Skepticism, statistical methods, and the cigarette: a historical analysis of a methodological debate

The discipline of modern "risk factor" epidemiology was in its formative stages in the early 1950s, when epidemiologic studies revealed a strong association between cigarette smoking and lung cancer mortality. Many medical scientists and physicians were reluctant to accept these studies as a demonstration of causation because the methods were "statistical" and involved data collected in uncontrolled conditions outside the laboratory. But a substantial number of senior biostatisticians and epidemiologists also voiced concerns, albeit more methodologically sophisticated, about the quality of the evidence at the time. Statistical methods were just beginning to work their way into medicine and public health, and many epidemiologists and statisticians were concerned about the potential misuse of these methods by untrained investigators. When studies of smoking and lung cancer gained increasing publicity and were being used to recommend public health policies, some prominent epidemiologists and statisticians highlighted this debate in their efforts to pursue methodological reform. Participants in the debate over smoking and lung cancer saw the need for explicit and rigorous standards for evaluating etiologic hypotheses, but they held conflicting views about what those standards should be. These diverging views reflect an underlying tension within the discipline of epidemiology between the search for "objective" methods of scientific inference and the practical needs of public health research that persists today.