"The Great Debate? Smoking, Lung Cancer, and Cancer Epidemiology GERRY HILL WAYNE MILLAR JAMES CONNELLY Abstract. During the first decades of the 20th century the lung cancer death rate increased sharply in developed countries, including Canada. An association with cigarette smoking was suspected. Between 1950 and 1960 many epidemiological studies confirmed this association. The validity of these studies was attacked fiercely by some renowned statisticians. The resulting debate sharpened the methodological and analytical rigour of epidemiological studies. Epidemiology became an indispensable tool in cancer prevention. RCsumb. Pendant les premi6res d6cennies du XXe siecle la mortalit6 attribuable au cancer du poumon augmentait nettement dans les pays industrialids, le Canada compris. On a soupconne une relation avec la fum6e de la cigarette. Entre 1950 et 1960 ungrand nombre &etudes 6pidemiologiques ont corrobor6 cette relation. Plusieurs statisticiens r&ncnnmesont attaque fhrocement la validit6 de ces etudes. Le d6bat suivant a aiguis6 la rigeur m6thologique et analytique des etudes Cpidemioloqiues. IXpid6miologie est devenue un outil indispensable pour la prhention du cancer.

,

In his magisterial survey of the major epidemics of the 20th century, Sir Richard Doll2 identified lung cancer and coronary thrombosis as the two most important causes of premature death. Although both epidemics have begun to wane, lung cancer is still the leading cause of cancer mortality in Canada, accounting for about 30% of cancer deaths in males and 20% of cancer deaths in females.3 It is now acknowledged that the epidemic of lung cancer was due primarily to cigarette smoking, and some progress has been made in the campaign to reduce the prevaGerry B. Hill, MD, PhD, FRCPC, Dept of Community Health and Epidemiology, Queen's University. Wayne Millar, MA, MSc., Health Statistics Division, Statistics Canada. James Connelly, MB, BS, MSc, Nuffield Institute for Health, University of Leeds.

368

I

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

lence of smoking, especially among males. However nearly a third of Canadians over the age of 15 remain addicted to nicotine,4 and the finance departments of Canadian governments remain addicted to the revenue from the sale of tobacco.5 The mid-century studies which established the link between smoking and lung cancer, and the scientific debate which they provoked, were instrumental in the development of cancer epidemiology and cancer prevention. Before the 20th century knowledge about the etiology of cancer came from the observations of astute clinicians, who noticed the increased frequency of particular types of cancer in special groups of people. In 1713Ramazzini of Padua6 noticed that cancer of the breast was more common among nuns (a finding which was confirmed by Stern in 1842,7 using death records in Verona). In 1761 John Hill) an English physician, found snuff users to be prone to cancer of the nose (probably the first indication of the carcinogenic properties of tobacco). In 1795 the renowned English surgeon Percivall Pottg described the high incidence of cancer of the scrotum among chimney sweeps. During the 19th century several occupational cancers were identified by clinicians: skin cancer in copper smelters and workers exposed to paraffin, tar and mineral oil; lung cancer in Schneeberg miners (later shown to be due to radon gas); and bladder cancer in dye workers.10 Early in the 20th century the scope of the discipline of epidemiology, previously concerned exclusively with the study of communicable diseases, was broadened to include non-communicable diseases such as cancer. This development was spearheaded in Britain by Major Greenwood at the London School of Hygiene, and in the United States by epidemiologists working in the federal, state and local health departments.11 The first national epidemiological studies of cancer were based on mortality statistics, which became available in England and Wales in 1837, and later in most developed countries.12Prior to 1911, however; mortality statistics were published only for cancer as a whole, and epidemiological debate was focused on the apparent increase in mortality from all types of cancer, in particular whether this could be a statistical artefact due to increased diagnosis of cancer in organs not readily diagnosable by inspection or palpation.13After 1911 mortality statistics for cancer in specific organs became available, and it was possible to study trends over time and differences in relation to sex, age, occupation, and place of residence.14 These broad descriptive studies based on mortality statistics were later supplemented by more detailed etiological studies involving comparisons of cancer patients and controls. In 1929 Fry15 published the results of a study in which he compared the Wassermann reactions of patients with cancer of the digestive tract with those of patients with other diseases; and in 1933Stocks16 reported a large, multicentre study of

"The Great Debate"

369

the characteristics of cancer patients and controls, the first study to identdy the protective effect of eating vegetables. Fifteen years later a second, and more precise type of etiological study was introduced, in which cohorts of workers exposed to arsenic" and chromatesls were followed up to determine their mortality from cancer. The problems of designing, analysing and interpreting epidemiological studies of cancer and other chronic diseases took several decadeito resolve. This process was catalysed by the debate over the role of cigarette smoking in the epidemic rise in lung cancer. In this paper we describe this epidemic, the studies incriminating cigarette smoking, the scientific debate which ensued, and the contribution of the debate to the development of chronic disease epidemiology. THE EPIDEMIC OF LUNG CANCER

I

Autopsy Studies Although cancer of the lung had been identified as a pathological entity by Morgagni in the early 18th century,19and studies of Schneeburgminers in 1879 had shown them to be at increased risk,m prior to 1920 the disease was believed to be rare among the general population. In 1859 Rudolf Virchow,21 published statistics on 3,390 autopsied deaths in Wurzberg in 1852-55. In only one case (0.03%)was cancer of the lung reported. However, by 1924 Walter JGkuthF2 in a study of autopsies in Hamburg, found that the proportion with lung cancer increased from 0.07% in 1880-99 to 0.4%in 1900-1911, and 0.7% in 1912-23. JGkuth concluded that the concept of the rarity of cancer of the lung could no longer be maintained. The subsequent literature on the prevalence of lung cancer at autopsy was the subject of several reviews.23 The results are summarized in Figure 1, with a trend line fitted by logistic regression. Although there is considerable variation between studies (some due to the small numbers of cases) there is a clear upward trend from 0.6% in 1900 to 5% in 1947. Understandably some of the earlier authors were sceptical about the possibility of an increased frequency of lung cancer. Even as late as 1967, Rupert Willis questioned the reality of the apparent increase, suggesting that it could be explained by a "greater 'surgical interest in the disease, diminished surgical interest in some other kinds of cancer, and greater competence of pathologists in the recognition of the less obvious lung cancers at autopsy."24 A general finding in the autopsy studies was that the frequency of lung cancer increases with age (though the disease does occur in childhood), and that males are affected more than females with a ratio of about 4 to 1.

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

Figure 1 Lung Cancer at Autopsy: Combined Results from 18 Studies 8

Per cent of autopsies

1860

*

1870

1880

1890

1900

l910

1920

1930

1940

1950

Year Observed

I

+fitted

Mortality Statistics The Registrar General of England and Wales began publishing the numbers of deaths for specific cancer sites in 1911.W The death rates for cancer of the lung from 1911 to 1955were published by Percy Stocks.26The rates increased exponentially over the period: 10%per year in males and 6% per year in females. Canadian rates for the period 1931-52 were published by A. J. Phillips.27 The rates were consistently lower in Canada than in England and Wales, but also increased exponentially at 8% per year in males and 4% per year in females. The British and Canadian rates are shown in Figure 2. The rates (a)for males, and (b) for females have been age-standardized,28 and the trends extended to 1990, using data published by Richard Peto and colleagues, 29 and by Statistics Canada.30In British males the rates reached a maximum in the mid-1970's and then declined. In Canadian males the initial rise was more prolonged, reaching a maximum in 1990.Among females the age-standardized rates continue to climb in both countries, the rise being steeper in Canada than in Britain. The fact that mortality was lower at first in Canada than in Britain may be explained by the difference in smoking in the two countries. Percy Stocks31 cited data on the annual consumption per adult of cigarettes in various countries between 1939 and 1957. In 1939 the consumption in Canada was only half that in the United Kingdom, while in 1957 the consumption in Canada was 5% higher than in the United Kingdom. The trends in the age-standardized rates in most developed countries between 1953and 1992have been published recently by the International

"The Great Debate"

371

Figure 2(a) Mortality from Cancer of the Lung in Males Rate per 100,000

120 l00

-

80

-

60 40

-

20

-

0 l910

1920

1930

1940

1960

1960

1970

l980

l990

2000

Year

-England & Wales

Canada

+

United Kingdom

+

Standardlzed to European Population

Figure 2(b) Mortality from Cancer of the Lung in Females Rate per 100,000 36

30

-

26

-

20 l6 l0

-

0 1910

1920

1930

1940

1960

l960

1970

1980

1990

Year

-England

(L

Wales

Standardlzed to European Population

Canada

+

+ -

United Kingdom

2000

372

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

Union Against Can~er.3~ In all but three of the countries (Japan,Poland and Yugoslavia) there is a clear quadratic trend in the rates for males. In all countries there is a continuing upward 'trend in the rates for females. In both England and Wales,33and in Canada34 cohort analysis of the age-specific mortality rates showed that the increased risk of dying from lung cancer began in cohorts born around the turn of the century. The analysis for Canada has been extended in Figure 3.35In both sexes the increase reached a plateau for cohorts born between 1930 and 1950 and subsequently declined. It is interesting to note that this decline occurred in both sexes, though the age-standardized rate for females continues to increase. The figures for the earliest and most recent cohorts are, of course, based on incomplete data. Figure 3 Cohort Mortality from Cancer of the Lung in Canada Mortality Ratio (1911 1) 2

1

1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970

Birth Cohort

-Males I

Females

-+

Etiological Hypotheses The authors who reviewed the increasing frequency of lung cancer in autopsy and mortality statistics also discussed possible causes. The general theory of oncogenesis at the beginning of the 20th century was a combination of the chronic irritation theory of Rudolf Virchow and the theory of hereditary disposition derived from the animal breeding experiments of Maud Slye.36 In 1929 Car1 Weller predicted that the increase "will be found due to (1)an inheritable intrinsic predisposition which may be activated by (2) a variety of chronic irritative factors. These may be mechanical, chemical, bacterial, thermal or radioactive."37 The possible irritative factors included tuberculosis, the 1918influenza pandemic, dusty occupations, gassing of men in the First World War, automotive

"The Great Debate"

373

vehicle exhausts, and use of oil and tar on the roads. Weller mentions tobacco only in the context of industrial exposure to organic dust: "The number of cigar-makers affected by pulmonary carcinoma has seemed disproportionate to the number engaged."38 Clinical opinions incriminating smoking began as early as 1923.39 In 1927 an editorial in the Lancet on the increasing frequency of lung cancer at autopsy included the following comment: "Tobacco smoking is much more prevalent than it was, and it is one of the signs of the times that the class from which general hospital patients are mostly drawn have changed pipes for cigarettes."40 EPIDEMIOLOGICAL STUDIES OF SMOKING AND LUNG CANCER

1

Case-control Studies 41 The use of the case-control approach to study smoking and lung cancer began independently in Germany, the United States, and Britain during, and immediately following, World War 11, and subsequently spread to other countries. Anti-tobacco campaigns, based on moral and eugenic concepts, had developed in the United States in the late 19th century, and in Germany in the early 20th century. In Germany, during the 1 9 3 0 ~ these ~ efforts were strengthened by the work of physicians such as Lickint, and drew state support from the racist ideas of the Nazis and the personal asceticism of Adolf Hitler.42In 1939, Franz Muller, a young physician at the Ufiversity of Cologne, published the first case-control study of smoking and lung cancer.43 His study was based on the smoking histories, obtained from questionnaires, of 86 lung cancer patients and 86 controls. Muller classified his subjects into non-smokers and five levels of tobacco consumption (cigarette, pipe or cigar). The results were presented only as percentages in each category, without statistical analysis, but in fact the overall differences between lung cancer patients and controls were significant (p < .01), with a highly sigruficant increasing risk with amount smoked (p < .001). Muller's study was replicated four years later by E. Schairer and E. Schoniger of the Jena Pathological Institute.P4Their study involved 109 cases of lung cancer, 318 cases of cancer of the gastrointestinaltract and prostate, and 270 controls. The results, in terms of the comparison between lung cancer and controls, were almost identical with those of Miiller. The smoking histories of the other cancer patients were not different from those of the controls. Anti-tobacco campaigns in the United States lost momentum during and after the World War I, due to the popularity of the cigarette in the armed forces. Although the increasing use of cigarettes by women was deplored on moral grounds, the majority opinion among physicians

374

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

was that smoking was not injurious to health, and may have psychological benefits.45However, in the 1930s the biometrician Rayrnond Pearl showed that smoking reduced life expectancy, and some physicians, such as William McNally, Pearl and Alton Ochsner,identified smoking as the cause of the increasing incidence of lung cancer.%Ochsner had been trained by Evarts Graham of St. Louis, a pioneer of thoracic surgery. Although Graham, a cigarette smok r who subsequently died of lung cancer, was sceptical of the smoking ypothesis, one of the young surgeons on his service, Ernst Wynder, persuaded ,Graham to undertake a multicentre case-control study, supported by a grant from the American Cancer Society. Wynder was not, at that time, an epidemiologist, but he received help and advice from the epidemiologist Lester Breslow, the Dean of Public Health at the University of California in Los Angeles!' Wynder and Graham's study was published in the Journal of the American Medical Society in May, 1950.48 The same issue of the journal included another case-control study of smoking and lung cancer, by Levin, Goldstein and Gerhardt of the Roswell Park Memorial Institute in Buffalo.49 Although the two papers were published simultaneously it appears that the Roswell Park study deserves priority status.50 Morton Levin, an epidemiologist who had studied with Wade Hampton Frost at Johns Hopkins, developed an epidemiology program at Roswell Park in the rnid1930s. This involved a comprehensive system of data collection, including smoking histories, for all patients.51 The two studies were the first to include a formal statistical analysis. In both, the proportion of smokers among the lung cancer patients was significantly higher than among patients without cancer. In the study by Wynder and Graham there was a marked trend with the amount smoked, which is highly significant, though the trend was not tested by the authors. In Britain there was little interest in the adverse health effects of tobacco prior to 1947, when the Medical Research Council, at the instigation of Percy Stocks, convened a conference to investigate the cause of increased mortality from cancer of the lung. Although atmospheric pollution was considered as a likely factor, the conference eventually recommended the commissioning of a case-control study of smoking habits by the Council's Statistical Research Unit, headed by Austin Bradford Hill. Bradford Hill recruited Richard Doll, a clinical research assistant at the Central Middlesex Hospital to assist in the study.52 The study was based initially in 20 large hospitals in London. Patients with cancer of the lung, stomach and large bowel, together with control patients without cancer, were interviewed by hospital soaal workers, using a specially designed questionnaire. An initial report on the London patients was available by 1949, but it was decided that the results would not be published until further data could be obtained from hospitals in other cities. However, the publication of the studies in the United States in 1950 led

K

I

375

"The Great Debate"

'

to the publication of the London results later that year73 and the complete study, involving 1,465pairs of lung cancer patients and controls, in 1952.54 The results were consistent with earlier studies, but the large number of patients involved, and the more detailed questions asked, led to a more searching analysis and the elimination of possible confounding factors. After 1950 the number of case-coytrol studies mushroomed. The accumulating evidence from these studies was reviewed successively by Richard Doll75 Harold Dorn,56 and the Surgeon General of the United States.57 Initially the results of the case-control studies were expressed simply in terms of the proportions of smokers among cases and controls (p1 and p2,say), but in a seminal paper in 1951Jerome Cornfield58 showed that the odds ratio, pl/(l - pl): p241 - p2), in a case-control study is an estimate of the ratio of the risk of lung cancer among smokers to that among non-smokers. This paper was the first major contribution to epidemiological methods produced by the problem of lung cancer and smoking. Figure 4 summarizes the results of 24 case-control studies in males and 12 such studies in females. Many of the studies were small so that the estimates of the odds ratio varied considerably. In Figure 4 a weighted mean is shown, calculated by a method described by Barnet Woolf in 1955.59 The risk of lung cancer is greater in males than females, and increases with the amount smoked. Figure 4 Smoking and Lung Cancer Case-control Studies Odds Ratio* '0

1

"

Lees than 20

20 or more

Male8 Weighted mean

mFemales

All

376

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

Cohort Studies 60 Cohort studies, though less prone to bias, are much more difficult to perform than case-control studies, since it is necessary to assemble many thousands of individuals, determine their smoking status, and follow them up for several years to determine how many develop lung cancer. Four such studies were mounted in the 1950s. The subjects used were British doctors,61United States veterans,62 Canadian veterans,63 and volunteers assembled by the American Cancer Society.@All four used mortality as the end-point. Figure 5 shows the combined mortality ratios for cancer of the lung in males by level of cigarette smoking. Two of the studies involved females, but the numbers of lung cancer deaths were too small to provide precise estimates. Since all causes of death were recorded in the cohort studies it was possible to determine the relationship between smoking and diseases other than lung cancer. Sigruficant associations were found in relation to several types of cancer (e.g. mouth, pharynx, larynx, esophagus, bladder) and with chronic respiratory disease and cardiovascular disease. Figure 5 Smoking and Lung cancer Cohort Studies in Males 25

Mortality Ratio*

-1

Less than 10

10 to 19 ,

20 or more

welghied mean of 4 etudles

The epidemiological studies of smoking and cancer of the lung stimulated debates within two areas of discourse: (1) the scientific literature, and (2) the medical establishment and government agencies. The first of these is most relevant to the development of epidemiology as a scientific discipline.

'

"The Great Debate" THE SCIENTIFIC DEBATE

The initial challenges came from three distinguished statisticians:Joseph Berksont5 Jerzy Neyman,66 and Ronald Fishet67All three had made major contributions to both theoretical and applied statistics. Their criticisms can be grouped together under the following counter-propositions: 1.The apparent ass,ociationbetween cigarette smoking and lung cancer is spurious due to "bias" (in the sense of a biased coin rather than a biased judgement), in the selection of subjects, or in the information obtained. 2. The risk ratio (odds ratio, mortality ratio) is an inappropriate measure of association. 3. If the association is not a statistical artefact then it can be explained by a joint association of smoking and lung cancer with some other factor (confounding). 4. The finding that smoking is associated with increased risks of diseases other than cancer makes a causal relationship implausible. 5. There is no clinical (e.g., pathological) or experimental evidence to incriminate cigarette smoking as a cause of lung cancer.

I

Bias Before the lung cancer controversy began, Berkson68had criticized the use of hospital data to claim an association between two diseases, arguing that people with both diseases are more likely to be hospitalised than people with only one disease, producing an apparent association among hospital patients even if none existed in the general [email protected]* suggested that a similar selection bias might occur in smoking studies, not only the hospital-based case-control studies, but also in the recruitment of subjects to cohort studies. Errors in the ascertainment of smoking status or the cause of death could also produce a spurious association. Neyman70 pointed to another, subtler, possible bias in case-control studies. Since the lung cancer patients in the study are survivors from the disease, if those who smoked were more likely to survive (an hypothesis which he admitted to be implausible) then smoking would appear to be associated with the disease, even if the incidence of the disease was no greater among smokers than non-smokers. Risk Ratios vs Risk Differences In two papers in the Journal of the American Statistical Association, the second written with Lila Elveback, Berkson challenged the use of the mortality ratio in the context of competing risks, for example lung cancer and heart disease. In the first article Berkson argues verbally: The exclusive use of this index (the mortality ratio)...tends to bolster up the theory being advanced-the association between lung cancer and smoking-

378

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

and to tone down the association with other causes of death."7*In the second article, which made an important contribution to the theory of competing risks, Berkson and Elveback showed how to estimate, in the context of a cohort study, the net probability of death from lung cancer, i.e., the risk with other causes excluded, which they claimed was the natural measure ohhe effect of smoking. They showed that the mortality difference approximated the net probability of death more closely than the mortality rati0.n

I

Confounding The possibility that the observed association between smoking and lung cancer might be due confounding was emphasized by Fisher. He listed three possible explanations for the association-an enumeration which has since become a routine for epidemiologists. In addition to the hypothesis that smoking causes lung cancer Fisher proposed: "The two classes of alternative theories which any statistical association, observed without the precautions of a definite experiment, always allowsnamely (1)that the supposed effect is really the cause, or in this case that incipient cancer, or a precancerous condition with chronic inflammation, is a factor inducing the smoking of cigarettes, or (2) that cigarette smoking and lung cancer, though not mutually causative, are both influenced by a common cause, in this case the individual genotype."73 Fisher, a geneticist as well as a statistician, pursued the secondalternative theory with data on smoking among monozygotic (identical)and dizygotic twins provided to him by von Verschuer74 and Eliot Slater." Combining the two data sets, 80% of 104 pairs of monozygotic twins were concordant with respect to smoking compared with 41% of 49 pairs of dizygotic twins, a statistically significant difference. Further credence to the genetic hypothesis was provided by the psychologist Hans Eysenck76 who, using a quota sample of 2,360 men aged 40 to 70, showed that cigarette smokers were more extraverted than non-smokers and pipe smokers. Multiple Associations Berkson argued that the finding of multiple associations in the cohort studies cast doubt on the main hypothesis. Smokers appeared to have higher death rates, not only from cancer of the lung, but also from many other cancers, and from other diseases. Either the results are spurious, or they point to some more fundamental biological phenomenon.77

Lack of Clinical and Experimental Evidence Berkson, unlike Neyman and Fisher, was a medical statistician who was more interested in the clinical rather than the public health applications

"The Great Debate"

379

of statistics. His major contributions to the discipline were in the areas of follow-up studies and bioassay. This might explain the apparent incongruence of his scathing attack on the lung cancer studies as being "so exclusively statistical" and "lacking observations of the pathological process of which the statistics are only the supposed reflection."78 In fact, as Berkson acknowledged in a postscriptal footnote, such evidence was beginning to appear. For example, Auerbach and his colleagues, in an autopsy study, found epithelia1metaplasia and atypia (a precursor of cancer similar to that found in screening for cancer of the cervix) to be much more frequent in smokers than non-smokers.79 However, the relevance of Auerbach's findings were subsequently disputed by Little,80on the grounds that similar changes occur in children, and are also found in the trachea, where cancer is extremely rare. Berkson also emphasized that "in science there is no substitute for experiments," and urged "an adequate program of experimentation with animals be accomplished before definitive conclusions are drawn regarding the association between smoking and cancer."al Such experiments, using mice, were, in fact, underway, but they succeeded only in producing laryngeal cancer, bronchial atypia, and pulmonary adenoma.82

I

The Rebuttal The main response to the above criticisms came from a group of American epidemiologists led by Jerome Cornfield.83The authors acknowledged the possibility of selection bias, as suggested by Berkson and Neyman, but pointed out that the latter would only apply to case-control studies, and the former only to hospital-based case-control studies and the initial phases of cohort studies.84 With regard to the choice between relative and absolute differences, the authors argued that both measures served a purpose. The relative measure is more useful in assessing the causal significance of an agent, while the absolute measure is more important in appraising the sign& cance of a causal agent from the public health perspective.85 The methodological superiority of the relative measure was substantiated algebraically in mathematical appendixes, One of these has become a standard criterion in the assessment of confounding variables: "Cigarette smokers have a nine-fold greater risk of developing lung cancer than nonsmokers ....Any characteristic proposed (as an explanatory) cause common to both smoking status and lung cancer risk must therefore be at least nine fold more prevalent among cigarette smokers than among nonsmokers.. ..No such characteristic has yet been produced despite diligent search."86 To counter Berksonls argument regarding multiple associations the authors used the example of the Great Fog of London in 1952, which

380

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

increased the death rate from a number of causes. Since tobacco smoke contains many combustion products it is not unreasonable to suppose that it could have many effects.87 In a later paperF8 Bradford Hill listed nine criteria for deciding when an observed statistical association can be interpreted as causal: the strength of the association; its consistency between studies; its specificity; an appropriate temporality; a "dose-response" effect; biological plausibility; coherence with the natural history of disease; consistency with experimental studies; and, in some circumstances, analogy with related established causal associations. Bradford Hill comments: "None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non. What they can do, with greater or less strength, is to help us make up our minds on the fundamental question-is there any other way of explaining the facts."89 THE MEDICAL DEBATE

l

As the scientific debate progressed-so did discussions within medical institutions and research bodies. Eventually conclusions in favour of a probable causal relation between cigarette smoking and lung cancer were reached by the British Medical Research Council;90 a study gr&p appointed by the U.S. National Institutes of Health and the American Cancer Society;91 the National Cancer Institute of Canada;92 the U.S. Public Health Servi~e;~3 the Royal College of Physicians of London;94 and finally,the first of a long series of reports by the Surgeon General of the United States.95 Some insight into the internal debates which preceded these pronouncements, and their eventual translation into public health policy, has been provided by recent historiography in the United States,96 and in Britain.97 The gist of these papers is summarized nicely in a letter to Ernst Wynder from Sir Richard Doll: I think the sceptical reaction of the medical and cancer research scientists was partly because they smoked themselves and partly because they were unaccustomed to the interpretation of epidemiologic data and tended to judge causality by Koch's postulates. Advisors to the government were pathologically scared of causing cancer phobia by undertaking any publicity about cancer, even to the extent of opposing education about the need for early diagnosis. Within government there was anxiety about the effects of reduced sales on tax income and there was certainly a desire to work with the industry rather than against it.98

"The Great Debate" THE AFTERMATH OF THE DEBATE

I

By 1960 the debate was virtually over from the epidemiological perspective. A few dissidents remained. The genetic hypothesis was pursued by Car1 Selzer, who found anthropomorphic differences between smokers and non-smokers.99 However, studies of blood groupslw and maternal age101 found ho differencesbetween lung cancer patients and controls. The arguments against a causal interpretation were restated by a medical physicist, Philip Burch, as late as 1978.102 An "echo" debate among epidemiologists was stimulated by Jan Vandenbroucke's 1989 article entitled "Those who were wrong."l03 As Vandenbroucke concluded: "The reading and re-reading of those great scientists who were wrong, along with the papers of those who were right, will teach us much about the power of the epidemiologic method, which is so heavily rooted in the smoking-lung cancer controversy."l" Furthermore, as Mervyn Susser105 has emphasized, the papers by Cornfield,l06 Woolf,'w Cornfield et al.,lm Bradford Hill,'@ together with a 1959paper by Nathan Mantel and WilJiam Haenszelllo on the analysis of case-control studies, provided the kernel of text books on epidemiologic methods. Academically, epidemiology has displaced public health in medical schools, and in Britain the Medical Officer of Health no longer exists, having metamorphosed into a Specialist in Community Medicine, whose role is essentially that of an epidemiological consultant.111 As Dorothy Porter writes in her recent review of the history of public health: "The analysis of the relationship between smoking and lung cancer gave epidemiology a new credence within the biomedical model of chronic disease. It became a critical heuristic device and legitimated a new approach to prevention which hiwghted the role of individual behaviour and lifestyle in sharp contrast to the preventive model promoted by nineteenth century epidemiology, which focused on social structural methods of preventing epidemics."ll2 The "Great Debate" was the crucible in which the tools of a new discipline were forged. This new branch of medical science laid the foundation for a new approach to the control of cancer and other non-communicable diseases. NOTES 1 Editorial, "The Great Debate," New England Journal of Medicine, 264 (1961): 1266. 2 Sir Richard Doll, "MajorEpidemics of the 20th Century: From Coronary Thrombosis to AIDS," Journal of the Royal Statistical Society, Series A, 150 (1987):373-95. 3 Health Canada, "Lung Cancer in Canada," Cancer Updates, 2001, p. 1. 4 Health Canada, "Lung Cancer," p. 4. 5 Statistics Canada, Canada Year Book 1999, (Ottawa:Minister of Industry, 1998), p. 241. 6 Bernadino Ramazzini, De Morbis Art@curn Diatriba (Geneva, 1713). 7 Rigoni Stern, Giornale della Pathologia e della Terapeutica, 2 (1842):507-17.

382

I

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

8 John Hill, Cautions against the Immoderate Use of Snuff (London, 1761). 9 Percivall Pott, Chirurgical Observations Relative to the Cataract, the Polypus of the Nose, the Cancer of the Scrotum etc. (London: Hawes, Clarke and Collins, 1775). 10 Sigismund Peller, Cancer in Man (New York: International Universities Press, 1952), p. 108. 11 Milton Terris, "The Society of Epidemiological Research (SER) and the Future of Epidemiology," American Journal of Epidemiology, 136 (1992): 909-15. 12 Arthur Newsholme, Vital Statistics, 3rd ed. (London: Swan Sonnenschein, 1899), p. 19. 13 Newsholme, Vital Statistics, p. 241-45. 14 Percy Stocks, "Cancer Mortality Trends in England and Wales," in Ronald Raven, ed., Cancer, Vol. 3 (London: Butterworth, 1958), p. 184-207. 15 H. J. B. Fry, "Syphilis and Malignant Disease, A Serological Study," British Journal of Hygiene, 29 (1929): 313-22. 16 Percy Stocks and Mary N. Karn, "A Co-operative Study of the Habits, Home Life, Dietary and Family Histories of 450 Cancer Patients and of an Equal Number of Control Patients," Annals of Eugenics (London), 5 (1933): 237-80. 17 Austin Bradford Hill and Lewis Faning, "Studies in the Incidence of Cancer in a Factory Handling Inorganic Compounds of Arsenic. I-Mortality Experience in the Factory," British Journal of Industrial Medicine, 17 (1948): 493-504. 18 W Machle and F. Gregorius, "Cancer of the Respiratory System in the United States Chromium Producing Industry," Public Health Reports, 63 (1948): 1114-27. 19 I. Adler, Primary Malignant Growths of the Lungs and Bronchi (New York: Longmans, Green and Company, 1912). 20 F. H. Harting and W. Hesse, "Der Lungenkrebs, die Bergkrankheit in den Schneeberger Gruben," Vierteljahrschrift fiir gerichtliche und 6ffentliche Medizin, 30 (1879):296-309; 31 (1879): 102-33,313-37. 21 Rudolf Virchow, "Sitzung der physikalisch-medizinischen Gesellschaft &U Wiirzburg," reprinted in Collected Essays on Public Health and Epidemiology, Vol. 1, translated by L. J . Rather (Canton, Mass.: Watson, 1985), p. 515-40. 22 Walter Kikuth, "Uber Lungencarcinom," Virchows Archiv fiir pathologische Anatomie und Physiologie undfiir klinische Medizin, 255 (1923): 107-28. 23 Car1 V. Weller, "The Pathology of Primary Carcinoma of the Lung, Archives of Pathology, 7 (1929): 478-519; Frederick L. Hoffman, "Cancer of the Lungs," American Review of Tuberculosis, 19 (1929): 392-406; B. M. Fried, "Primary Carcinoma of the Lung," Medicine, 10 (1931): 373-508; Georgiana M. Bonser, "The Incidence of Intrathoracic Tumours in Leeds," Journal of Hygiene, 34 (1934); 218-34; Alton Ochsner and Michael DeBakey, "Carcinoma of the Lung," Archives of Surgery, 42 (1941): 20958; and M. E. Daff, R. Doll and E. L. Kennaway, "Cancer of the Lung in Relation to Tobacco," British Journal of Cancer, 5 (1951): 1-20. 24 Rupert A. Willis, Pathology of Tumours, 4th ed. (London: Butterworth's, 1967), p. 354. 25 Registrar General of England and Wales, Statistical Review (London: Her Majesty's Stationery Office, 1923), Text Table XLIV. 26 Percy Stocks, "Cancer Mortality Trends," p. 200. 27 A. J. Phillips, "Mortality from Cancer of the Lung in Canada," Canadian Medical Association Journal, 71 (1954): 242-44. 28 By the direct method, applying the age-specific rates to the IUCC standard European Population. 29 Richard Peto, Alan Lopez, Jillian Boreham, Michael Thun, and Clark Heath Jr, Mortality from Smoking in Developed Countries 1950-2000: Indirect Estimates from National Vital Statistics, (Oxford: Oxford University Press, 1994) p. 526-27. 30 Statistics Canada, Mortality-Summary List of Causes Catalogue 84-209 (Ottawa: Minister Responsible for Statistics Canada, various years). 31 Percy Stocks, "Recent Epidemiological Studies of Lung Cancer Mortality, Cigarette Smoking and Air Pollution, with Discussion of a New Hypothesis of Causation," British Journal of Cancer, 20 (1966): 595-623.

"The Great Debate"

I

32 International Union against Cancer, Age-Adjusted Death Ratesfor Malignant Neoplasm for Selected Sites, by Sex, in 33 Countries of the World (from 1953-57 to 1988-92) http://publications.iucc.org/99stats/partl.pdf 33 Stocks, "Cancer Mortality Trends," p. 206. 34 A. J. Phillips, 'h Analysis of the Increase in Lung Cancer in Canada," Canadian Medical Association Journal, 95 (1966): 1172-74. 35 The cohort born around 1911 was taken as the reference cohort. For other cohorts the ratio of the mortality rate at each age to that of the reference cohort at the same age was calculated. The geometric mean of the ratios for each cohort was plotted. 36 James Ewing, "The Causal and Formal Genesis of Cancer," in British Empire Campaign, Report of the International Conference on Cancer (Bristol: John Wright and Sons, 1928), p. 1-13. 37 Weller, "Pathology," p. 499. 38 Weller, "Pathology," p. 497. 39 Comment by Fahr (1923) cited by Ochsner and Debakey, "Carcinoma of the Lung," p. 219. 40 Editorial, "Cancer of the Lung," Lancet, 2 (1927): 125. 41 A study in which the prevalence of prior exposure to a potential etiological factor among those with a particular disease (cases) is compared with the prevalence of exposure to the factor among those without the disease. 42 Robert N. Proctor, "The Nazi War on Tobacco: Ideology, Evidence, and Possible Cancer Consequences," Bulletin of the History of Medicine, 71 (1997): 435-88. 43 Franz H. Miiller, "Tabakmissbrauch und Lungencarcinom," Zeitschrift fur Krebsforschung, 49 (1939): 57-85. 44 Eberhard Schrairer and Erich Schoniger, "Lungenkrebs und Tabakverbrauch," Zeitschriftfiir Krebsforschung, 54 (1943): 261-69. a 45 Cassandra Tate, Cigarette Wars: The Triumph of the Little White Slaver (New York: Oxford University Press, 1999), p. 93. 46 James T. Patterson, The Dread Disease: Cancer and Modern American Culture, (Cambridge, Mass.: Harvard University Press, 1987), p. 205. 47 Ernst L. Wynder, "Tobacco as a Cause of Lung Cancer: Some Reflections." American Journal of Epidemiology, 148 (1997): 687-94. 48 Ernest [sic] L. Wynder and Evarts A. Graham, "Tobacco Smoking as a Possible Etiologic Factor in Bronchogenic Carcinoma. A Study of Six Hundred and EightyFour Proved Cases," The Journal of the American Medical Association, 143 (1950): 32936. 49 Morton L. Levin, Hyman Goldstein, and Paul R. Gerhardt, "Cancer and Tobacco Smoking. A Preliminary Report," The Journal of the American Medical Association, 143 (1950): 336-38. 50 H. K. Armenian and M. Szklo, "Morton Levin (1904-1995): History in the Making," American Journal of Epidemiology, 143 (1996): 648-49. 51 Warren Winkelstein Jr., "Observations on the History of Epidemiology in Western New York, 1843-1960," American Journal of Epidemiology, 146 (1997): 896-906. 52 Sir Richard Doll, "The First Reports on Smoking and Lung Cancer," in S. Lock, L. Reynolds and E. M. Tansey, eds., Ashes to Ashes: The History of Smokingand Health, (AmsterdadAtlanta, Georgia: Editions Rodopi, 1998), p. 130-40. 53 Richard Doll and A. Bradford Hill, "Smoking and Carcinoma of the Lung. Preliminary Report," British Medical Journal, 2 (1950): 84-92. 54 Richard Doll and A. Bradford Hill, 'K Study of the Aetiology of Carcinoma of the Lung," British Medical Journal, 2 (1952): 1271-86. 55 Richard Doll, "Bronchial Carcinoma: Incidence and Aetiology," British Medical Journal, 2 (1953): 521-27,585-90. 56 Harold F,Dorn, "The Relationship of Cancer of the Lung and the Use of Tobacco," American Statistician, 8 (1954): 7-13.

.384

I

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

57 Surgeon General, Health Consequences of Smoking. A Reference Edition (Atlanta, Georgia: US. Department of Health Education and Welfare, 1976). ' Method of Estimating Comparative Rates from Clinical Data. 58 Jerome Cornfield, !A Application to Cancer of the Lung, Breast and Cervix," Journal of the National Cancer Institute, 11 (1951):1269-75. 59 Barnet Woolf, "On Estimating the Relation between Blood Group and Disease," Annals of Human Genetics, 19 (1955):251-53. 60 A study in which the subsequent incidence of, or mortality from, a particular disease among subjects exposed to a potential etiological factor is compared with the incidence or mortality among subjects not exposed to the factor. 61 Richard Doll and Sir Austin Bradford Hill, "Mortality in Relation to Smoking: Ten Years Observations of British Doctors," British Medical Journal, 1 (1964): 1399-410, 1460-67. 62 Harold Kahn, "The Dorn Study of Smoking and Mortality among U.S. Veterans. Report on Eight and One-Half Years of Observatians," in William Haenszel, ed., Epidemiological Approaches to the Study of Cancer and Other Chronic Diseases (Bethesda, Md.: National Cancer Institute, 1966), p. 1-125. 63 E. R. Best, G. H. Josie, and C. B. Walker, '!A Canadian Study of Mortality in Relation to Smoking Habits. A Preliminary Report," Canadian Journal of Public Health, 52 (1961):99-106. 64 E. Cuyler Hammond, "Smoking in Relation to the Death Rates of One Million Men and Women," in Haenszel, Epidemiological Approaches, p.127-204. 65 Joseph Berkson, "The Statistical Study of Association between Smoking and Lung Cancer," Proceedings of the StaffMeetings of the Mayo Clinic, 30 (1955):319-48; "Smoking and Lung Cancer: Some Observations on Two Recent Reports," Journal of the American StatisticalAssociation, 53 (1958):28-38; "Smoking and Cancer of the Lung," Proceedings of the StaffMeetings of the Mayo Clinic, 35 (1960):367-85; with Lila Elveba&, "Competing Exponential Risks, with Particular Reference to the Study of Smoking and Lung Cancer," Journal of the American Statistical Association, 55 (1960):415-28. 66 Jerzy Neyman, "Statistics-Servant of All Sciences," Science, 122 (1955): 401-6. 67 Ronald A. Fisher, "Dangers of Cigarette-Smoking," British Medical Journal, 2 (1957): 43, 297-98; "Lung Cancer and Cigarettes," Nature, 182 (1958):108; "Cancer and Smoking," Nature, 182 (1958):596. 68 Joseph Berkson, "Limitations of the Applications of Fourfold Table Analysis to Hospital Data," Biometrics Bulletin, 2 (1946): 47-53. 69 Berkson, "Statistical Study," p. 324-28. 70 Neyman, "Statistics," p. 404-5. 71 Berkson, "Smoking and Lung Cancel;" p. 30. 72 Berkson and Elveback, "Competing," p. 423-28. 73 Fisher, "Dangers," p. 297-98. 74 Fisher, "LungCancer and Cigarettes," p. 108. 75 Fisher, "Cancer and Smoking," p. 596 76 H. J. Eysenck, Mollie Tarrant, Myra Woolf, and L. England, "Smoking and Personality," British Medical Journal, 1 (1960): 1456-60. 77 Berkson, "Smoking and Cancer of the Lung," p. 383-84. 78 Berkson, "Statistical Study," p. 339. 79 0. Auerbach, J. B. Gere, J. B. Farmer, T G. Petrick, H. J. Smolin, G. E. Muesham, D. Y. Kassouny and A. I! Stout, "Changes in the Bronchial Epithelium in Relation to Smoking and Cancer of the Lung: A Report of Progress," New England Journal of Medicine, 256 (1957):97-104. 80 Clarence C. Little, "Some Phases of the Problem of Smoking and Lung Cancer," New England Journal of Medicine, 264 (1961):241-45. 81 Berkson, "Statistical Study," p. 342. 82 John A. Baron and Thomas E. Rohan, "Tobacco," in David Schottenfeld and Joseph F. Fraumeni Jr., eds., Cancer Epidemiology and Prevention, 2d ed. (New York: Oxford University Press, 1996), p. 277.

"The Great Debate" 83 Jerome Cornfield, William Haenszel, E. Cuyler Hammond, Abraham M. Lilienfeld, Michael B. Shimkin, and Ernst L. Wynder, "Smoking and Lung Cancer: Recent Evidence and a Discussion of Some Questions," Journal of the National Cancer Institute, 22 (1959): 173-203. 84 Cornfield, et al., "Smoking and Lung Cancer," p. 181-83. 85 Cornfield, et al., "Smoking and Lung Cancer," p. 194. 86 Cornfield, et al., "Smoking and Lung Cancer," p. 194. The argument is the equivalent, for discrete variables, of the following proposition for continuous variables: if the (positive) correlation between two variables is to be explained by their (positive) correlations with a third variable, i.e., the partial correlation coefficient r12.3is zero, then rlz r l 3 r ~must be zero, i.e. rl3 = r1drz3 >r12. 87 Cornfield et al., "Smoking and Lung Cancer," p. 196. 88 Sir Austin Bradford Hill, "The Environment and Disease: Association or Causation," Proceedings of the Royal Society of Medicine, 58 (1965): 295-300. 89 Bradford Hill, "Environment," p. 299. 90 Medical Research Council, "Tobacco Smoking and Cancer of the Lung. Statement by the Medical Research Council," British Medical Journal, 1(1957): 1523-24. 91 Study Group on Smoking and Health, "Smoking and Health," Science, 125 (1957): 1129-33. 92 National Cancer Institute of Canada, "Lung Cancer and Smoking," Canadian Medical Association Journal, 79 (1958): 566-68. 93 Larry E. Burney, "Smoking and Lung Cancer. A Statement of the Public Health Service," Journal of the American Medical Association, 171 (1959): 1829-37. 94 Royal College of Physicians, Smoking and Health. A Report of The Royal College of Physicians of London on Smoking in Relation to Cancer of the Lung and Other Diseases (London: Pitman Medical Publishing, 1962). 95 Surgeon General, Smoking and Health: Report of the Advisory Committee to the Surgon General of the Public Health Service, Public Health Service Publication No. 1103 (Washington, DC.: US. Department of Health, Education and Welfare, Public Health Service, 1964). 96 John C. Burnham, "American Physicians and Tobacco Use: Two Surgeons General, 1929 and 1964," Bulletin of the Histoy of Medicine, 63 (1989): 1-31; Ernst L. Wynder, "Tobacco as a Cause of Lung Cancer: Some Reflections," American Journal of Epidemiology, 146 (1997): 687-93; and Marc Parascandola, "Cigarettes and the US Public Health Service in the 1950s: American Journal of Public Health, 91 (2001): 196-203. 97 Sir Richard Doll, "The First Reports," p. 130-40; Sir Christopher Booth, "Smoking and the Royal College of Physicians," in Lock, et al., Ashes to Ashes, p. 192-97; Virginia Berridge, "Science and Policy: The Case of Postwar British Smoking Policy," in Lock, et al., Ashes to Ashes, p. 143-62; and E Cowan, "The Price of Coffins: Specious Arguments by Eminent Doctors against the Dangers of Tobacco," British Medical Journal (BMJ), 319 (1999): 1621-23. 98 Quoted in Wynder, "Tobacco," p. 692. 99 Car1 C. Selzer, "Morphologic Constitution and Smoking," Journal of the American Medical Association, 183 (1963): 639-45. 100 David J. B. Ashley, "Blood Groups and Lung Cancer," Journal of Medical Genetics, 6 (1969): 183-86. 101 Theodor Abelin and George K.Tokuhata, "Maternal Age At Birth and Susceptibility to Lung Cancer," Lancet, 2 (1965): 1121-23. 102 Philip R. J. Burch, "Smoking and Lung Cancer: The Problem of Inferring Cause," Journal of the Royal Statistical Society A, 141 (1978): 437-77. 103 Jan E Vandenbroucke, "Those Who Were Wrong," American Journal of Epidemiology, 130 (1989): 3-5. 104 Vandenbroucke, "Those," p. 5. 105 Mervyn Susser, "Epidemiologyin the United States after World War 11: The Evolution of Technique," Epidemiologic Reviews, 7 (1985): 147-77.

-

I

386

GERRY B. HILL, WAYNE MILLAR and JAMES CONNELLY

Cornfield, "A Method of Estimating." Woolf, "On Estimating." Cornfield, et al., "Smoking and Lung Cancer." Bradford Hill, "Environment and Disease." Nathan Mantel and William Haenszel, "Statistical Aspects of the Analysis of Data from Retrospective Studies of Disease," Journal of the National Cancer Institute, 22 (1950): 719-48. 111 Jim Connelly and Chris Worth, Making Sense of Public Health Medicine (Oxford: Radcliffe Medical Press, 1997) p. 6-7. 112 Dorothy Porter, Health, Civilization and the State: A History of Public Health from Ancient to Modern Times (London: Routledge, 1999), p. 300. 106 107 108 109 110