Controlling for potential confounding by occupational exposures.

Occupational exposure is an important potential confounder in air pollution studies because it is plausible that individuals who live in highly polluted areas also work in more polluted environments. While the original investigators made some efforts to control for possible confounding by occupational variables, it was felt that these could be improved upon. The reanalysis team attempted to control for occupational confounding by supplementing the original data sets with two new variables, an indicator of the "dirtiness" of a subject's job and an indicator of possible exposure to occupational lung carcinogens. The attribution of these variables was based on the job title recorded by the original investigators and on the judgment of our experts concerning typical exposure patterns in different occupations. We fitted Cox proportional-hazards models identical to those that had been used by the original investigators while also including one or both of the new occupational covariates in the models. In none of the analyses did the inclusion of the occupational variables materially change the results. It would therefore appear that, in general, the results reported by the original investigators were not distorted by inadequate control of occupational variables. We also carried out some analyses using the dirtiness index as a stratification variable to assess effect modification. There was some indication, albeit inconsistent, that the effect of air pollution on mortality was greater among subjects with dirty jobs than among those with clean jobs.