NIOSHTIC-2 Publications Search

Causal models for addressing the healthy worker effect in an occupational cohort study.

Chevrier J; Eisen E

Epidemiology 2009 Nov; 20(6)(Suppl):S94

https://doi.org/10.1097/01.ede.0000362993.74803.f5

20045214

Background and Objective: Individuals who are hired and remain at work are generally healthier than those who are unemployed or leave work. Due to this healthy worker effect, potential health consequences of occupational exposures may be underestimated. Methods: We present results obtained using three different methods to control for the healthy worker effect in a longitudinal mortality study which includes over 40,000 workers with potential exposure to metalworking fluids (MWF). Individuals who were hired between 1938 and 1981 in one of three Michigan automobile manufacturing plants were enrolled in the study. Their vital status and cause of death was ascertained using the National Death Index and state health records from 1941 to 1994. Date of birth, race and work history, including time off work, were obtained from company records. Annual exposure to mineral oil-based MWF was estimated based on work history and health status was approximated using the amount of time off work for every year of follow up. To adjust for time off work as a time-varying confounder, we applied standard Cox models and compared results with those obtained using two causal modeling approaches: Marginal Structural Models with Inverse Probability of Treatment Weights (IPTW) and Structural Nested Models using G-estimation. We considered exposure to straight MWF in relation to three outcomes: all-causes of death combined, all cancer mortality, and heart disease mortality. Results: We will demonstrate that despite the lack of direct comparability between methods, since they estimate different parameters, standard methods based on cumulative exposure are biased. The bias arises because leaving work is associated with mortality, determines future exposure and is predicted by past exposure and employment history. Conclusion: The g-estimation method is unique in that it takes into account the fact that health status is both a confounder and an intermediate variable between exposure and disease.

Workers; Work-environment; Worker-health; Exposure-levels; Metal-compounds; Metallic-compounds; Automotive-industry; Statistical-analysis; Morbidity-rates; Mortality-rates; Minerals; Oils; Diseases; Demographic-characteristics; Cutting-oils

EPIDEY

20091101

Abstract

Grant

2010

Grant-Number-R01-OH-008927

6

1044-3983

Epidemiology

CA

University of California, Berkeley