The Cigarette Century

by Allan Brandt

  There are undoubtedly great numbers of human beings who would continue the habitual use of a particular material they liked, even though it were absolutely and beyond any question or argument proved to be somewhat deleterious to them.93

  Although Pearl’s study might have strong statistical implications for life insurers, its meaning for individual smokers remained less clear. Did smoking lead to specific diseases that resulted in premature death? Or was it possible that less-healthy individuals might be predisposed to smoke? Were some individuals particularly vulnerable to the effects of the cigarette? Pearl’s study did not directly address these questions.

  Others, however, now suggested that such questions could be addressed through more systematic comparisons of smokers to nonsmokers. One such study, excerpted in the Journal of the American Medical Association, came from Germany. Franz Hermann Müller, a physician from the University of Cologne’s Pathological Institute, identified 96 individuals who had died of lung cancer. Of these individuals, he was able to determine that only three had been nonsmokers; about one-third smoked moderately. Müller’s study was important in that he had compared smokers to a control group who had died of other causes.94 Statistical observations like Pearl’s and Müller’s sustained concerns about the impact of smoking through the 1930s.

  Those who treated diseases of the lung saw smoking as a daily issue. Chest surgeons like Alton Ochsner in New Orleans and Richard Overholt in Boston drew attention in the 1930s with their observations that patients with advanced lung malignancies typically had smoked. Ochsner prohibited his staff from smoking and became known as a vigorous antitobacco advocate. He and colleague Michael DeBakey, assessing the increase in cases of primary carcinoma of the lung, concluded:In our opinion the increase in smoking with the universal custom of inhaling is probably a responsible factor, as the inhaled smoke, constantly repeated over a long period of time, undoubtedly is a source of chronic irritation to the bronchial mucosa.95

  According to most prevailing explanations of carcinogenesis, irritation played a crucial role in instigating the development of tumors, though exactly how was not known.

  But this statement powerfully revealed the limits of clinical observation, a hallmark of medical science and investigation. There was no indication that DeBakey and Ochsner systematically collected information on their patients’ smoking patterns. They possessed no method for turning clinical observations, however acute, from causal hypothesis into fact. Evarts Graham, a prominent surgeon at Barnes Hospital in St. Louis, who had trained Ochsner, told him, “Yes there is a parallel between the sale of cigarettes and lung cancer, but there is also a parallel between the sale of silk stockings and cancer of the lung.”96 The observations of thoracic surgeons like Overholt and Ochsner left the underlying causal hypothesis unresolved.97

  These clinical observations of the impact of smoking are, in retrospect, quite impressive. Almost all the risks that would later come to be attributed to smoking had been well documented by clinicians in the first decades of the century. Even the risks of passive exposure to cigarette smoke had been well articulated.98 Yet physicians and researchers could not move from such clinical observations to more powerful and generalizable assessments of the relationship of smoking to disease. Surgeons like Ochsner might well be convinced that tobacco had caused their patients’ malignancies, but their observations could never settle the larger questions of cause and effect.

  But standard experimental methods exported from the laboratory could not settle the question either. Not only would it be impossible to design the requisite experiment—randomizing children into groups of smokers and controls—it was widely recognized that such an approach would also be unethical.99 As Evarts Graham humorously explained in a 1954 commentary that appeared in the Lancet: One must grant that indeed no absolute proof has been offered. But what sort of proof is called for? To satisfy the most obdurate of the diehards it would be necessary to take the following steps:1. Secure some human volunteers willing to have a bronchus painted with cigarette tar, perhaps through a bronchial fistula.

  2. The experiment must be carried on for at least twenty or twenty-five years.

  3. The subjects must spend the whole period in air conditioned quarters, never leaving them even for an hour or so, in order that there may be no contamination by a polluted atmosphere.

  4. At the end of the twenty-five years they must submit to an operation or an autopsy to determine the result of the experiment.

  I will say to those who wish to volunteer for such an experiment, “please form a queue to the right; no crowding please.”100

  By the 1940s, several innovative and suggestive studies offered impressive support for the theory of cigarettes as a cause of disease. And yet, these studies did not constitute “proof ” in its scientific, medical, and social definition. Nor did they have any significant effect on consumption, medical views of the cigarette, or policy. They served as little more than scientific footnotes in an ongoing debate about the impact of smoking. Their implication for doctors and patients remained far from clear. Many physicians noted that medical studies of the health effects of smoking produced ambiguous and often contradictory findings. Although the moral opprobrium directed at the cigarette had dissipated, it left behind a substantial but quiet medical uncertainty. In this state of uncertainty, doctors and the public would be forced to confront a new epidemic: lung cancer.

  If the evidence incriminated only an article of diet, such as spinach, for example, probably nearly everybody would accept it as conclusive. I have never encountered any non-smoker who makes light of the evidence or is skeptical of the association between excessive cigarette smoking and lung cancer.1

  EVARTS A . GRAHAM, 1954

  Medical literature has numerous examples of such fallacious conclusions which have been proved to be wrong in the light of subsequent experience. This whole question of cause and effect deducted on a statistical basis is subject to the greatest fallacies. One way I like to emphasize it is to say that simply because one finds bull frogs after a rain does not mean that it rained bullfrogs.2

  MAX CUTLER, 1954

  Perhaps it is never possible to prove with absolute certainty

  that a naturally occurring event was the product

  of a specified set of naturally occurring conditions.

  However, such can often be established beyond reasonable

  doubt. In human affairs, important decisions

  must necessarily be based upon the preponderance of

  evidence.3

  E. CUYLER HAMMOND, 1955

  CHAPTER 5

  The Causal Conundrum

  IN 1933, AT BARNES Hospital in St. Louis, Dr. Evarts Graham performed the first successful pneumonectomy—total removal of a I cancerous lung. The procedure he devised required significant innovations in surgical technique and offered the best available treatment for bronchogenic carcinoma, well known to be uniformly fatal. Following his triumphant announcement of the procedure at the annual meeting of the American Association for Thoracic Surgery, pneumonectomy became the treatment of choice for lung cancers that had yet to spread to other organs. Graham’s professional reputation soared.4

  Despite his extensive experience treating individuals with cancers of the lung, Graham had greeted Alton Ochsner’s theory about the relationship of such cancers to smoking with a mix of skepticism and derision. A heavy smoker who had suffered no apparent ill effects, he believed that if Ochsner’s ideas were correct, surgeons would more typically see tumors in both lobes. Given that cancers were typically found in but one lobe, Graham reasoned, smoking was unlikely to be the cause. Why, when the smoke obviously entered both lobes, would only one be affected?5

  Despite these views, or perhaps because of them, Graham was sympathetic when a third-year medical student at Washington University, Ernst Wynder, approached him in 1947 to collaborate on a study of the relationship of smoking to lung cancer. Graham was eager to resolve the lurking suspicions that by t
hen had surrounded the cigarette for more than a decade. Wynder had spent the previous summer on his own initiative in New York collecting data from some 146 lung cancer patients about their smoking practices. Impressed by this student’s independence and fortitude, Graham agreed to participate. Even more importantly, he contacted the American Cancer Society (ACS), where he had a number of influential colleagues, to obtain modest funding for the work.6 Wynder threw himself into the investigation, pulling the senior surgeon along with him.

  Wynder quickly became something of a shoe-leather epidemiologist, traveling widely to interview lung cancer patients about their smoking histories and other toxic exposures. Wynder and Graham collected extensive data on 604 such patients, located at hospitals across the United States. Their approach to the question of whether cigarette smoking caused lung cancer was rooted in the historical tradition of the anecdotal collection of cases, well known to surgeons who would report successive surgeries to demonstrate their effectiveness. But following a practice that was growing increasingly common in clinical research, they also decided to apply rudimentary statistical techniques to their evaluations.7

  Unlike earlier researchers, who had separated smokers into crude categories of “moderate” and “heavy,” Wynder and Graham devised categories that carefully reflected levels of smoking over a twenty-year period, rating smokers in five groups from “light” to “chain.” They arranged for histological examination of cells from lung tissue in each case to confirm the cancer diagnosis.8 Cases of lung cancer without a history of tobacco use proved exceedingly rare, and in these instances Wynder often uncovered exposures to other inhaled carcinogens, such as gasoline fumes or insecticides.9 Wynder and Graham also attended to a number of the questions that had previously confounded investigators. They evaluated the types of tobacco their subjects used, noting that cigarettes were more frequently inhaled and used more heavily than pipes and cigars.

  Notably, Wynder and Graham established a control group of cancer-free individuals in hospitals for systematic comparison to their lung cancer patients. Interviewers carefully inquired about the smoking patterns of this group using the exact same questionnaire for lung cancer patients. “Two objects were realized by this control group,” they explained. “One was to learn of possible exposures to exogenous irritants of a large group of patients without cancer of the lung and the other to test the validity of the interviews made of those who knew the suspected diagnosis in a given case in advance.”10 Wynder and Graham assumed that the age distribution in the study group and control groups were comparable, and made adjustments where appropriate. They were eager that their findings stand up to statistical scrutiny.

  At the Journal of the American Medical Association (JAMA), editor Lester King reportedly greeted the resulting paper with considerable suspicion. He was eventually convinced of its merits by Morton Levin, a physician and epidemiologist at the State Institute for the Study of Malignant Diseases in New York, who noted that the paper’s statistical approach emerged from the historical tradition of case study and observation.11 JAMA published the paper on May 27, 1950.

  Wynder and Graham reported that lung cancer could occur among nonsmokers and that heavy smokers did not necessarily develop cancer. Therefore, they reasoned, “smoking cannot be the only etiologic factor in the induction of disease.” Nonetheless, “the temptation is strong to incriminate excessive smoking, and in particular cigarette smoking, over a long period as at least one important factor in the striking increase of bronchogenic carcinoma.”12 They offered four reasons to support this conclusion. First, it was very unusual to find lung cancer among nonsmokers. Second, among patients with lung cancer, cigarette use tended to be high. Third, the distribution of lung cancer among men and women matched the ratio of smoking patterns by gender. Finally, “the enormous increase in the sale of cigarettes in this country approximately parallels the increase in bronchogenic carcinoma.”13

  Wynder, idealistic and ambitious, was impressed and excited by the potential significance of his findings and saw them—correctly—as his chance to make a mark in medicine. “The data are most impressive,” he enthused in a letter to Graham. “Our final paper will ring forth startling as well as decisive news indeed. It lies within our hands to lower considerably the incidence of one of the major cancers.” 14 He had become so focused on completing the study that he apparently neglected his medical school work.15

  Born in Germany in 1922 to Jewish parents, Wynder had escaped with them from Nazi rule, arriving in the United States in 1938. He worked his way through New York University selling newspapers and waiting tables. During the war, he attained citizenship and entered the Army. His knowledge of German led to his assignment in a psychological warfare unit monitoring German newscasts. Following the war, Wynder followed his father into medicine, enrolling at Washington University School of Medicine. In Graham, he found a mentor eager to foster his strengths.16

  As the results of their study began to come in, Graham wrote to Ochsner, acknowledging that “I may need to eat humble pie.”17 Soon, Graham and Ochsner would become steadfast allies in their conviction that smoking was an important cause of lung cancer. With a push from his student, Graham employed the tools of epidemiology to transcend the limits of his own clinical observations. He would be forever associated as a key player in this path-breaking finding.

  As Wynder traveled around the United States, visiting hospitals and collecting data, two distinguished medical statisticians in Great Britain had independently embarked on a parallel study. Under the auspices of the British Medical Research Council (MRC), which had recently become a unit of the newly created National Health Service (NHS), A. Bradford Hill and Richard Doll began in 1947 to study the rising incidence of lung cancer. As they both noted, there was no easy way to account for the notable rise of diseases, such as lung cancer. On both sides of the Atlantic, the selection of this disease as an area for investigation reflected rising concern about the chronic systemic diseases of later life in the face of demographic changes clearly recognized at mid-century. Among the central tasks of the NHS would be to monitor and prevent chronic diseases, which had over the course of the twentieth century become increasingly prominent causes of morbidity and mortality. While Wynder and Graham initiated their study of the disease from a clinical perspective, their English counterparts worked within an epidemiological tradition deeply rooted in the investigation of vital statistics and population health.18

  Doll and Hill recognized that certain questions of great medical and public significance simply would not readily succumb to clinical evaluation and laboratory experiments. Nonetheless, the timeliness and importance of determining the factors leading to cancers warranted immediate attention. As a result, they joined together to utilize a sophisticated set of statistical and medical skills that they had developed independently over the previous decades to implement a rigorous investigation of the relationship of smoking to cancer.

  It had been expected that Hill, like his father before him, would pursue medical training and practice. But as a pilot in the Royal Navy Air Service in World War I, he contracted tuberculosis and was “sent home to die.” Defying the probabilities, Hill survived, but a medical career was now out of the question. Major Greenwood, the leading medical statistician in Great Britain and a friend of Hill’s father, recommended he pursue a correspondence degree in economics. This Hill accomplished while still bedridden. As his recovery proceeded, Greenwood arranged for him to undertake a field investigation of the reasons for high mortality rates in rural Essex. While conducting this study, Hill continued his education under Greenwood’s watchful eye; he also attended a course on statistics at University College taught by Karl Pearson, the noted mathematician and eugenicist.19

  In 1923, Hill was hired by the MRC to study occupational disease. While at the MRC, Hill lectured at the London School of Hygiene and Tropical Medicine on medical statistics and its applications. These lectures were published in the Lancet in 1937; collected i
n book form, they provided the basis for the epidemiologic techniques that would become the center of Hill’s subsequent investigations.20 The articles emphasized the need to make appropriate comparisons, to avoid bias, and to understand the role of chance—all principles central to modern epidemiologic methods. Hill urged the medical profession to apply new forms of scientific deduction and quantification in their assessment of medical practices and patterns of disease.21 Rather than dismissing clinical knowledge as anecdotal and idiosyncratic, Hill sought to develop systematic strategies for utilizing and assessing hard-won clinical observation. This work was a key element in the effort to place medicine, public health, and therapeutics on a new evidentiary foundation.22

  After World War II, Hill championed the utility of medical statistics for the evaluation of new treatments—following the pioneering work of the British statistician and geneticist R. A. Fisher, whose work in agricultural experimentation had been a great influence. One of Hill’s innovations was the first randomized, double-blind clinical trial, designed to reduce investigator bias in the evaluation of clinical outcomes. Attempting to assess the effectiveness of streptomycin in treating infectious tuberculosis, Hill divided patients with TB into two groups: one received streptomycin, the other a placebo. Neither researcher nor subject was informed about who was getting the drug—hence the term double-blind. As a result, evaluations of the drug’s efficacy would not be shaped by prior knowledge, expectations, or unknown biases. This method, which drew on Fisher’s agricultural experimentation in genetics, became a critical new tool for evaluating medical treatments.23 Hill would apply elements of this same framework to investigating the relationship of smoking to disease.