Doll and Peto's Quantitative Estimates of Cancer Risks: Holding Generally True for 35 Years

In late 1980, Sir Richard Doll and his colleague Richard Peto (now also Sir Richard) at Oxford University submitted a landmark review of factors known at the time to affect cancer risk ( 1 ). The report had been commissioned by the US Congress Office of Technology Assessment to help determine the percentage of cancer because of avoidable causes, and the 117-page document was published in the Journal of the National Cancer Institute in June 1981. The article became a reference standard when estimates were made of the relative roles of lifestyle vs environmental vs host factors as causes of cancer in the US population. Herein we review the estimates Doll and Peto provided, what new knowledge has accumulated since, and how the proportions of cancer estimated to be attributable to tobacco, diet, occupation, and infection compare now vs then. The Doll-Peto article is one of the few, if only, reports in the Journal’s 75 years that was generated by Congressional request. At the time there was considerable controversy over the burden of cancer associated with exposures to various chemicals in the workplace, air and water, in part because of a report by 10 National Institutes of Health researchers submitted to the Occupational Safety and Health Administration (hereafter referred to as the OSHA report) claiming that a minimum of 20% of all cancer deaths in the United States were because of occupational exposures, with up to 18% of all cancers arising because of asbestos exposure alone ( 2 ). Questions for one of us (WJB) following a seminar on geographic patterns of cancer with Doll’s group in Oxford in the summer of 1979 were not so much on the specific topic presented as on this OSHA report, its authors, and its evidence basis. Doll was the world’s leading epidemiologist of the day, having designed and led some of the original epidemiologic research studies documenting that smoking was the dominant cause of lung cancer and that groups heavily occupationally exposed to asbestos also were at elevated risk of this cancer, but the OSHA report projections seemed too high to the Oxford group. The issue of attribution of risks remained topical, and thus Doll and Peto were invited by congressional representatives to prepare a comprehensive report on the known causes of cancer and the amount of cancer in the United States because of tobacco, occupation, and other known etiologic factors. Below we comment on the specific cancer risk factors described in the Doll-Peto article and the evidence regarding the causes of cancer incidence and mortality then and now. We focus on four major exposures leading to cancer covered by Doll and Peto, namely tobacco (the most important), diet and nutrition (the most uncertain), occupation (at the time the most controversial), and infection (possibly the most looming [3]). We also comment on the role of chance, or bad luck, in the attribution of cancer causes, including the recent report that about two-thirds of a large group of cancers may be because of uncontrollable and naturally occurring random DNA replication errors during cell division ( 4 ). Doll and Peto listed cigarette smoking as the major single preventable cause of cancer in the United States, estimating that cigarette smoking accounted for about 30% of all cancer deaths in 1978. The overall estimate took into account estimated smoking attributions for specific cancers, eg, that 91% of the 71000 lung cancer deaths among men and 77% of the 24000 lung cancer deaths among women were attributable to smoking, and then added the numbers across cancer sites yielding estimates that 43% of all cancer deaths among men and 15% among women (and 30% overall) were because of tobacco smoking. Smoking prevalence in the United States had peaked in the 1960s, but data from the National Health Interview Survey (NHIS) show that approximately 42% of men and 34% of women age 35 to 64 years were smokers in the late 1970s ( 5 ). Smoking rates have since declined, with NHIS data indicating that after 2010 about 23% of men and 19% of women age 35 to 64 years are smokers ( 5 ). With much smaller percentages of smokers now than 35 years ago, the fraction of cancer attributable to smoking might at face value be expected to be lower today. However, the impact of current smoking rates will be expressed more in the future than now, with recent rates of lung and other cancers having been determined primarily by past exposures. Indeed, as summarized in the 2014 Surgeon General’s report “The Health Consequences of Smoking – 50 Years of Progress,” latest estimates are that between 2005 and 2009 among those over age 35 years 84% of lung cancers among men and 81% among women and 35% of all cancers among men and 24% among women (or 29% overall) were because of tobacco smoking ( 5 ). The percentages attributable to smoking are essentially the same as in the Doll-Peto report, despite the decline in smoking prevalence for several reasons, the most important being that the relative increases in cancer risk because of smoking are greater than first suspected ( 6 ). Doll and Peto cited a follow-up of US veterans where there was an 11-fold higher number of lung cancer deaths among men who did vs did not smoke cigarettes at the beginning of follow-up and used similar data from the American Cancer Society’s (ACS’s) Cancer Prevention Study One (CPS-I) study of a million persons in estimating the number of excess deaths because of smoking. Updated ACS data from a second cohort (CPS-II) show relative risk estimates for lung cancer nearly twice as high as in CPS-I, with a greater than 20-fold increase in the risk of this cancer among current smokers compared with never smokers ( 5 ). In addition, the number of cancers classified as smoking-related has increased over time. Doll and Peto listed smoking as a cause of lung, mouth, pharynx, larynx, esophagus, bladder, kidney, and pancreas cancers, but subsequent reviews by the US Surgeon General, as well as other authors, list additional cancers as inducible by smoking, including tumors of the stomach, large bowel, liver, uterine cervix, and acute myeloid leukemia ( 5 , 7 , 8 , 9 ). Lung cancer, however, remains the dominant cancer phenotype accounting for smoking-induced cancer deaths in both the Doll-Peto and recent Surgeon General calculations. Despite 50 years of progress since the first US Surgeon General’s report on Smoking and Health in 1964 ( 10 ), we as a nation are still experiencing a heavy cancer burden directly attributable to tobacco smoking. Policies aimed at smoking cessation have been successful at cutting the percentage of Americans who smoke by more than half, yet there are still over 40 million smokers in the United States ( 11 ). Continued efforts need be devoted to reducing this number, because quitting smoking has been documented to substantially and relatively rapidly reduce the risk of most smoking-related cancers (eg, risk of lung cancer at five to nine years after quitting smoking is only half what it would have been had the smoker continued to smoke [12]). Continued efforts also need be sustained towards preventing individuals from beginning to smoke. The mean age at smoking initiation in the United States is 15 years, with 87% of smokers having a first use of cigarettes before age 18 years ( 5 ), so programs aimed at youth avoidance of cigarettes remain crucial. Doll and Peto pointed out that epidemiological evidence regarding diet and cancer was largely indirect and noted the absence of reliable evidence regarding specific components of diet. After a lengthy review of various aspects of diet, including overnutrition, they “guesstimated” that 35% of cancer could be attributed to diet, but their uncertainty was emphasized by an accompanying wide possible range of 10% to 70%. Like the attributions for tobacco, Doll and Peto made their estimates by cancer site, indicating that perhaps 90% of stomach and colorectal cancers, 50% of endometrium, gallbladder, pancreas, and breast cancers, 20% of lung, upper aerodigestive, bladder, and cervix cancers, and 10% of other cancers may be diet related. Several of the cancer-specific estimates proffered by Doll and Peto are now known to be too high. Helicobacter pylori infection, a major gastric cancer risk factor in the United States and especially in developing countries, likely accounts for well over half of all stomach cancer cases among those infected with the bacteria, but H. pylori had not yet been discovered at the time of the Doll-Peto report ( 13 ). The 90% estimate also appears too high for colorectal cancer. In a recent follow-up of 347000 men and women, among whom over 3700 colorectal cancers were observed, only 16% of the cancers were said to be attributable to a combination of five healthy lifestyle (diet-, physical activity-, and obesity-related) components ( 14 ). Doll and Peto also appeared influenced by the fairly strong observational findings of lower risk of several cancers, including lung cancer, associated with higher serum levels or dietary intake of vitamin A or its precursor beta-carotene ( 15 ). By the mid 1990s, however, two large-scale trials of beta-carotene and beta-carotene/retinol found increased rather than decreased risks of lung cancer following supplementation with these vitamins ( 16 , 17 ). Not only has the hypothesized protection from vitamin A or carotenoids not been confirmed ( 18 , 19 ), but in contrast to numerous case-control studies, large cohort studies have found no overall protective effect for fruit and vegetable intake and cancer mortality ( 20 , 21 ). A possible contribution of fat intake to breast cancer risk was noted by Doll and Peto, but current overall evidence for such an association also is weak ( 22 ), and in one of the few randomized studies to evaluate diet and cancer risk, a protective effect of low-fat diet could not be confirmed ( 23 ). In its comprehensive review of research on diet, nutrition, and risks of specific cancers, the American Institute for Cancer Research (AICR) judged that the only food-cancer association for which evidence is “convincing” is the link between red and processed meat intake and colorectal cancer ( 22 ). Several broad-based composite dietary measures, such as the Healthy Eating Index, have been developed since the Doll-Peto report and have been shown to predict disease risk. Two such measures were significantly related to risk of incident cancer in the Nurses Health and Health Professionals Follow Up studies, but the relative reductions in cancer between the lowest and highest quintiles of the indices were small (6% to 10%), indicating that the attributable fractions would be even smaller ( 24 ). Today, in spite of an exponential increase in the number of papers on the nutritional epidemiology of cancer since 1981, the evidential basis for estimating the percentage of cancer associated with diet largely remains, as summarized by Doll and Peto, “a chronic source of frustration and excitement to epidemiologists.” In addition to the absence of definitive data on cancer risks associated with various diet and nutrition components, considerations of what to include as diet-related factors have tended to vary. Whereas AICR and others have included alcohol intake when examining dietary factors, Doll and Peto considered alcohol as a separate category, estimating that 2% to 4% of the total cancer burden was because of drinking. Evidence linking alcohol consumption and cancer risk has remained fairly consistent over time ( 8 , 25 ), with the one exception being that breast cancer risk is now considered to rise moderately with alcohol intake ( 26 ), so that Doll and Peto’s upper estimate for alcohol-induced cancer likely is still appropriate today. There also is considerably more information available now than over 30 years ago regarding the roles of physical and sedentary activity and obesity in cancer risk and mortality ( 27 , 28 ). Doll and Peto cited ACS CPS-I cohort findings on rising rates of cancer deaths with body weight, and incorporated effects from overnutrition when calculating guesstimates of dietary attributions of specific cancers (with high body weight apparently contributing, eg, to the 50% dietary attributions for endometrial, gallbladder, pancreas, and breast cancers). Doll and Peto made little mention of physical activity in cancer risk, but physical activity (or lack thereof) seems increasingly important as a direct factor in cancer occurrence and an indirect contributor via its influence on the rising prevalence of obesity in the United States over the past few decades ( 29 , 30 ). A recent study evaluated a score measuring diet, physical activity, BMI, and alcohol consumption, and the reported risk estimates suggest that the fraction of total cancer mortality attributable to these factors combined is less than 20% ( 31 ). All things considered, although some specific food items and broad-based dietary patterns, as well as obesity and physical activity, have been convincingly linked to cancer risk, it appears that the percentage of cancer attributable to diet (and diet-related factors) will eventually be found to be closer to 20% than the 35% guesstimated by Doll and Peto ( 32 ). A major motivation for the Doll and Peto article was the claim in the OSHA report that at least 20%, and possibly as much as 38%, of cancers were the result of occupational exposures. After briefly critiquing the OSHA report and pointing out errors leading to overestimation of cancer risk among workers exposed to suspected carcinogens (eg, the OSHA report’s disregard of dose and duration of exposure in predicting risk for exposed workers), Doll and Peto proceeded to estimate the proportion of cancer deaths they thought likely to be explained by occupational exposures. For organs, such as the lung and bladder, affected by exposure to occupational hazards considered definitely to result in increased cancer risk, site-specific estimates by sex were based on a review of the existing epidemiologic literature tempered by their professional judgment. Doll and Peto estimated that such occupationally-induced cancers accounted for about 3% of all cancer deaths, with lung cancer accounting for over two-thirds of deaths attributed to occupation and asbestos being the single agent associated with the largest number of cancer deaths. For organs possibly associated with occupational hazards, they estimated that 1% of cancer deaths among men and 0.5% of cancer deaths among women could be explained by occupational exposures. They concluded that in total 4% of all cancer deaths were explained by occupational exposures and stated that this estimate was unlikely to be off by over a factor of two in either direction. Despite the many uncertainties confronting Doll and Peto in making their estimates based on 1981 information about potential occupational risk factors, their estimates served to temper a growing anxiety at the time about a possible cancer epidemic resulting from occupational exposures. In the nearly 35 years since publication of the Doll and Peto article, both the number of occupational epidemiology papers published and the number of workplace exposures classified by the International Agency for Research on Cancer as known carcinogens have increased greatly. Nonetheless, the Doll and Peto estimate that 4% of US cancer deaths are explained by occupational exposures has stood up well. A recent attempt to estimate the percentage of US cancer deaths attributable to occupation reported an estimated range of 2.4% to 4.8% ( 33 ), and because of reductions in the levels of workplace exposure to known carcinogens resulting from regulatory action and industry concerns, it has been estimated that the percentage of cancer deaths in the United States may be as low as 1% in the future ( 34 ). More recent publications have estimated that the percentage of cancer deaths explained by occupation is 5.3% in Britain ( 35 ) and 2.6% in Sweden ( 36 ). Occasional efforts continue to argue that the percentage of cancers resulting from occupational exposures is considerably larger than that estimated by Doll and Peto. The President’s Cancer Panel published a report which opined that the Doll and Peto estimate of 4% was “woefully out of date,” and further stated the belief that it significantly underestimated the true toll of cancers related to occupation ( 37 ). No effort was made to estimate the percentage of cancer attributable to occupation, but the report detailed a long list of concerns about various possible cancer-causing agents in the workplace and the environment. The report generated a flurry of news articles, but it’s apparent that it was, as Doll and Peto concluded about the OSHA report, written more for political than In addition, two recent reported that the percentage of cancers attributable to environmental exposures, including workplace exposures, is between and 19% ( , ). for this range of estimates is however, and the both to document the to 19% range ( 40 , ) not data the the Doll and Peto estimate that 4% of cancer deaths are explained by occupational exposures still appears to be consistent with the available epidemiologic evidence from the United States and other The of Doll and Peto regarding the percentage of US cancer deaths attributable to infection was made before studies confirmed some important causes of cancer (eg, as the major cause of and before important cancer-causing agents had been (eg, an important cause of cancer, and H. pylori , the major cause of gastric ( 13 , ). had been by 1981 between infection and cancer, infection and and cancer, and and ( , ). Doll and Peto that further causes of cancer were yet to be and so estimated that 10% of US cancer deaths were attributable to infection, with a after the a estimate of 1% and a indicating the upper of the range of estimates to their In the decades since the Doll-Peto report, in addition to the H. , and cancer the has the rise of the epidemic and its associated increase in by as well as more (eg, cancers ( ). also has been documented as a cause of and cancers, with now as many or more as cancers in the United States ( ). Doll and Peto’s estimate that 10% of all cancer deaths are because of infection is for the United States and other developed A recent review concluded that the percentage of cancers attributable to agents is approximately 23% in less developed and in more developed ( ). seems that additional to cancer risk from agents remain to be especially with research on the and the percentage of cancers attributable to infection may ( ). Doll and Peto did not an estimate of to cancer risk, that is not and they did not an estimate of the overall percentage of all cancer deaths attributable to avoidable The of their estimates for tobacco, diet, and the other specific factors considered to Doll and Peto noted that there may be in the estimates between some of the causes considered (eg, that smoking and alcohol to increase risk of and so that the influence of one factor may in reduce the impact of the article, similar at the time ( , ), the that most cancers from the environmental and lifestyle exposures this likely both the of the available evidence and a not to the impact of of cancer risk. Indeed, in a recent Peto ( ) concluded that 50% of all cancer deaths in the United in 2010 could be accounted for by lifestyle and environmental factors. factors also a because there are where host directly to cancer (eg, as with and breast and and because factors influence and to what exposures are the body ( 50 , ). the large of cancers have an cause or are more likely to be the result of was made by a that most of the in cancer risk among the 25 organs or evaluated was explained by or that by random DNA replication errors during cell ( 4 ) rather than causes those considered by Doll and Peto. reported a strong between the number of cell in an and the cancer risk in the United States, with the 2 of leading the to claim that about two-thirds of the in cancer incidence was explained by in cell and only because of avoidable on an risk the classified cancers by affected or as either affected by or or most to random errors during DNA with most of the cancers evaluated in the to a of vs cancers, however, will be and such a may be of value in cancer are in the of a and it is likely that many of the are random but that some are the result of or from environmental The both of random and of with or environmental causes likely increase with total number of cell so that of both would to the between cell and cancer risk. A the relative of random vs environmental or to cancer rates the numbers of cell at various organs or are likely to among and thus even if most in the of cancers are the result of random replication the large geographic in cancer rates at most organs that the percentage of cancers arising by such random errors in the United States is relatively Doll and Peto’s article a and review of the known causes of cancer and the proportions of cancer deaths in the American attributable to these avoidable exposures. estimates have true over the 35 is a rather than a and new information from research will both on the relative contribution of vs and to the of and on the estimates made by Doll and Peto for the continue to as we progress into the 21