by Allan Brandt
Thus, there remained a significant gap between ongoing assessments of the risks of smoking—based principally on case studies, animal experiments, and laboratory research—and the ability to categorically demonstrate these risks. It was one thing to suggest that smoking might harm susceptible individuals, and quite another to claim that it caused serious disease. The elusiveness of this causal link confounded investigators. The problem involved some of the deepest issues in science: How do we know? Are there alternative ways of knowing? What constitutes proof?72
During the first half of the twentieth century, historical strategies of investigation and explanation within medicine had been marginalized by the elegant experimental cogency of the laboratory. But the question of smoking and disease was being raised not in a laboratory but in a complex environment shaped by powerful interests. What did it mean to identify the cause of disease? No longer would it suffice to explain that a disease appeared to be airborne or waterborne, contagious or constitutional. No longer was it adequate to associate patterns of disease by socioeconomic status, climate, or geography. According to many investigators, proving causation now required the identification of a “specific” mechanism under laboratory conditions. This approach had been highly successful in identifying the causes of infection; from the 1880s to 1910, over thirty “causal organisms” were found for specific diseases utilizing the famous postulates set forth by German physician and researcher Robert Koch in 1884. The very meaning of the term cause had been revolutionized.73
Koch’s four postulates, in which a specific organism could be (a) identified, (b) isolated, (c) grown in culture, and (d) utilized to induce disease, now served as the fundamental basis for determining causality in instances of infection. The tradition of environmental and behavioral investigation that had once characterized the search for “cause” was now deemed by many researchers to be primitive and imprecise. The center of action had shifted to the laboratory. The “field” was now beneath the lens of the microscope.74
When they were first formulated, Koch’s postulates had offered an important model for understanding causality. But their limitations became apparent almost immediately. Even Koch cautioned against their uniform and rigid application. For example, he failed to satisfy all the postulates for important diseases like cholera and typhoid, even though the causal organisms were isolated in the laboratory. The late nineteenth century witnessed several important scientific critiques of the postulates, especially since it was clearly understood that many who became infected with a microorganism nonetheless remained healthy. Such cases clearly demonstrated that the one-to-one relationship between cause and disease was not absolute. Other factors had to be involved in pathogenesis. Koch found that while he himself was infected with the tubercle bacillus, he had no signs whatsoever of the disease of tuberculosis.75
Early in the twentieth century, it also became clear that the postulates often could not be satisfied for other infections, especially those caused by viruses. As a result, there was an extensive medical literature on the limitations of the postulates and the need to modify them. Many scientists argued that it would be unscientific to let a commitment to the postulates inhibit medical and scientific advances. Finally, it was widely accepted that chronic diseases called for alternative approaches to investigating and understanding causality. Since these diseases were so obviously multicausal, involving not only specific agents but the attributes of the host and the environment as well, the postulates were a weak and limited approach to understanding them. Most physicians and researchers grasped these distinctions quite well. By mid-century the notion that the postulates constituted a “gold standard” for determining causality in both infectious and chronic diseases no longer reflected the status of the postulates or the history of their application in twentieth-century medical science.76
Nonetheless, mechanism became something of a fetish in the modern medical sciences. The municipal laboratory became the new focus of public health. Even when researchers identified environmental or behavioral risks, they now generally focused on the mechanism of disease. The whole notion of statistical inference was marginalized as research came to center on the cellular level. In this respect, exposure to a carcinogen was often equated with exposure to an infectious organism. Identifying the health risks of a particular behavior like smoking fit this model poorly. The length of time before the disease developed was protracted (and equated to an “incubation period”); in addition, the large number of intervening variables confounded the emerging notion of specific causality. Everyone “exposed” did not get the disease, and most did not; and some who were not exposed, did. Also, there was broad cultural discomfort with notions of comparative risk assessment. How dangerous was the cigarette? How did this danger rate vis-à-vis other risks? Finally, medical theory offered few persuasive models for understanding systemic and chronic diseases; the anomalies of cigarette smoking did not fit the reductionist biomedical model’s ideal of specific causality.
In this context, traditional epidemiological methods of inference based on close observation of patterns of disease—central to nineteenth-century medical thought—fell into disrepute. In the first decades of the twentieth century, epidemiology—eager to associate its work with the cutting-edge findings of the bacteriologic revolution—focused attention on the identification of infectious microbes in the community. In this respect, the cultural barriers to determining the harms of cigarette smoking actually grew. The precise mechanics of the postulates made other, more abstract historical approaches to causality—like statistical inference—seem outmoded.
The risks posed by cigarettes were especially difficult to assess. Early in the twentieth century, when so many Americans began smoking, the most significant health effects had yet to develop. Early studies often failed to find clear evidence of serious pathologies and, thus, had the ironic effect of exonerating the cigarette. By the late 1930s, the long-standing hypothesis that smoking caused disease remained only a hypothesis. Both cause and effect appeared so imprecisely defined that pinpointing a relationship between them might prove impossible. Smoking was a complex behavior, by its nature difficult to study, confounded by human variability. Some individuals could smoke for a lifetime with apparently no ill effects.
Researchers also committed the tactical error of studying smoking’s effects in young people, in whom the consequences were difficult, if not impossible, to discern. Time would show that the risk of smoking increases with exposure. During the first decades of the twentieth century, when most cigarette users were young and had only recently taken up the habit, the most serious health impacts had yet to result. But it was this generation who would ultimately provide the crucial data substantiating the relationship between smoking and disease.
Despite the many obstacles to investigation, suspicions about the health impact of cigarettes persisted, especially as long-term smokers accumulated. Summarizing medical opinion in 1937, Dr. James J. Walsh noted that cigarette smoking was making once-rare ailments fairly common.77 “I am deeply persuaded,” he wrote, “that we are now reaching the limit that nature can stand of the various harmful substances inevitably associated with excessive cigarette smoke.” Walsh’s observation centered attention on a critical insight: that it could take many years, even decades, for smoking to cause disease. Again, the relatively short latencies associated with infections had directed researchers away from studying long-term, cumulative exposures.
The causal conundrum of smoking and disease could be “solved” only by a truly significant increase in disease. This certainly occurred with lung cancer. Not until the 1940s would the full health implications of the mass consumption of cigarettes become statistically visible. Many of the health effects of smoking simply did not appear on the radar screen of epidemiologic surveillance until the 1940s. But by then, many had already concluded that cigarette use posed little or no risk.
Moreover, it took a dramatic epidemiologic transition—the decline of infectious
disease as a dominant cause of mortality—to make the harms of cigarette smoking fully explicit. In the first half of the twentieth century, the most significant causes of death shifted from infectious diseases, such as tuberculosis and pneumonia, to chronic diseases like cancer and heart disease. This shift was accompanied by an impressive increase in life expectancy. In 1900, life expectancy at birth for men in the United States was approximately forty-eight; by 1970, it approached seventy and would continue to climb for the remainder of the century. Other developed countries enjoyed similar increases: in France, expected longevity for men improved from forty-five in 1900 to sixty-nine in 1972, in England and Wales from fifty-two in 1910 to sixty-nine in 1970.78 The shift in patterns of disease and the increase in life expectancies made new risks possible—and ultimately visible.
Even so, the “visibility” of the harms of smoking required new strategies for observation and assessment. By the 1940s, discussion of the consequences of cigarette smoking had shifted from individual clinical “tolerance” and the perceived effects on growth and development to the possibility of cancer and premature death. But even this transition demanded new technical capacities. Was the apparent increase in cancer and heart disease attributable to the dramatic rise of cigarette smoking? Or were these diseases simply more prominent because individuals now survived longer rather than succumbing to infection earlier in life? Or perhaps, as some argued, the apparent rise in rates of lung cancer was an artifact of new technical abilities to clearly diagnose diseases previously invisible to both medical science and public health.79
There was, of course, a long historical tradition of investigating disease in populations. Researchers throughout the ages had engaged in the systematic process of developing hypotheses about the causes of diseases and testing them against clinical observations of the trajectory of disease and individual patients’ specific medical histories and exposure. A strong historical lineage connected James Lind’s investigation into the causes of scurvy; John Snow’s remarkable demonstration that cholera in London was waterborne; and Joseph Goldberger’s illuminating experiments proving that pellagra was a diet-deficiency disease.80 These and many other examples are evidence of a powerful tradition that predated the germ theory; nonetheless, such investigations of medical causality had important limitations.
Researchers investigating the relationship of smoking to disease would draw upon and expand this traditional approach. Proponents of the “new epidemiology” that flowered at mid-century saw their discipline as a strategy for resolving causal questions that could not be answered through clinical observation or laboratory experiment. Although some scientists saw such approaches as “soft” compared to the precision and replicability of the laboratory, it had become clear that many fundamental questions about causality simply would not succumb to either standard experimental methods or clinical observation.
New, more sophisticated approaches to the epidemiology of cancer began to emerge by the 1920s. In particular, it became increasingly clear that focusing only on those who had cancer was inadequate; any full investigation would require “controls”—matched individuals free of disease. “We feel that any study of the habits of individuals with cancer,” explained Herbert L. Lombard and Carl R. Doering of the Massachusetts Department of Public Health in a 1928 article in the New England Journal of Medicine, “is of little value without a similar study of individuals without cancer.”81 They collected a cohort of more than two hundred cancer patients and a similar number of healthy controls, matched for age, economic status, and race. By organizing their study in this way, Lombard and Doering demonstrated that cancer was not contagious and that it was not associated with poor housing or, as some had suggested, with constipation.
Although Lombard and Doering did not focus specifically on smoking and cancer, they did examine the relationship of “heavy smoking” to disease in the study population, finding a 27 percent increase in overall rates of cancer among the smokers. This increase, they concluded, was “highly significant which suggests that heavy smoking has some relation to cancer in general.”82 Still, sharply aware of the limitations of their study, the authors advocated more systematic assessments:Although we realize that the figures in this study are too small and incomplete for significant conclusions to be drawn, they are represented to show the methods used in order that others may conduct similar studies. We feel that other independent samples collected in a like manner would do much to either prove or disprove our findings.83
But sustained attempts to use statistical inference and epidemiological investigation to explicate the relationship of smoking and health would not come for another two decades.
By the 1930s, the relationship of smoking to cancer was a topic of unresolved debate. It was the life insurance industry, which like the tobacco corporations grew by leaps and bounds in the first half of the twentieth century, that took the lead in understanding the effects of smoking on health.84 Insurance companies had a vested interest in prevention, behavioral aspects of health, and social and environmental influences on patterns of disease. With the ongoing decline in the authority and resources of public health agencies, expertise on the health of populations—bringing together statistical, epidemiologic, and demographic data—was often most prominently found within the insurance industry. Health experts, such as Louis Dublin, statistician at Metropolitan Life Insurance Company, Edgar Sydenstricker of the U.S. Public Health Service, and Frederick Hoffman, a statistician at Prudential, made these subjects their primary focus.85
Hoffman, in particular, took the lead in investigating the rising prevalence of cancer and its relationship to environmental conditions, offering a comprehensive analysis of the evidence for smoking as a cause of cancer in 1931.86 Already well-known for his comprehensive San Francisco Cancer Survey, published annually from 1924 to 1934, Hoffman was well aware of the methodological dilemmas inherent in any attempt to determine the impact of smoking on health.87 Did smoking cause serious disease, especially cancer? Hoffman noted that it was one thing to determine that cancer patients might have a tendency to be smokers, quite another to obtain thorough, reliable data about their smoking practices. What did it mean, for example, to call someone a “heavy” or “moderate” smoker? Many smokers used pipes or cigars as well as cigarettes. Did this affect their risk of disease? And it was difficult to ascertain how long and with what regularity an individual had smoked. All these questions made systematic evaluation of the impact of smoking especially difficult. Like all researchers attempting to tie behavior to disease patterns, Hoffman found himself confronting the limits of smokers’ own testimony. In 1930, when he conducted his surveys, a sixty-five-year-old man with oral cancer might well have smoked since youth, but if his smoking patterns followed historical trends, he most likely began with pipes and cigars, only to shift later to cigarettes. Even if he accurately recalled exactly when he took up cigarettes, this typical pattern made it even more difficult to draw causal inferences with any confidence.
Hoffman nonetheless noted the impressive rise in cases of lung cancer in the United States during the first two decades of the twentieth century. In 1915, the rate stood at 0.7 per 1,000; it rose to 1.1 per 1,000 in 1920, 1.6 in 1924, and 1.9 in 1926. Although some observers attributed the increase to improved diagnosis, Hoffman disagreed. “I am strongly inclined to think that the increase is directly connected with the much wider spread of cigarette smoking habits,” he noted, “including the inhaling of smoke which must enter the lungs to a considerable extent in many cases.”88 Finally, he realized that “the injurious effects of tobacco smoking in their relation to cancer probably require quite a long period of time to become noticeable.”89
Yet Hoffman was not prepared to recommend abstinence. Rather, he invoked moderation, the watchword of Progressive hygiene. “Moderation in smoking,” he concluded, “commenced in adult life and carried on with reasonable safeguards is, in all probability, free from serious danger. . . .”90 If this were not so, he reasoned,
given the dramatic rise in cigarette smoking, one would likely have seen an equally dramatic “increase in cancers of the specific organs and parts most affected by smoking habits,” such as the mouth and lip.91 The obstacles to a more definitive assessment of the relationship of smoking to lung cancer—and other diseases—would remain largely intact until mid-century.
Source: American Cancer Society
CHART 2 Cancer rates by site, males
Source: American Cancer Society
CHART 3 Cancer rates by site, females
In 1938, Raymond Pearl, the eminent Johns Hopkins biologist and population geneticist, examined the relationship of tobacco use to longevity. Unlike everyone who studied the health impact of smoking before him, Pearl went for the bottom line: was tobacco use correlated with a shortened life span? Using the data collected in his Family History Records of nearly 7,000 white males, he concluded that “the smoking of tobacco was statistically associated with the impairment of life duration, and the amount of this impairment increased as the habitual amount of smoking increased.”92 His findings anticipated a dose-response relationship; the greater the sustained exposure, the shorter anticipated life expectancy. Pearl refused to speculate as to why this might be so. But he did point out that the use of pharmacologically active agents for pleasure—despite potential harms—was a long-standing historical tradition.