Tobacco Documents Online

1202 Doll and Peto currence is nearly always recorded, at least in middle age, for it is nearly always fatal. It is much more difficult to be sure about changes in the incidence of many other types of cancer. The common basal cell carcinomas of the skin, for example, are also easy to: diagnose but are often not registered at all~ as they seldom cause death~ and may be treated effectively outside the hospital. What appears to be a change in incidence may, therefore, be a change only in the completeness of registration. Cancer of the pancreas, by contrast, i's almost always fatall but is easily misdiag- nosed, perhaps as cancer of some other organ, unless it is specially looked for. What appears to be an increased incidence may, therefore, be wholly or partly due to improvements im diagnosis, in the availability of medi- cal services, or (as for all other types of cancer) in the readiness of physicians to inform cancer registries of any cancers they find! Such changes are particularly, likely to affect the cancer incidence rates recorded for people over 65 years of age; as many terminally ill old people used not to be intensively investigated (some- times, it must be admitted, to their advantage). As most cancers are commoner among the old than among the young, these spurious changes in ol& age are liable to distort overall rates quite considerably and' (if attention is not restricted to people under 65 years of age) may conceal a stable or even a decreasing incidence at younger ages at which cancer has been reasonably well diagnosedl for several decades. Despite these difficulties, some changes during periods when no large improvements in relevant diagnostic tech- nology were introduced have been so gross that there can be no doubt about their reality. These changes include the increase in esophageal cancer in the black populatiom of South Africa, the continued increase in lung cancer throughout most of the world, the increase in mesothelh'oma of the pleura in males in indus- trirali¢ed' countries, and the decrease in cancer of the tongue in Britain and in cancers of the cervix ute6 and~ stomach throughout Western Europe and North Amer- ica. Worldwide changes in the mortality attributed to cancers of the lung and stomach in the last 25 years are given in table 5. Detailed U.S. data for these and many, other types of cancer are discussed in section 4.1 and im appendixes C, D, and E. 2.4 Identitication of Causes The simplest evidence of the preventability of cancer would be the demonstration by scientific experiment that a particular action actually leads to a reduction im the incidcnce of the disease. Even .+,here such evidence could in principle have been sought by means of randomized! trials, this has not in general been done; and so we often have to be content with the type of strong circumstantial evidence that would be sufficient to obtain a conviction in a court of law. Action, based on such evidence, has in practice oftvn been followed by the desired result-for example, a reduction in the incidence of bladder c:utcer in the chc•ntical industry has bec-n seen since stoplring the ntanulacture and use TABLE 5.-International changes since 1950 in death certification rates for cancers of stomach and hrng Cou ntry Period Percent change in mortality' from, cancer of: Stomach Lung Australia 1950-51 to: 1975 -53 +146 Austria 1952-53 to : 1976 -53 -8 Chile 1950-51 to 1975 -56 +38 Denmark 1952-53 to ~1976 -62 +87 England and Wales 1950-51 to 1975 -49 +33 West Germany 1952-53 to ~1975 -50 +36 Ireland 1950-51 to~1975 -54 +177 Israel 1950-51 to 1975 -49 +58 Japan 1950-51 to 1976 -37 +408 The Netherlands 1950-51to 1976 -60 +89 New Zealand 1950-51 to 1975 -54 +137 Norway 1952-53 to 1975 -59 +118 Scotland 1950-51 to ~ 1975 -46 +44 Switzerland 1952-53 to 1976 -64 +72 United States 1950-51to 1975 -61 +148 ° Average of a and 9 rates at ages 35-84 yr. standardize& for age as in IARC' (1976). of 2-naphthylamine, while the progressive increase in lung cancer risk that regular cigarette smokers suffer is avoided byy people who give up the habit of smoking. Cancer research workers throughout the world have therefore accepted that the type of human evidence that has been obtained, sometimes but by no means invari'r ably (see section 4.2)~combined with laboratory evidence that some suspect agent is carcinogenic in animals, is strong enough to j,ustify the conclusion that a means of avoiding some cases of human cancer has been iden- tified. There are, of course, many borderline instances. where reasonable differences of opinion exist, while even for the well-established causes a few critics can always be found who will argue that causality is not established. A majorityy of students of the subject are agreed that a few d'ozen agents or circumstances have al'ready been shown to cause or prevent cancer in humans and that, in a number of other instances, the conditions that give rise to an increased incidence of cancer have been cl'osely defined without a specific agent having yet been identified (IARC Working Group, 1980). These agents an& conditions are listed in table 6. Exposure to some agents, it will be noted, has been on only a small scale, as in the case of a drug introduced briefly for the treatment of a rare disease, whereas exposure to others has been intensive and widespread, and hundreds of thousands of cancers have been caused each year. The extent to which these liste& agents and conditions are now affecting the incidence of cancer in the United States is discussed in Section 5. 2.5 Role of Genetic Factors, Luck, and Age Some people of a given age will: develop cancer in the near future,, and some will not. The deterntinants of'who will and who will non develop cancer are best tlividt•d into three catef;ories„ not only thc usual J"(4 Vt,t.. fk,: NO 6: JU:.t, t9Ht

Avoidable Risks of Cancer In the U.S. 1203 TABLE 6.-Established human carcinogenic agents and eireumstanees°'D Agent or circumstance Exposure' Occupa- Medical Social tional Aflatoxin Alcoholic drinks Alkylating agents: Cyclophosphamide Melphalan Aromatic amines:, 4-Aminodiphenyl + Benzidine + 2-Naphthylamine + Arsenic°' + Asbestos + Benzene + Bis(chloromethyl) ether + Busulphan Cadmium° + Chewing (betel, tobacco, lime) Chromium + Chlornaphazine. Furniture manufacture (hardwood) + Immunosuppressive drugs Ionizing radiations' + Isopropyl alcohol manufacture + Leather goods manufacture + Mustard gas + Nickeld + Estrogens: Unopposed Transplacental (DES )~ Overnutrition (causing obesity) Phenacetin Polycyclic hydrocarbons + Reproductive history: Late age at 1st pregnancy Zero or low parity Parasites: Schislosoma haematobium Chlonorchis sinensis Sexual promiscuity Steroids: Anabolic (oxymetholone). Contraceptives Tobacco smoking UV light + Vinyl chloride + Virus (hepatitis B) + Liver + Site of cancer Mouth, pharynx,Jarynx, eosphagus, liver + Bladder + Marrow Bladder u k + Skin, lung Lung, pleura, peritoneum Marrow Lung + Marrow Prostate + Mouth Lung + Bladder Nasal sinuses + Reticuloendothelial system + Marrow and probably all other sites Nasal sinuses Nasal sinuses Larynx, lung Nasal sinuses, lung + Endometrium Vagina + Endometrium, gallbladder Kidney (pelvis) Skin, scrotum, lung + + + + Breast + Ovary + Bladder + Liver (cholangioma) + Cervix uteri + Liver + Liver (hamartoma) + Mouth, pharynx, larynx, lung, esophagus, bladder + Skin, lip Liver (angiosarcoma) + Liver (hepatoma) ° Expanded from IARC working group, 1980. b By restricting this table to firmly established causes, we undoubtedly have omitted some of the more important determinants of human cancer. (A few borderline cases might not command uniform agreement; e.g., we have on balance just included cadtnium and just excluded beryllium.). ` A plus sign indicates that evidence of carcinogenicity was obtained& d Certain compounds or oxidation states only. ` For example, from X-rays, thorium, thorotrast, some underground mining, and other occupations. ture" and "nurture" but alAo "luck,"' or the play of chance. "Nacurc" relittts to a person's genetic makeup at conception, and this cerrtainJv affects the risk of some types of cancer. For exarnple, other things being t•yuul!, a whitc-skinned person isanore likc•ly, to c)evelcup skim canrcr in resptrnst to sunli~;ht tiran is a black- skinnect l.>crson• whiiY• 1>t•crple• who havc inhe•rited xero- derma pigmentosum, a very rare genetically determined inability to repair the normal effects of sunlight on the skin (:Robbins et al., 1974), are likely to develop several skin cancers per person. "Nurture," which is the subject of this whole report, relates to what people do or have done to them (in the womb; in childhood, or in adult life) and is of public interest as a determinant of cancer J.Nct. vot-, 66. NO. u, JUNE t9ar

1204 Doll and Peto risk because it is the only thing that can be influenced by personal or politicali choice.' Finally, "luck" takes care of the remaining differences in outcome that both observation and theory lead us to expect (Peto, 1J77b)i perhaps by determining the con- catenation of events that brings about specific changes in particular molecules in individual cells at particular times. Somewhat similarly, luck involves some of us but not others in traffic accidents. Even among genetically identical laboratory animals kept under conditions that are as closely uniform as possible, some will die of'f cancer in middle age, while others will live on into old age with no cancer. (t?.nalogously, the fact that some people die of'. lung cancer at 40 years of age while other people live on in apparently similar circumstances to 80 does not of ilself provide any suggestion at all as to whether or not there are any genetic factors which affecr lung cancer risks, for variation in age at onset of disease would be expected in either case.) Nature and nurture affect the probability that each individual will develop cancer, and luck then deter- mines exactly which~ individuals will actually do so. However, although for each single individual the role of luck is enormous, in a: population of a hundred thousand or more (e.g., the population covered by one particular cancer registry) the role of luck is smaller, and in determining the annual number of cancers in the whole United States Ittck has a completely neg- ligible effect, for the larger the population the more the good and bad luck will tend to average out. Conse- quently, in the comparison of national cancer rates only nature and nurture are important~ Much of the evidence outlined above (changes of cancer incidence with migration, changes over the decades within one country;, and' the identification~ of particular causes of cancer) points to an important role for "nurture." However, this does not deny an equally important role for "nature." For example, the stomach cancer risks in certain countries differ markedly from each other, and most are decreasing rapidly (table 5), both of which observations point to the relevance of nurture. How- ever, in both high-risk and low-risk countries people whose "ABO" blood group (a factor that is determined purely genetically) is of type "A" have a stomach cancer risk some 209'o greater than that of their com- patriots of type "0." In this instance, as for skin cancer, nature and nurture seem to multiply each ' One difficulty of terminology with the disutxtion between nature and nurture is where to classify a genetically inherited tendency to behave in certaimm ways (e.g., to overeat or undereat): From a public health point of view it is probably most appropriate to attribute the net results of tendency-plus-beh:rvior ta "nurture,"' since ftw such compulsions can he so rigid that social factors will not also affect the behavior pattern. Another difficuGy in identifying "nurture" " as "thar whit:h might Ixr avoidable"' is that sorne day s<•lective alionion. (or, more sfx•eulatiivcly, wlecti've conception) may Ik• prrsilile to avoid thr birth uf a ft•w babies wiih a near ccrtaioty of druth tronr rauce.r. other's effects. If many other genetic factors are relevant to stomach cancer, then maybe two compatriots chosen at random would be likely to differ quite widely in their genetic susceptibility to the external causes of stomach cancer, although it is still possible that there is much less individual genetic variation than many people suppose.8 Whether most Americans are of simi- lar susceptibility or whether there is typically wide variation in susceptibility makes little difference to the net effects of changes in nurture on the total number of cases in the nation as a whole and is therefore of little immediate public health relevance. (In either case, if the causes of stomach cancer are halved, then the stomach cancer rates will be roughly halved, as has been happening every 20 years.) Moreover, even if individuals do vary widely in their genetic suscepti- bility to stomach cancer, nhis does not suggest that different countries will vary, widely in the averages of the genetic susceptibilities of their citizens, for in each such average all the large variations between com- patnots will be ironed out. Fot a few types of internal cancer the differences between countries may be chiefly due to large differences in genetic susceptibility (e.g., the shortfall of chronic lymphocytic leukemia among the Chinese an& Japanese or the excess of cancer of' the nasopharynx among the southern Chinese), but this seems likely to be the exception rather than the rule. For example, taking the three types of cancer which are currently commonest in the United States (lung, colo- rectal, an& breast cancers); lung cancer was less than half as common a quarter of a centuryy ago, which shows that most cases are avoidable, while for both breast and colorectal cancers there are striking corre- lations between the rates in particular countries and various aspects of those countries' life-style (e.g., fat consumption; text-fig. 1), It is most implausible that international variations in daily fat consumption are chiefly determined genetically,, and if it is accepted that they are not, then the striking correlations between dietary factors and the onset rates of certain types of cancer show that the large international differences in onset rates are not chiefly genetic in origin. [Note that these correlations merely suggest that these cancers are r It is sometimes suggested that because a percentage of smokers do not get lung cancer, there must be other catLses, or genetic variability. The conclusion may or may not be correct, but the argument for it is bogus. Conversely, it is often argued that because the relatives of patients with a particular type of cancer have onlyy moderate rather than marked excess risks of that type of cancer (although no excess of cancer inm general);, the amounr of simplyy inherited genetic susceptibility must also be moderate rather than marked. This argument sounds reasonable, but in fact quite marked genetic variation usually leads to surprisingly moderate excess risks in relatives (Peto J, 1980)4 , so this argument too is bogus unless the analysis is of people with two or more relatives affected by one particular type of cancer (and makes tlue allowance for familial sitnilarities in life•style and' environment). At present, the relevance of genetic susceptihility to tlit• cowttnon types of cancer rrtnains obscure. ea JN(:I', VCJI. 66, NO. ti. .ft:Nt: 1981

Avoidable Risks of Cancer fn the U.S. 1205 A 50 >0 m .ro. .•.~ rt1 vlK . 'N • .~ll W. . Cr. - .Co ~ .D0 on 120 tsD :a0 t.D :aD170. ItR CAiUt,DAIEY MEAr CONSUMPttON- GAAM.S rEnnAtE .K.~faAW~ ers~ay ,~... ..«,.,..^ .~. . .~. a t~ 11L -0nG..u.w C5 ~..~.,~ ...~ ...~.. ~,.,,, om .Ofi0 00 too 120 B TOT14 pETAftYFnT :NTGKE ( Fj / Doy 1. t - - --- - . .~ 1,~ ~rEXT-FlGURE 1'~.-A). Correlation between colon cancer incidence in various countries and meat consumption (Armstrong and Doll, 1975a; reprinted with permission of British Journal of Preventive and Social Medicine and R. Doll). B) Correlation between breast cancer mortality in various countries and fat consumption (Carroll, 1975; reprinted with permission of Cancer Research and'~ K. K. Carroll)j These sviking, age-standardized correlations do not' necessarily suggest that either meal or some type of' fat'are major determinants o/ either colon or breast cancer, but they do suggest that mani¢ ulab& determi.nants of these cancers do exist- Iargely avoidable (except perhaps among those few people with the extremely rare genetic conditions of a strong predisposition to colon cancer or to breast cancer at an early age) but do not mean that avoidance of dietary fao would achieve this.] Turning finally to the role of age itself, it is sometimes suggested that because cancer is ten or a hundred times more likely to arise in the coming year in old people than in young people,, aging per se should be thought of as an important determinant of cancer. We rather doubt whether this viewpoint is a scientifically fruitful one (Doll, 1971; Peto et al., 1cJ75),, and in any case we are concerned in this report with avoidable causes of cancer, among which we can hardly count old age. 3. PROPORTION OF U.S. CANCERS THAT ARE KNOWN TO BE AVOIDABLE If the foregoing is accepted :Is justifying the belief that much huntan cancer is avoidable, tUen a crude estimate of'~ the proportion of cases that might be avoided in any tlne conttrtunity c•an lx• obtained by comparing for each separate type of cancer the inci', dence in that community with the lowest reliable incidence that is recorded elsewhere. For this purpose, the calculation is best confined to figures for men and women under 65 years of age, because the data on older people are unreliable (see also Appendix C): The proportion of avoidable cancers in older people is best estimated indirectly (see below). For certain types of tumor we have also thought it wise to omit rates for those communities that are believed to have low rates largely because of genetic insusceptibility. Finally, we have omitted the common non-melanoma skin cancerss entirely as, although they vary in incidence even more widely than most other types of cancer, reliable figures for their incidence are not generally available and they are, in any case, easily treated and seldom fatal. Before incidence rates in different communities can be compared meaningfully, however, they must first be corrected for the fact that some communities have a higher proportion than others of young people (among whom cancer is everywhere extremely rare). This is allowed for by "age standardization," which we have done by calculating what the incidence in each com- munity would have been expected' to be if the propor- tions of young people in each had been the same as in the respondents to the 1970 ULS. census. Details are given in appendix A, and age-standardized rates for different communities can differ only if the incidence rates observed among people of a given age really differ between the different communities. To estimate the proportion of all cancers that might have been avoided, we have taken, as an example, the population under 65 years of age in Connecticut during 1968-72 and have compared the incidence of each type of cancer (other than non-melanoma skin" cancer) in that population with the incidence rates recorded in the populations listed in table 7. For example, the age-standardized rate for cancer of the esophagus among men under 65 years of age in Connecticut was 34.6 per million, while that in rural Norway was only 6.5 per million. Similar calculations were made for 37 other types (or groups of types) of cancer in men and for 40 types (or groups of types) in women. In selecting low rates, we confined ourselves to data from about 1968 to 1972 from registries selected by the IARC (1976) as being reasontrbly reliable. The results are shown in table 7, and the total of these low incidence rates is contrasted with the corre- sponding totals for all types of cancer (except non- melanoma skin cancer) in Connecticut and in many other parts of the United States in table 8. The comparisons in table 8 suggest that in most parts of the United States in 1970 about 75 or 80 0 of the cases of cancer in both sexes might have been avoid- able. The proportion could be more, as the lowest rates that have been used almost certainly include somee avoidable cancers, especially since sornee of the countries that differ most markedly in various ways from the United States do not have a good cartcer registry and so havee not beert uscd in table 7. (Nloreovc•r, the propor JNia. \'t)1-. lifi;, NU, ti. ll°Nk 1981.

1206 Doll an6 Peto tion in 1980 will probably, be about one percentage point larger than that in 1970 due to the steady increase in tobacco-induced lung, cancer in the United States.) However, the proportion that might by prac- ticable means be avoidable may well be somewhat less than is suggested by tabla's 7 and' 8, partly because in a developed area such as the United States some lumps may have been counted that, although histologically "cancer," were biologically benign (appendix C); but, more importantly, because even if means of modifica- TABLE 7.-Cancer rates in selected low.incidenee areas among people under 65 years of age°'° Male rates in: Registry, with lowest reliable incidence for: Female rates in: Type of cancer Connect- icut registry Low-in- cidence registry Males Females Connect- icut registry Low-in- cidence registry Lip 11.8 4.1 United Kingdom, southern metropoli: tan region United Kingdom; Birmingham 0.8 0.4 Tongue 19.8' 4.1 New Mexico: Spanish Israel: Q Jews 6.7 2.7 Salivary gland 7.3' 2.3 Japan, Miyagi Japan, Miyagi 6.7 1.2 Mouth 31.3' 0.8 " " 11.8 2.4 Oropharynx 13.9 1.1 " " 6.0 0.8 Nasopharynx 11 5.6 2.4 East Germany East Germany 1.1 1.1' Hypopharynx 10:7 1.4 " " 2.9 0.2 Esophagus 34.6 6.5 Norway, rural Norway, rural 8.3 1.8 Stomach 66:2 28.0 New Mexico: whites United States, Iowa 26.7 16.6 Smalli intestine 6,4! 3.0 Israel: Jews Israelc Q Jews. 5.0 2.5 Colon 137:2 13.7 Nigeria, Ibadan Nigeria, Ibadan 140.7 1L6 Rectum 98.6 14.1 "' " 66.1 17.2 Liver 11.8 6.0 United Kingdom, southern metropoli- tan region United Kingdom, Oxford 5.3 1.0 Gallbladder, plus ducts. 9.0 3.3 Norway, rural Norway, rural 11.2 6.7 Pancreas 45.11 21.0 Nigeria, Ibadan Nigeria, Ibadan 30.9 14.9 Nose 3.2 2.2 United States, Iowa United States, Iowa 2.4 1.5 Larynx. 54!7 11.5 Japan, Miyagi Japan, Miyagi 8.2 0.4 Bronchus 325.8 9.0 Nigeria, Ibadan Nigeria, Ibadan 96,9 8.7 Bone 9.3 7.3 Puerto Rico United States, Iowa 7.5 5.2 Connective tissue 20.0 12.5 United Kingdom„ Birmingham United Kingdom, southern metro- politan region 14.6 6.4 Melanoma 40.8 8ff United Kingd'om,. Liverpool Uhited Kingdom, Liverpool 38.6 18:4 Breast 3.5 1.7 Finland Israel: Q non-Jews 593.7 100:9 Cervix Israel: Q~Jews 90.4 42.5 Choriocarcinoma United Kingdom, Oxford 1.2 0.2 Other uterine cancers Japan, Miyagi' 150.6 11.1 Ovary 104! 8 25.9 Other female genital organs 11 16.0 2.3 Prostate 92.3 5.3 Japan, Miyagi Testis 26.6 7.1 Penis 2.0 0.2 Israel: Jews Bladder 113.1 17:8 Japan, Miyagi Japan, Miyagi 32.8 7.3 0 Kidney 59:6 9;0 Nigeria. Ibadan Nigeria, Ibadan 23.2 2.5 Eye 4.3 2.0 Japan, Miyagi Japan, Miyagi 4.3 0.5 Brain and CNS 54.9 12.2 " " 35.2 8.9 Thyroid 12.4 3.6 United Kingdom, southern metropoli- tan region United Kingdom, Oxford 34.0 8.8 Other endocrine cancers 2.5 1.4 Puerto Rico Puerto Rico 2.2 0.6 Lymphosarcoma 39.8 13.1 " " " " 25.5 6.4 Hodgkin's disease 37.4 6.2 Japan, Miyagi Japan, Miyagi 28.1 3.5 Other reticuloses. 11.3 1.8 Israel: Jews Israel: 9 Jews 7.6 1.9 Myeloma 15.1 1.8 Japan, Miyagi Japan, Miyagi 9.6 3.3 Leukemia 57.9 40.8' New Mexico: Spanish " " 41.1 36.3 Polycythemia 4.8 0.6 Japan. Miyagi " " 1.6 0.3 All other cancers 89.9 33.7 New Zealand: whites New Zealand: whites 74.6 23.5 Total, all cancers 1,590 321 1,775 408 ' For all tumors except those of benign or unspecified malignancy an& non-melanoma skin cancers (which, collectively; accounted for <2% of all cancer deaths in the United States in 1978). ° From IARC (1976). ` Annual rates/million people <65 yr old, standardized for age as deseribe& in appendix A. JNC. vOt.. rA. hc). 6. Jt7Nt 1981

Avoidable Risks of Cancer In the U.S. 1207 TABLE 8'-Com}tarison of total'tumor incidence rates'6 abserred in various American cancer regist'ries, circa 1970' Area in United Male tumor incidence Female tumor incidence States covered bv tumor registry° Observed Minimal,' as % of observed Observed Minimal,' as % of observed Alameda, Calif. (W) 1,589 20 2,108 19 San Francisco, Calif- (W) 1,668 19: 2,137 19 Connecticut 1,590 20 1,775 23 Iowa 1,422 23 1,594 26 Detroit, Mich: (W) 1,498 21 1,737 23 New Mexico (W) 1,469 22 1,784 23 New York, upstate 1,372 23 1,481 28 El Paso„Tex. (W) 1,245 26 1,682 24 Utah 1,215 26 1,464 28: "Ten areas" from TNCS (a study covering a modtrately representative tenth of~ the whole United States): TNCS White 1,519 21 1,702 24 TNCS Non-white 1,906 17' 1,721 24 TNCS White and non-white 1,557 21 1,705 24 ` Annual rates/million people <65 yr old, standardized for age as described'& in appendix A. ° See table 7, footnote a, for excluded tumors: " The totall of the lowest reliable rates for each type of cancer listed in table 7 was 321 (a) and 408 (Q), which is a crude indi- cation of the minimal incidence that might be achieved., ° W=whites only: avoidable may simply be tiie proportion that is avoid- able among middle-aged people. The foregoing estimates refer to all, malignant tumors, both fatal and non-fatal' (excluding only non~rnela- noma: skin cancer): Directestimation by similar methods (but on the basis of national, death certification rates instead of, as in tables 7 and 8, registered incidence rates) of the proportion of fatal cancers that are avoidable might be misleading. This is because many underdeveloped countries enumerate causes of death so inaccuratePy that comparison of their certified death rates from particular types of cancer witK the corre- sponding rates in the United States might overestimate the proportion of U.S. cancer deaths that is avoidable. However, the two types of cancer (lung and' large intestine) that currently kill the largest numbers of Americans have incidence rates that vary particularly widely between the United States and certain other countries and the U.S. deaths from these two types are therefore largely avoidable. The same is true of many other types of cancer that currently kill large numbers of Americans, and it is reasonable to suppose that the proportion of fatali cancers whose onset could have been avoid'ed' will be approximately the same as the avoidable proportion of all cancers discussed above, i.e., more than 75 or 80 o in principle but perhaps less in practice for many years to come. 4. ATTRIBUTION OF RISK tion of cancer risks can be identified, these may not be socially acceptahlt•. This might obviously be a seriouss limitation if preventive measures had perforce to be limited to ways whereby different countries already differ,, for affluent people will not be persuaded to ad'opt certain aspects of the life-style of the impover- ished. But there may be many different simple or highly technical ways of preventing the same cancer (see subsequent sections), some of which have not been inadvertently adopted by any country with a good cancer registry, at least one of which ways may be both practicable and acceptable. About half of the cancers diagnosed in the United States are found among people 65 or more years old, and we have made no explicit estimate in table 7 of what proportion of these might be avoidable. This iss because data from cancer registries become very unreli- able in old age, not so much in the United States nowadays as in those countries where the contrasts with the U.S. life-style and environment may be greatest. Consequently, any similar analysis of rates among older people might be severely biased. There is, however, little reason to suppose that the proportion of Li.S. cancers that would be preventable differs greatly above and below the age of 65 years as long as lungg cancer (which is relatively slightly tnore comrnon among the old)i and other cancets are considered separately (see sr•(tion 5:1 and appendix 1•:): f'ara- doxically, thcn•frrtc•.;, rhc most reli:rble available rstitnate rrf; tttc• propur,ir)n of c:utccr :rrnonf; oltlcr fx•oplc that is 4.1 Increases and Decreases in U.S. Cancer Rates If there were currently an "epidemic" of cancer in the L'nited States (by which we mean rapid increases in the probability of people of a given age developing most particular types of cancer), this might suggest that the search for avoidable causes for the cancers that we observe today should be directed chiefly toward various aspects of the modern environment that were much less widespread half a century or more.ago. If,, conversely, most of the cancers that are now common have been common for many decades, then, although this would not be evidence as to whether our new habits will eventually increase or decrease future cancer risks, it might suggest that the cancers that are cur- rently common, and that will continue to be common unless we do something about them, have been largely determined by long-established aspects of the AmericanN life-style or environment. ,N~ Practical Difficulties in Gauging Cancer Trends Cancer is certainly much more noticeable nowadays than it was a decade or two ago, bur this is not in itselfi evidence that cancer rates are increasing as there art~ several factors that influence public awareness abou cancer. First, especially when active treatment is bein undertaken, the friends and' relatives of cancer patient (or the general public, if the patient is a public figure) nray discuss the disease openly, whereas previouslv such~ matters often used to be hushed~ ttp and the .)NCt. VOt.. u,: NO. 6. JUNt_ 1981

1208 Doll and Peto TABLE 9. Deatie certification rates/1,000 Americans,° 1935 and 1875 All neo- piasmsb Respir- All causes except except atory ' All causes Sex Years neoplasms respir- tract atory cancers cancers Rate 'lu` Rate %` Rate 91i` Rate %` Male 1933-37 15:12 91.0 1.42 8.5 0:09 0.5 16.63 100. 1973-77 &91 81.0 1.41 12.8 0:69 6_2 11.01 100' Female 1933-37 11.92 87.6 1.65 12.1 0:03 0.2 13.60 100 1973-77 4.96 78.8 1.17 18.6 0:16 2.5 6.29 100 "' All ages, standardized for age to U.S. 1970 census (see ap- pendix A). For most scientific purposes, separate examination of the trends above and below the age of 65 is preferable (see ap- pendixes C and' D);, since many deaths from cancer half a cen- tury ago may have been miscertified as due to other causes;, particularly among older people. °' Benign ar.d malignant tumors are included'in this table, as elsewhere throughout this text. ` Rate as percent of corresponding all-causes rate in last column: diagnosis perhaps withheld even from the victim. Second, some cancers are now diagnosed that might previously have gone unnoticed in the medical treat- ment (and subsequent death certification) of dying people, especially of the elderly. Third, cancer has become relatively more common as a cause of'~ death chiefly because of the prevention or cure of so many other diseases. This is nicely illustrated by the data for 0.8% 0.7. x 0.61 h t WW D< .5% Y Y q 4: ~M 0.41 6 0 ~~ 0,39 2z y 4 0.21 0.1. DECREIISING 110N-IteSPIR/1TORy C7INCER.. yENAllS.. DECREJISING VASCULAR DISEIISESINCE.1970 '--+---~.. .. L fS/r11A70NYfJJCOI (N) - - - - - 11rs.rWl7onT cwNCeR (r) 1935 1940 1945 1950 1955 1960 1965 1970, 1975 CGrlSTJWT N0M-RESPIRATORY CENT.iA:. YE/WI (centrek y..ro! 5-ye.r period •tadled, 1933-37 to 1973-77) 7}:xr-r1cuRl: 2.-Annual aKe-swndardiit•d dialh ra(es„ I'J9'3-77, anhong Americans under 65 years of agc. CJINCBR NP LBSS females in 1935 and 1975 (table 9). The non-respiratory cancer death rates decreased substantially, but the death rates from all other causes decreased even more sub- stantially. Therefore, the percentage of female deaths attributable to non-respiratory cancer is actually greater now than it was 40 years ago, even though among women of a given age the absolute cancer risks are lower nowadays. If attention had been restricted to people under the age of 65 years (text-fig. 2); then the contrast between declining absolute rates and increasing percentages would have been even more marked. Fourth, there is a larger proportion of old people nowadays, and cancer risks are ten or a hundred times greater among old people than among young people. Finally, cancer has become a highly political issue, and con- sequently discoveries (perhaps using modern ultrasen- sitive analytical methods) of even quite small amounts of carcinogens in various everyday contexts attract vigorous media coverage, as d'o various other aspects of cancer research. We shall therefore review in this section, and in our appendixes C, D and E, some of the objective evidence concerning the upward and downward trends in the U.S. death rates from, and incidence rates of, various cancers. Epidemic increases in lung cancer are clearly taking place, as would be expected as a result of the widespread adoption of cigarette smoking earlier this century, but apart from this we can see no good evidence of a cancer "epidemic"' in the above sense. Unfortunately, both cancer registration rates (a "can- cer registry"' tries to count all the new cancer onsets in a particular area,, such as the State of Connecticut) and cancer death certf fication rates are subject to large JNCI, VC/l.. (7;, NO. (:, JUtNF:. 1981

errors; more unfortunately, these errors are not con- stant witK time so that artifactual trendsin theregistered' incidence or certified mortality rates for particular cancers may be superimposed on the true trends. The problem with:any comparisons of cancer rates in different decades is that these artifactual trends may be of the same order of magnitude as the trends in real cancer onset rates that one wishes to study. The chief sources and likely magnitudes of such biases are discussed in appendix C. Reduction of Bias: Trends in Mortality in Middle Age The data suggesting moderate improvements in rela- tive 5-year survival rates (e.g., from 60 to 68% for breast cancer)i are also discussed in appendix C(see table C2 on page 1278); where it is suggested that parr at least of these moderate apparent improvements is artifactual, due to progressively more complete enumeration of the non-fatal cases. Changes in treatment for many types of cancer have chiefly improved palliation rather than cure of the disease, and the true cure rates for many of the common types of cancer have probably changed very little since 1950. For these types of cancer the trends in death certification rates among people under 65 years of age, at which ages treatment of the curable and medical investigation of! the causes of death of the incurable have for decades been reasonably careful, may paradoxically y,ield' a mucK more reliable (and representative) indication of the real trends in cancer onset rates than can the superficially more attractive study of any of' the currently available data on registered incidence rates. The need to restrict attention to death certification rates for people under the age of 65 years arises because many people who died of cancer in past years never hacf' their disease diagnosed and might have been certified as dying of pneumonia, senility, or the wrong type of' cancer. Progressive correction of such errors over the past several decades has resulted in large artifactual trends (some upward, some downward) in the death certification rates for certain types of cancer„ especially during the first half of this century or, since 1950, especially among old people. (These and various additional biases also affect the trends in disease registration rates, where a registry tries to count both all fatal and' non-fatal cases of cancer: see appendix C.), However, for most types of cancer the trendk since the 1950's among middle-aged American death certification rates seem likely to yield a reasonable indication of! the true underlying trend's in the corresponding real dis- ease onset rates. Increase in Middle-Aged Mortality From Respiratory Cancer Either by examining the lower lines in text-figure 2 with a magnifying glass, or by referring to appendix tables I)1 and D2 (pp. 1282-1283) from which text- figure 2 was derived, it can be seeu that rnale respira- tory cancer death rate•.s appear to h,tve been rising stc~~adily for at Icast half a ce.ntrary and that fernale Avoidable Risks of Cancer In the U.S. 1209 respiratory cancer death rates started to rise a quarter of a century ago and are now increasing alarmingly rapidly. The trends in respiratory cancer are discussed in more detail in appendix E, where we conclude that before 1950 airitost the whole of the apparent increase in female lung cancer and some of the apparent increase in male lung cancer were artifactual, due to more accurate detection of lung cancer, but that some of the pre-1'950 male increase and virtually all of the more recent increases in both sexes are real and are largely or wholly caused by the delayed effects of the adoption, decades ago, of the use of cigarettes (see also section 5.1). (The long delay between cause and full effect arises because even among people who have smoked regularly throughout most of their adult livess the degree of exposure of the lungs to cigarette smoke during their late teens or early twenties remains a surprisingly important determinant of~ lung cancer risks in middle or old age. See text-fig. El on page 1292.) Lack of Generalized Increase in Middle-Aged Mortality From Non-respiratory Cancer Text-figure 2 and tables D1 and D2 (pp. 1282-1283) also indicate that the aggregate of all non-respiratory cancers has takem a fairly constant toll among males for half a century (with about a 10% decrease among younger men in the past decade), but that the total non-respiratory cancer death rate among females has been decreasing rapidly for half a century, due not chiefly to improved treatment but rather to d'ecrease& onset rates among women of a given age. For age- specific details, see text-figs. C1 and C2 on page 1272. (All the overall comparisons we make are based on "age-standardized" rates, which can change onlyy be- cause of changes in the risk of cancer among people of a given age; increases or decreases in the proportions of old people will not affect them. This is not true of "crude" cancer rates nor of "percentages of all deaths attributable to cancer," and these should never be used to characterize trends: see appendix A.) However, non- respiratory cancer is an aggregate of many completely different types of'~ cancer, some of which are increasing and some of which are decreasing. Text-figures 3 and 4 describe, for males and for females, respectively, changes in mortality (or, more strictly, death certification rates) during the past quarter century for various types of cancer. 14oree detailed data are presented in table D3 on page 1284, together with a separate discussion of the apparent changes in mor- tality from various particular types of cancer among people under 65 years of age. Corresponding details for people aged 65 years and over appear in table D4 on page 1285. A11 the changes are small in comparison with the large increases in the smoking-related cancers of the respiratory and upper digestive tracts, although the decreases in mortality from cancers of the stomach and uterus are also important. ln apfx•ndix I) we also present the recent (Jf1(i8-"!ti) trends in drath svrtificatiotr rates among Arnericans in Jwct: vt,t.. ta;: NO. t;. )tt.rvt: t9r'<t

1210 Doll and Peto ROIR'H. iSOPAGbS PHARINX OR LARYNX LUNG STOtlACH INTESTINES (INC. RLCTUM) LIVER. GAI.i. BLA'JDER • BILE DUCTS PAN'CREAS BONE SttiN BLADDER 1lIDNEY PRUSTATi BRAIN • NERInOS itUKA1:HIA r "J' ~ HODGKIN'S DISEASE & MORTALITY -7 XeY ~ 1 195%JJ+] 19)J+7 Hale., age 0, 6Iyeera, certlfled nortallty OIHER RETICULO- EN7()'71:ELI.4 OTHER A(@+Ir:EIY UNSPECIFIED SITiS TEXT-FIGURE 3-C:ertified mortalitr^ per 100 million males, ages 0-64 years (standardized for age to U.S. 1970 population as de- scribed inI appendix A): early middle life (35-44 years of age), as it is here that the first effects of'any changes for better or worse in the causes of cancer might first be clearly evident. Reassur- ingly, no unexpected upward trends emerge (see table D6 on, page 1287), while significant downward trends are seen ini mortality from many types of cancer. For males, the sites where there are now significant de- creases in mortality at, ages 35-44' years include the pancreas, lungs (presumably chiefly due to decreasing tar yields per cigarette: See appendix E), and genitalia. For females, they include the intestines, genitalia, reproductive system, and breast (~the latter decrease due perhaps to a protective effect of early childbirth on the mothers of the 1950's glut of'babies). Overall, cancer mortality among young adults in the United States is decreasing quite rapidly, and much of the decrease cannot plausibly be attributed to improved therapy. Trends in Incidence, as Assessed by Cancer Registry Data "I'urning (with some trepidation, because of the greater likelihood of bias) from trends in certified mortality to trends in registered incidence, we are immediately confronted with the problem of exactly which incidence data to study-those from particular cancer "registries" that have operated for decades, trying to list all the cases, fatal or otherwise, of cancer in New York or Connecticut, those from comparison of the Second Natitlnal' (:'tncer Survey (SI\r(S) in 1947 or 1948 with the Third National C»icer Survey (7'\C:S1 in 1969-71, or those from comparison of the -I-\CS with the Surveillance: Epidemioli;y, and End Rc-sults (SI`F.R) pn)l;rarrr of the rnid-197(1's' (.SN(.S, 0 2000 6000 10 800, 15 000 70000 25000 30 000 I ( I I t~~. I'~. I r 1 ESTIMATED RaTES AMONGNON-SMDKERS TNCS, and SEER all tried to monitor cancer incidence in about one-tenth of the entire U.S. population.) Unfortunately, many, of the above comparisons suffer from such large artifactual irregularities and biases (see discussion in appendix C and text-figs. C3 to C5, pp. 1274-1276) that for most types of cancer they yield much less reliable information about long-term trends in real disease onset rates than the mortality data do. The only one of these comparisons of cancer inci- dence rates that is at all compatible with the mortality data is than of the SNCS (in 1947 or 1948) with the TNCS (in 1969-71). This comparison has been de- scribed by Devesa and Silverman (1978, 1980). From their 1978 paper we have abstracted text-figures 5 and 6, describing the changes in registered incidence rates for each of the major types of cancer. The overall pattern of change indicated by these text-figures is, of course, roughly similar to that indicated by the mor- tality data, for this was why we selected! this particular comparison of incidence rates for study. Consequently, we would not strongly disagree with anyone who argued that even the comparison of incidence rates in text-figures 5 and 6 is so uninformative that it would be preferable to rely chiefly, on mortality data (although some of the striking differences. between certain of the trends in incidence may be informative as, for example, the apparent decrease in cancer of the cervix but not of the endometrium). MOUTH,, ESOPHAGUS, PHARYN% OR LARYNX LUNG STOMACH INTESTINES(INC. RECTUM) LIViR, GALLBLADDER AND BILE DUCTS PANCREAS BONE SKIN BREAST BLADDER KIDNEY CERVIX UTERII ENDOMETRIUM AND OTHER UTZR:`S OVARY BRAIN AND NERVOUS LtUxAEMIA HODGKIN'S DIStASE OTHERRCTICULO- ENDOTHELIAL 0 2000 6000 10 000 15 000 . 20 000 I I I I I li I I ~ ESTIMATED RATES AdOHGNON-SMOKERS ~P .] y MORTALITY r ~ Key 197)-7 Female, age 0-64 years, certtfled RR:r2ality OTHERAND (CHIEFLY UNSPECI~F'IEDSITES TF.xt;FictfRe. 4.-Certified monality per 100 millinn females, ages 0-64 years (stmdardiztnl for age to U.S. 1970 tx/pnla11U11 as de- xrilx•d in aptxmdix A). J\(SL VOL. (a6;, NO. b. 11'1~F~. 1901 ~.

Avoidable Risks of Cancer In the U.S. 1211 lqUTH„ ESOPHAGUS,. PH,ARYNX OR LAR:Y NX. LUNG STOMACNINTESTINES.(r.NC. RECTUM) LIVER, GALLULADDEA IUIDD aSLE DUCTS PANCREAS. HONE' MELANOMA ULADDER KIDNEY' PROSTATE BRAIN AND NERVOUS KAEMIA EU 0 100 200 JUD, 400 50D 600700 ' I 1 t t 1 1 t t J ] t d INCIDENCE Key~ snCS, 19oYB' TNCS, 1969-71' L . Mste.,. alli aqes, registeredlncidence HODGKIN'S'. DISEASl: OTHER IIETICULO- [NDOTHE I.I AL OTHER AND UNSPEC- IFILD SO:iID, EXCEP~TSKIN TExT.Ftcvae 5.-Registered incidence rates per million males, all ages' (st-tndbrdized for age to U.S. 1950 population and for race to 90~'a white). However, even if the detail of the incidence trends is uncertain, the general picture is clear: a) the most important absolute increases have been in cancer off the lung, b) the mosr important absolute decreases have been ini cancers of the stomach andl uterinecensix, and'c) less reliably, there seem to be no large changes in the aggregate of the incidence of all nonrespiratory cancerq(for which the age-standardized incidence regis- tration rates d'ecreased! between 1947-38 and 1969-71 by 3% for males and' by 19% for females.). Comparison With Interpretations by Others ln suntmary, the trends since 1950 in mortality in middle age, sornewhat reinforced by, the trends in incidence between the Second and Third National Cancer Surveys, suggest that, apart from the effects of smoking (and perhaps asbestos: See section 5.6), there are no major epidemic increases in cancer. Unfortu- nately, our conclusion is not shared by all commen- tators. Epstein (',1981b)) whose book, TheYofil':c.sofCancer10 (Epstein, 1978, 1979), was based on the ° N!e have'e!xcdudrd; sinre• ttie surn•eyti did rtot attc•rnpt to register it; non-mrlanunta skin cancer, Non-ntclanoma skin cancer is diagnosed mort• contntunly than any othrr type of c:utcer, but it is nearly always so easily cured that it is one of' ttie• Ic•:tst common fatal ca ncc•rs. lo' For a wide'-ran/;in}; cumm(•nt on Epstein's (1978, 1979, 1981a, 198115) lx•rslx•ctive on (hr caust•sol eanrer, which will make clear our re•asons fnr rtot drlwing on it in uur present report, srr Peto. I980! The par(ieular qur'stiun of thc• rol(• nf cxcupatii)n:J la((ms will la+ dealt with in w•cti(/n rr/i' :/nd apfx ndix F, where s(ronl; reasons fln distruti(inl; F:ps(ein'. 11!IR1aai) suw(r. :uw y;iv(•n: assumption that Americans live in an era of genuinely and rapidly, increasing cancer rates over and above the increase dite to tobacco, rejects in out of hand without acknow1'edging or explaining wh.• the trend in U.S. mortality from non-respiratory cancer in middle age is actually downward, and without serious discussion of the potentiali biases in trends in death certification rates among older people (or, still' more so, in trends in the registered incidence rates of tumors) that we have emphasized' in appendixes C, D, and E. The Toxic Substances Strategy Committee (TSSC) in their 1980 report to the U.S: President also came to a conclusion directly opposite to ours, namely, that "even' after adjustments for age ... recent figures show that both incidence (new cases) and mortality (deaths) rates are increasing," and later that "when the effects of cigarette smoking are corrected for, the recenu trends in incidence show an increase." Their conclusions about rising incidence rates were based on the data of Pollack and Horm (,I980)1 and on the interpreration of these data by Schneiderman (197'9) and rested heavily on a com- parison of the incidence rates recorded in the TNCS during 1969-71 with those recorded in the ongoing SEER program that began in 1973. In appendix C we show that this particular comparison yields estimates of trends in reali disease onset rates that are grossly discrepant with more reliable data. An even more serious error in the TSSC (1980) report is the committee's peculiar method of "allowing" for the effects of! cigarette smoking on the recent trends in MOUTK, ESOPHAGUS.,, PNARYNX..OR LARYNX LUNG STOMACH INTESTINES (INC. RECTUM) LIVER, GALI~tLADDER AND BILE DUCTS PANCREAS BOME MELANOMA BREAST BLADDER KIDNeY' CERVIX',UTERI ENDOMETRIUM AND OTHER UTERUSOVARY' BRAIN:ANDNERYOUSIEUKAEMIA. xODGKIN'S DISEASE OTHER RETICULD- F:NDDT)IELIAL Ot'NER.AND tN'SPL'C- I PI ED S0L1 D, EXCEPT SKIN. J ~ 8 100 200 300 400 500 600 700 1 II I I I I I 1 Key Q INCID$NCE ~ sMCr, 1W7/B'. TKt, '.%9-71' Penrl., all aqe., reqlstered', 1nclA.nce TEX7'-rtctleE 6.-Rt•gistert•d'' incidence r:ues per million feutales- a1l ages (standardinYf for age to U.S: 1950 fxlpul:uion artd for race to 90°. white). a\fa. VUl.. tli, 1t). (r; JtrRl:' I!IN1i