Towards causal inference in occupational cancer epidemiology--I. An example of the interpretive value of using local rates as the reference statistic.

A brief overview is made of the criteria currently applied for establishing causation in occupational cancer epidemiology, and further criteria or 'desiderata' are proposed. These supplement the present somewhat simplistic ones for 'sufficient evidence of carcinogenicity' advocated by the International Agency for Research against Cancer in their monograph series. A frequent cause of confounding, not always appreciated, is the limited choice of a reference population for the statistical analyses. The study population is often an industrial one with marked regional and urban biases, especially for some of the cancers under scrutiny: these biases can influence markedly the resultant Standardized Mortality Ratio (SMR) calculated, and thus lead to misleading conclusions. The British Rubber Manufacturers' Mortality Study, 1946-1980, provided an opportunity to extract the data set of one of the 13 factories involved. Both local and national death rates were applied as a reference, and the use of the former substantially reduced the high SMRs previously recorded for lung and stomach cancers. This emphasized the need for further analyses when making other internal comparisons in order to accommodate more of the 'desiderata' listed. Ultimately causation can be inferred only after weighing up all the evidence, and must be based on a 'balance of probabilities'.