How do we judge what causes cancer? the meat controversy.

Abstract

The fact that obesity and the intake of processed meat are considered “carcinogenic to humans” – according to the criteria used by the International Agency for Research on Cancer (IARC) and/or the World Cancer Research Fund (WCRF) – may look disconcerting to many, and calls for an explanation of how the conception of causality has developed in medicine in the last 50 years. In this evolution IARC itself has played a pivotal role, as testified by the recent publication of a volume on its history1, and the long series of Monographs on the carcinogenicity of specified agents to humans2. The gold standard of causality in medicine since the 1940s has been the randomized clinical trial (RCT), used to test pharmaceutical drugs and other interventions intended to be beneficial. The emphasis of RCTs is on single agents as causal factors (medical treatments or otherwise “actionable” agents). This can be aligned with the microbiological tradition, in particular the criteria for the identification of pathogenic micro-organisms developed by Henle and Koch in the 18th and 19th centuries. According to these criteria, an agent was a cause of disease if it was always present in the diseased individuals, it was always absent in the healthy subjects and the disease could be reproduced by inoculation in animals. The first RCTs were on antibiotics, i.e. antimicrobial agents, to affirm the relationship between the successes of microbiology and the development of a powerful strategy to investigate drugs. Also the early years of research on chemical carcinogenesis entertained the idea of most cancer cases being attributable to single causal agents. That was, however, abandoned already in the 1950s mainly due to epidemiological studies demonstrating then-inexplicable marked differences in cancer incidence between communities worldwide. The idea of a single causal agent, applicable perhaps to smoking and a few like agents, may still be at the heart of community thinking about cancer causation generally, and may be at the basis of recent misunderstandings concerning the classification of processed meat. When tobacco smoking was found to be carcinogenic by sound epidemiological research in the 1950s, the tobacco industry asked for more than evidence of an association, specifying RCTs as the gold standard. Of course RCTs could not be undertaken for ethical reasons (we cannot knowingly expose humans to putative hazardous agents). To develop a sound response to this claim from the tobacco industry, the English statistician Sir Austin Bradford Hill modified Henle-Koch's criteria to apply them to non-infectious diseases such as cancer. The guidelines developed by Bradford Hill - and used in a modified version also in the US Surgeon General reports on tobacco smoking – date back to the English philosophy of science developed in particular by John Stuart Mill and founded on concepts such as consistency of observations in different populations, strength of the statistical association between exposure and risk of disease, the presence of a quantitative relationship between level of exposure and quantification of the risk and the biological plausibility of a causal association (based for example on knowledge of intermediate mechanisms or work in animals and in vitro systems). It is with similar criteria that IARC Working Groups have come to the conclusion that chemicals or other exposures are “carcinogenic to humans”. While for many of these the causal nature of the association with cancer is obvious because of the strong increase in risk, for example asbestos and mesothelioma or aromatic amines and bladder cancer, for other exposures the conclusions of IARC have been more controversial, usually showing that the critics have little understanding of current concepts of causality in medicine. One of the Henle-Koch's criteria was the idea of necessary cause, meaning that the disease will not appear in those unexposed to the agent in question: for example, the Ebola virus is necessary to develop the disease known as Ebola. However, the public usually does not appreciate that cancer and other non-infectious diseases are often due to a combination of factors, none of which is sufficient nor necessary. Cancer is multifactorial, that is it arises because exposure to several agents together with certain biological processes (including for example DNA damage or cell proliferation as explained below) constitute a pathway leading to malignancy. Multiple agents are involved almost by definition. Being obese or eating red meat do not necessarily lead to cancer, and in fact the causal association with cancer is weak (meaning a mathematically small increase in risk) but well established. We know only one necessary cause for cancer (i.e. an agent without which a certain type of cancer would not occur), that not by chance is an infectious agent, human papilloma virus (HPV) for cervical cancer (in this particular case we also have the equivalent of RCT evidence through vaccination). For non-infectious diseases, the vast majority of known causal agents belong to the category of non-necessary causes, such as tobacco smoking, asbestos, high blood pressure, high cholesterol and so on. Thus everyone is aware when a lifelong never smoker is diagnosed with lung cancer. Just to give some figures, the lifetime risk of lung cancer in a non-smoker is 1% and in a heavy smoker it is 25%; the lifetime risk of myocardial infarction or other cardiovascular diseases in people without other risk factors but with a cholesterol/HDL ratio greater than 6 is 19%, compared to 16% in those with a normal ratio (up to the age of 75); the risk of colorectal cancer in frequent red meat eaters is 6% vs. 5% in moderate eaters. Following the historical evolution described above, cancer causation is now understood as the complex interplay of numerous agents and biological response patterns that act along distinct mechanistic pathways, thus creating the opportunity for several steps critical to the carcinogenic process to be affected. Also “weak” risk factors, i.e. those for which the evidence of causal association with cancer is clear but the increase in risk is 20-50% rather than 10-fold (or 100-500-fold like for HPV), can contribute to the process if they are in combination with other agents. This is the case for example of processed meat, that increases the risk of colorectal cancer by 18% for every 50 grams of intake per day (that gives the difference between 5% and 6% reported above). This is well established, but the absolute increase in risk is low. The public (and the media) should be sufficiently informed as to recognize the difference between degree of evidence (strong for processed meat) and strength of the risk increase (modest for 50 grams of processed meat per day). This concept is well-known in clinical practice: response to therapies is much more complex than we thought 50 years ago (when antibiotics dramatically saved lives), and response of patients depends upon pathways and networks that include co-morbidities and individual susceptibility. Therefore, a drug can definitely improve survival or well-being (the evidence may be strong) only in some – but not all – patients. Thus, in both etiologic (ie. causation-based) research and clinical medicine we acknowledge that an occurrence in a single patient can be due to the concomitant action of several component “causes” within a causal complex. A direct and immediate overlap between recognized causes of cancer and single agents identified as carcinogens is challenged by current understanding of cancer causation. Complex exposures or behaviours can be recognized as posing a carcinogenic risk in the absence of the isolation of the single specific causal agent. The IARC Monograph evaluations have addressed circumstances of exposure in which the discrete carcinogenic agent(s) have not been identified, a situation exemplified by occupational exposure as a painter. The WCRF specifies that, for example, ‘The evidence that greater body fatness is a cause of colorectal cancer is convincing’: a cause of cancer identified in the absence of any reference to a relevant carcinogen(s). Only a proportion of persons exposed to a given level of carcinogen for a specified period develop cancer, corresponding to the concept that single carcinogens are usually “insufficient “causes. Similarly, with some exceptions, only a proportion of patients respond to a drug otherwise recognized as effective. Reasons for such circumstances are largely unknown, but include variation in biological susceptibility and more broadly defined characteristics including socio-economic status and race, in addition to the impact of other carcinogenic exposures. Considered in relation to the sequence of events leading to cancer development, genetic determination of susceptibility may be characterized as operating prior to carcinogen exposure. Epidemiology is currently based on a model, popularized by Kenneth Rothman's use of “pie” diagrams to illustrate sufficient causal complexes resulting from individual insufficient components3. For example, lung cancer in an individual may be explained by a “pie” made of smoking, plus past exposure to asbestos plus inherited susceptibility related to variants in DNA repair genes: none of these is sufficient in itself but their complex can in fact be sufficient to cause cancer in a single individual. No single component is generally necessary, i.e. it will not be found in all causal complexes (pies); but it is necessary in the single “pie” insofar as its elimination will make that complex ineffective and thus prevent disease in that particular individual. In this context it is clear that also “weak” risk factors can be recognized as contributing causes. Consideration of biological responses (‘mechanisms’) provoked by agents specified as causing cancer in humans, that is Group 1 agents in the IARC Monographs parlance, indicates a range of parameters encompassing, but well beyond, the widespread recognition of DNA-damaging carcinogens. They include the ability of an agent to: (1) be metabolized or broken down to a reactive intermediate; (2) damage DNA; (3) alter DNA repair or increase the occurrence of mutations; (4) induce heritable alterations apart from mutations; (5) induce oxidative stress; (6) induce chronic inflammation; (7) be immunosuppressive; (8) modulate endocrine-related pathways; (9) cause cells to divide indefinitely rather than for a finite number of times; and (10) alter cell proliferation, cell death, or nutrient supply4. Many, but not all of these characteristics involve patterns of response able to be identified with reference to a single cell. The mechanisms by which obesity is thought to mediate cancer development involve multi-cellular or tissue-based processes: perturbation of sex hormone metabolism, of insulin and related intra-cellular signaling, and effects due to hormone-like pathophysiology and inflammation. In the case of red meat there is evidence of genotoxic activity of carcinogens formed during cooking practices (polycyclic aromatic hydrocarbons, heterocyclic aromatic amines) or via the reaction of heme iron with nitrite (sometimes generated from nitrate) to form nitroso compounds; but also others among the mechanisms listed above are at work. The difference between single agents able to act on single cells, perhaps through a relatively simple pathway perturbation, and exposures such as obesity - that act indirectly via a number of multicellular pathways - explains why it has been easier to establish the carcinogenicity of single agents in the past. Knowing that a particular tumour type is caused by a specified carcinogen immediately reveals the potential for cancer prevention. From a global perspective, the highest priorities for cancer prevention are known: tobacco control, eliminating infection by hepatitis B and HPV, reversing trends in overweight/obesity and sedentary lifestyle, drinking less alcohol, and avoiding exposure to occupational and polluting carcinogens. Some, but not all of these measures involve exposure to an agent readily identified as a carcinogen. The decreasing incidence of lung cancer in many high-income countries epitomizes the relevance and centrality of cancer epidemiology to cancer control: an outcome achieved despite the intransigence of the tobacco industry. The identification of tobacco smoke as carcinogenic to humans ultimately encompassed both mechanistic data and epidemiological data including the decline in lung cancer after smoking cessation as well as lung cancer attributable to second hand smoke. Though understood to be a modifiable risk factor, consideration of obesity or red meat in this context clearly indicates greater complexity compared to occupational carcinogens or tobacco. The causal association of obesity with cancer at several sites is established unequivocally and the implicit preventive potential is recognized as a public health imperative. The inter-related risk factors of high caloric diet, overweight/obesity and physical inactivity intuitively suggest cancer prevention through dietary modification, weight control and exercise. If relevant change can be achieved, reduced cancer rates may be anticipated or implied as a matter of common sense. In fact, evidence for the efficacy of measures such as dietary change, albeit with reference to trial-based populations, is becoming available. For example, adoption of the Mediterranean diet has contributed to the primary prevention of breast cancer in some studies. In the case of red meat, two RCTs exemplify intervention: the Women's Health Initiative and the Polyp Prevention Trial (PPT, focused on polyp recurrence as outcome).5, 6 In the first a reduction in meat intake was included in more complex dietary and behavioural changes, so that meat intake as such cannot easily be evaluated. In PPT the reduction of processed meat intake in the course of the intervention was low for processed meat, and the follow-up was only 4 years. Telling people what causes cancer and how to avoid it is not only a public health imperative; the information is often identified with a ‘right to know’. Concerning the disease which is feared most, the community demands to be informed, relying in large part on relevant statutory authorities and trusted non-government agencies. Media reports about risk of cancer causation, commonly characterized as ‘cancer scares’, usually involve the potential for exposure to a recognized carcinogen(s) through its presence in food or a consumer product. Reporting of such exposure, and an implied risk of cancer, can be justified by reference to the principle summarized as ‘no safe level of exposure to a known carcinogen’. There is however confusion about the degree of evidence available, the strength of association with risk (“strong” and “weak” carcinogens), and between single carcinogenic agents and complex exposures and behaviours. Some causes of cancer, such as smoking or exposure to asbestos, are easily recognized. Others not clearly supported by scientific evidence, such as stress or pesticide residues in food, may be accorded recognition by the media ahead of, for example, obesity and alcohol drinking. In relation to single chemical carcinogens, widespread understanding of ‘no safe dose’ serves the community well, particularly when necessary distinction is made between regulations preventing exposure to certain carcinogens at low level and lifestyle changes warranted to prevent an evident burden of cancer. In the case of complex exposures and behaviours such as obesity or meat intake, the degree of evidence is strong but they cannot be treated as single carcinogens are, for example because the concept of “no safe dose” cannot be used for them. Though prevention is certainly necessary for these, translation of science into regulation is much more complex. We believe that the current procedures used by IARC are scientifically sound. However, a very clear distinction should be made between a “carcinogen”, that entails “no safe dose” and agents or behaviours that are “causes of cancer” such as obesity or red meat, where the concept of no safe dose is meaningless and the evaluations of carcinogenicity need to be put in context. These complex issues may be addressed when the Monograph Preamble, which specifies procedure and terminology, is next revised. Paolo Vineis and Bernard W. Stewart Paolo Vineis is Professor of Epidemiology at Imperial College, London Bernard Stewart is Professor of Medicine at the University of New South Wales, Sydney