Controlling for the healthy worker survivor bias with g-estimation to assess the potential benefits of reducing occupational exposure limits

摘要

Background: To prevent healthy worker survivor bias, g-estimation of an accelerated failure time model is a good analytic method for occupational cohorts with longitudinal exposure data. Although technical challenges make the method difficult to implement with quantitative exposures, it is useful for evaluating the benefits of hypothetical interventions. Aims: To assess the potential impacts on all-cause mortality of lowering the occupational exposure limit (OEL) for respirable particulate matter (PM3.5) composed of metalworking fluid to a series of different values. Methods: For each OEL value, we used g-estimation to calculate the total counterfactual years of life saved by an intervention in which annual exposure was not permitted to exceed the OEL. The years of life saved by each intervention were computed relative to the observed outcomes, which were common to all cutoffs. Because the effect measure in each analysis relates counterfactual outcomes under an intervention to observed outcomes under no intervention, results from the different cutoffs may be compared. Results: Beginning with a PM3.5 cutoff of 0.15mg/m3, the impact of the intervention progressively increased as the cutoff approached 0. For example, the interventions that allowed no one to be exposed above 0.15, 0.10, and 0.05mg/m3 would have saved the population of 38,666 workers a total of 4336, 6587, and 8128 years of life, respectively. The greatest impact was observed for a cutoff of 0-i.e., for the most extreme intervention, in which nobody is ever exposed: in that case, a total of 22,652 years of life would have been saved. Conclusion: This method allows for a comparison of the impacts of hypothetical interventions limiting workplace exposures to below various levels. Interventions had greater impacts for lower OELs; there was no evidence of a threshold.