Occupational exposures and lung cancer: adjustment for unmeasured confounding by smoking.

BACKGROUND: Occupational cohort mortality studies rarely include information on smoking history. Consequently, smoking is often an unmeasured potential confounder in analyses of associations between occupational exposures and lung cancer. Several authors have recommended sensitivity analyses to assess confounding by smoking, which require speculation about the prevalence of smokers in different occupational exposure groups.
METHODS: A method of adjustment for confounding by smoking is proposed in which the target parameter of interest (the log hazard ratio for lung cancer contrasting exposed to unexposed workers, adjusted for smoking) is approximated by the difference between the crude exposure-disease associations for lung cancer and chronic obstructive pulmonary disease. A polytomous logistic regression method is used to derive appropriate confidence intervals. The performance of this adjustment approach is assessed via direct calculations.
RESULTS: Under the scenarios considered, 90% or more of the bias due to confounding by smoking was removed via this adjustment in the absence of smoking data.
CONCLUSIONS: This approach to adjustment for confounding by smoking can be employed without explicitly positing the distribution of smoking with respect to occupational exposure. The approach is easily implemented in analyses of occupational cohort data and should facilitate quantitative assessments of bias due to unmeasured confounding by smoking in occupational studies of lung cancer.