by Allan Brandt
They followed this group forward, noting deaths through the Registrars-General Office in the United Kingdom. The prospective study demonstrated Doll and Hill’s experimental approach to epidemiology. By gathering two groups similar in every respect except for their smoking behaviors, they had created an experiment—in effect, a randomized trial in which the “intervention” was cigarette use rather than a therapeutic agent. Now they awaited the effects of the intervention. As the data came in, it proved fully consistent with their previous research: they found an impressive excess of deaths among the doctors who smoked. Strikingly, heavy smokers had death rates 24 times higher than nonsmokers.47
In a preliminary report on the study, published in the British Medical Journal in 1954, they remarked
We thought it necessary, in view of the nature of our results, to lay these preliminary data before the survivors of the 40,000 men and women who made them possible.48
Impressing these survivors was central to legitimating both their findings and their methods: if the medical profession was convinced, it would persuade individual doctors to advise patients against smoking and also provide authority to the new epidemiological methods of inquiry.
In the United States, statisticians E. Cuyler Hammond and Daniel Horn simultaneously conducted a similar prospective study under the auspices of the American Cancer Society. The antismoking advocate Alton Ochsner had become president of the ACS, and he now pressed the organization for a more systematic assessment of changing rates of cancer mortality. For this work, the ACS turned to Hammond, who had received his doctorate at the Johns Hopkins School of Hygiene and Public Health, where he had come under the influence of Wade Hampton Frost, Lowell Reed, and Raymond Pearl, dominant figures in pre-war epidemiology and medical statistics. In 1946, he went to the ACS to lead its Department of Statistical Research. Hammond had been among the most vociferous critics of Wynder and Graham’s initial retrospective study. In December 1950, a frustrated Evarts Graham called Hammond a “two for a nickel guy” who had “done everything [he] could to obstruct this work.”49
Now, however, Hammond and Horn, a psychologist with training in statistics, worked to design a trial free of the potential limitations of retrospective studies. They focused on two primary ways to reduce bias: a prospective design, like that used in Doll and Hill’s physician study, and perhaps more significantly, the largest study population anyone had yet assembled. Over nearly four years, Hammond and Horn followed a sample of 200,000 men between the ages of fifty and sixty. During this period, 12,000 died. Not only was lung cancer a far more prevalent cause of death among those who smoked (twenty-four times more common than among nonsmokers), so too was heart and circulatory disease.50
Even though Hammond had been dismissive of the retrospective studies, as his prospective data came in he grew increasingly convinced of the causal relationship. “All the evidence collected to date,” he wrote to Graham in 1954, “certainly points strongly to the conclusion that cigarette smoking does increase the probability of developing lung cancer.”51 That September, Graham wrote to Ochsner concerning Hammond’s change of heart:I am very happy that Hammond has completely reversed himself from the position which he took in 1950 when he told me that our work was no good and that if he had anything to do with it we would not get a nickle [sic] for any renewal of our work.52
Hammond was also increasingly attuned to other serious health consequences. “I am strongly suspicious,” he noted, “that cigarette smoking increases the death rate from causes other than lung cancer.”53 When the study was published, Hammond and Horn had added cardiovascular disease to the short but growing list of the cigarette’s harms. “Tobacco heart” reappeared in modern form, confirmed by these substantial epidemiological investigations.54
Hammond’s investigation also addressed two questions that had been raised by Doll and Hill’s study. He showed that the risks of smoking did not vary between urban and rural areas, and that even light smokers incurred higher risk of disease and death than nonsmokers.55 With these findings, Hammond countered the frequent arguments of skeptics that environmental confounders, such as air pollution, were more important than cigarettes in causing cancer and that only “excessive” smokers were at risk.
Doll would later praise Hammond’s study for introducing much larger data pools and showing the wide range of maladies associated with smoking, far more than had been anticipated. Hammond’s data demonstrated that smoking might be responsible for up to 40 percent of mortality among smokers. Although the retrospective studies had focused on lung cancer (by starting with lung cancer patients), the prospective studies offered the important opportunity of identifying a wide range of other potential health outcomes among smokers.
Hammond subsequently pointed out that the relationship between smoking and disease could be found only in a particular historical and social context:We are concerned here with a restricted set of conditions—human populations where death from infectious and parasitic diseases is uncommon and where violence and accidents account for a relatively small proportion of all deaths. It is only in such populations that a remarkable degree of association has been found between death rates and amount of cigarette smoking.56
Hammond offered an important rationale for the discovery of the harms of smoking. Only in developed nations that had experienced a decline in infectious disease and possessed the affluence of a consumer culture would the harms of smoking become fully visible.57
While Doll and Hill moved to employ additional epidemiological strategies to confirm their initial findings, Wynder and Graham tried to uncover the biology of carcinogenesis through animal studies.58 Wynder reasoned that such studies, if successful, would be an important step in confirming the cancer-tobacco link. He and Graham believed that the crucial question was now centered on what came to be called the “biological plausibility” of the causal claim. In the same year that they reported their epidemiological findings, they initiated a study in which they painted mice with condensed tars from tobacco smoke. Wynder, accompanied by Graham’s assistant Adele Croninger, traveled to Bar Harbor, Maine, to visit the renowned Roscoe B. Jackson Memorial Laboratory, founded in 1929 by geneticist C. C. Little, to learn about techniques of experimentation on purebred mice. Even after Wynder had gone to Washington, DC, to do his internship at Georgetown University and then on to the Sloan-Kettering Institute in New York, Graham and Croninger continued painting mice with the chemical tars distilled from tobacco smoke.
After a year, 44 percent of the mice thus treated had developed cancers. Wynder found benzopyrenes, arsenic, and other carcinogens in the tars but could not determine which chemical specifically caused the animals’ cancers. “It remains to be seen whether removing a small percentage of the tar will decrease the carcinogenic activity of this material,” he wrote. “The answer to this question cannot be given until such time as we know what the active carcinogenic component of tobacco tar really is.”59
Even so, Wynder and Graham’s studies on mice offered critical support to the emerging epidemiological studies. Applying tobacco tars to laboratory animals had been attempted in the past, with similar results.60 But in the context of the epidemiological data, these findings took on new importance. No longer could skeptics—whether within the industry or within science—claim that the evidence linking smoking to disease was “merely” statistical; Wynder and Graham had given the connection the crucial quality of biological plausibility. The production of tumors in lab animals offered a powerful indicator that something in cigarette smoke could account for the epidemiological findings.
Although some researchers had difficulty replicating these experiments—there was confusion about dosing, length of observation, and the character of the evolving tumors—the wide recognition of the existence of carcinogens in tobacco nonetheless provided critical scientific support for the conclusion that smoking did cause cancer. The first published account of the Wynder, Graham, and Croninger mouse experiments appeared in the December
1953 issue of Cancer Research.61 The paper galvanized medical and public attention.
By early 1954, many physicians, scientists, and public health professionals were convinced of the hazards of smoking. Graham would proclaim that “those who have ventured to express doubt on the significance of our findings can almost be counted on the fingers of one’s two hands.”62 Following the publication of the prospective studies by Doll and Hill and Hammond and Horn a year later, Wynder wrote to Graham, “It is interesting to see how the circle is beginning to close.”63 He had a right to feel vindicated. Noting Hammond’s early skepticism, Wynder wrote that “one can relish in the thought that one’s original stand has been accepted.”64 Graham, for his part, expressed resentment that Hammond’s findings—which he saw as merely confirming his own—had generated so much attention and praise.65 But Graham perhaps underestimated that there were two simultaneous processes at work: the validation of the causal relationship of lung cancer to smoking and the legitimation of epidemiology as a tool of medical science. Clinical medicine remained antagonistic to quantitative analysis.66
By the mid-1950s, researchers employing a range of approaches had substantially advanced medical and scientific knowledge of the harms of cigarette use. Collectively, they had reached a conclusion of signal importance about the relationship of smoking to lung cancer. This conclusion emerged from three distinct but related domains of medical knowledge: clinical observations, population studies, and laboratory experiments. The questions associated with cigarettes as a cause of disease illuminated the relative strengths and weaknesses of these approaches to medical knowledge as well as important connections among them. Demonstrating that smoking caused disease ultimately required important insights integrating clinical, epidemiological, and laboratory investigations.
As we have seen, clinical observations concerning the possible harms of cigarette use proved crucial to investigations like Ochsner’s and DeBakey’s, but they alone could not resolve the hypothesis they helped to generate. Physicians, healers, and other health care providers on the front lines may observe symptoms, make diagnoses, offer therapeutics, and draw conclusions about what causes disease. Often, physicians have related particular patterns of disease to the environment, noting that some environments appear to be comparatively healthy while others seem to foment disease. They have also drawn attention to the role of particular behaviors in the development of disease—including nutrition and diet, exercise, and the use of stimulants. And they have long known that individuals vary in their innate resistance to diseases. Physicians have often written up specific cases from which they drew conclusions about health and disease, and about prevention and causality.
But the difficulty of making reliable generalizations from such observations had historically been a crucial limitation of clinical knowledge. Given the well-recognized variation among patients, how could a physician be sure that other individuals would respond in similar ways to “causes”? Without rigorous statistical methods developed for epidemiological research, they could not. As Bradford Hill would explain, it was the very nature of variability that required additional modes of investigation. How else could medicine reach broad conclusions beyond the observation that every patient is indeed different? “Far, therefore, from arguing that the statistical approach is impossible in the face of human variability,” he explained, “we should realize that it is because of variability that it is often essential.”67
The epidemiologists whose work proved so central in demonstrating the harms of tobacco use drew on a deep historical legacy of exploring the causes of disease. In the centuries before the ascendance of the laboratory and the microscope, careful observations of patients, populations, and their behavior and environments, sustained by the collection and evaluation of vital statistics, had been central to understanding causality. This tradition’s crucial contributions included the association between poor health and poverty studied by Edwin Chadwick in Great Britain and Lemuel Shattuck in the United States, and the remarkable efforts by John Snow and William Farr to assess the causes of epidemic cholera during the nineteenth century.68
The breakthroughs of the germ theory, most sharply articulated by Louis Pasteur and Robert Koch during the last decades of the nineteenth century, disrupted this approach to the investigation of disease. Their research situated the laboratory, rather than the clinic, at the top of the epistemological hierarchy of medical knowledge. Even more importantly, the identification of specific organisms as necessarily related to specific diseases radically reconfigured assumptions about the nature of disease causality. Following his discovery of the tubercle bacillus in 1882, Koch’s conclusion that a specific pathogen was invariably associated with a specific infectious disease would radically reorient medical thought and practice concerning disease causality. Although the power of his postulates and their experimental elegance transformed the biological sciences, their reductionist assumptions soon became handicaps in the effort to understand other kinds of disease.69
As infectious diseases gave way to chronic systemic diseases like heart disease and cancer as causes of death in developed nations over the first half of the twentieth century, these handicaps grew more significant. A model that assumed a simple, straight path from cause to disease lacked sufficient explanatory power. From tracking microbes and their impact on the cellular level, investigators would now come to investigate risks and their impact on population health.70 The investigations into the harms associated with tobacco would be crucial in this transformation of medical ways of knowing.
Such changes do not arise without conflict. Among the many appeals of laboratory experimentation was that it appeared to replace probability with something approaching “proof ”—and by precisely identifying the cause of a disease, it also held out the promise of certain cure. By mid-century, with penicillin and other antibiotic drugs newly on the scene, this promise looked close to being fulfilled. Scientists who were steeped in the values of the laboratory, with deep intellectual and cultural commitments to controlled experimentation, often voiced skepticism about field investigations dependent on the collection and assessment of statistics drawn from populations. As a means of data collection, the patient interview was woefully imprecise when compared to the carefully designed experiments of the laboratory. For those in search of a “definitive” demonstrative experiment, notions of probabilistic, quantitative findings were anathema. Many researchers now pointed out, however, that much in medicine and science could not necessarily be confirmed in the laboratory. “In short,” concluded statistician Jerome Cornfield of the National Institutes of Health (NIH).
if we insist on direct experimental demonstration on humans there are many widely held beliefs that must be regarded as without solid foundation. . . . The truth of the matter appears to be that medical knowledge (and one suspects, many other kinds as well) has always advanced by a combination of many different kinds of observation, some controlled, and some uncontrolled, some directly and some only tangentially relevant to the problems at hand. Although some methods of observation and analysis are clearly to be preferred to others when a choice is possible, there are no magical methods that invariably lead to the right answer. If we cannot specify exactly what has been learned in medicine from the study of statistical associations, we can at least say that we could not have accumulated the knowledge we have without them.71
Epidemiological findings like those of Doll and Hill would come under attack from scientists unilaterally committed to experimental laboratory investigation. But the lab offered no way of resolving the question of smoking’s harms. Even if scientists could have decided on the most appropriate animal model for the investigation of smoking, the production of disease in animals could not perfectly replicate pathogenesis in humans. In the end, resolving the lung cancer-smoking relationship would require a new and more sophisticated understanding of the very character of medical knowledge.
Even those advocating laboratory science must have understood
that such observations only could be preliminary. As British physician George White Pickering explained, “Any work which seeks to elucidate the cause of disease, the mechanism of disease, the cure of disease, or the prevention of disease, must begin and end with observations on man, whatever the intermediate steps may be.”72 In the case of smoking and lung cancer, Hill argued, the ultimate answers could never come from the laboratory:Yet in this particular problem what experiment can one make? We may subject mice, or other laboratory animals, to such an atmosphere of tobacco smoke that they can—like the old man in the fairy story—neither sleep nor slumber; they can neither breed nor eat. And lung cancers may or may not develop to a significant degree. What then? We may have thus strengthened the evidence, we may even have narrowed the search, but we must, I believe, invariably return to man for the final proof or proofs.73