Weak Associations in Occupational Epidemiology and the Criteria for Deducing Causality

Abstract

Study of the incidence of cancer in men and women employed in different occupations has been the most prolific method for establishing the carcinogenicity of chemicals to Man. The specific risks of cancer that have been observed have commonly been large, either absolutely in the case of common cancers, like cancer of the lung attributed to bis-chloromethyl ether, or relatively, like angiosarcoma of the liver attributable to vinyl chloride, and it has been easy enough to determine that the chemicals to which the employees were specifically exposed were the causal agents, once certain elementary rules had been defined for the scientific conduct of epidemiological research. With the care that is now taken to evaluate the toxicity of chemicals before they are used on a large scale and the higher standards of housekeeping in industry, which diminish the risk of prolonged and intensive exposure to agents whose effects are not fully understood, there is little chance of situations occurring in the future like those that caused such large risks in the past and the problems with which we are now faced are much more difficult to solve. These are mostly posed by the discovery that very large doses of an agent to which people have been exposed at work in much smaller amounts for many years are capable of causing cancer in one or more species of animal. Studies then have to be undertaken to see if any effect can be detected from the occupational exposure and, as often as not, some types of cancer are found to have occurred more often than would be expected: judged, that is, by comparison with some selected control population. We are then faced with the problem of deciding whether the observed excesses should be regarded as due to chance, bias in the conduct of the study, confounding with some other factor, or the carcinogenicity of the agent of interest.