improved study designs, and in particular by combining large prospective studies in multiple, heterogeneous populations and by increasingly incorporating into future studies the use of biochemical markers of diet and metabolism.
- 2World Cancer Research Fund, American Institute for Cancer Research. Food, nutrition and the prevention of cancer: a global perspective. Washington, DC: American Institute for Cancer Research, 2004; 1–670.
- 3IARC Working Group on the Evaluation of Cancer-Prevention Strategies. IARC handbooks of cancer prevention: weight control and physical activity. Lyon: IARC, 2003; 1–315.
- 4IARC Working Group on the Evaluation of Cancer-Prevention Strategies. IARC handbooks of cancer prevention: fruit and vegetables. Lyon: IARC, 2003; 1–376.
- 15IARC Working Group on the Evaluation of Cancer-Prevention Agents. IARC handbooks of cancer prevention: carotenoids. Lyon: IARC, 1998. 1–326.