Skip to main content
eScholarship
Open Access Publications from the University of California

UC Berkeley

UC Berkeley Electronic Theses and Dissertations bannerUC Berkeley

Estimating Causal Effects of Occupational Exposures

  • Author(s): Izano, Monika A
  • Advisor(s): Eisen, Ellen A
  • et al.
Abstract

Estimates of the risk of occupational exposures are typically based on observational workplace studies that are subject to bias due to the healthy worker survivor effect (HWSE), a ubiquitous process that results in the healthiest workers accruing the most exposure. This body of work is concerned with the estimation of causal effects of occupational exposures from observational workplace studies, in the context of the HWSE.

We estimate the effect of cumulative exposure to straight, soluble, and synthetic metalworking fluids (MWFs) on the incidence of colon cancer in the United Autoworkers-General Motors (UAW-GM) cohort. We use longitudinal targeted minimum loss-based estimation (TMLE) to compute the 25-year risk difference if always exposed above compared to if always exposed below an exposure cutoff while at work. Exposure cutoffs were selected a priori at the median of exposed person-years among colon cancer cases. Risk differences are 0.038 (95% CI = 0.022 to 0.054), 0.002 (95% CI = -0.016 to 0.019), and 0.008 (95% CI = 0.002 to 0.014) for straight, soluble, and synthetic MWFs, respectively. By control of the time-varying confounding on the casual pathway that characterizes the HWSE, TMLE estimated effects that were undetectable in earlier reports.

Most workers in UAW-GM were hired decades before the reporting of incident cancers began. Incident cancers that occurred before the start of reporting were left filtered. We show that if ignored, left filtering can lead to downward bias in exposure effect estimates. Further, we propose a novel delayed-entry adjusted Kaplan-Meier estimator that controls for time-varying confounding, and permits delayed risk-set entry. The estimator results in little bias in simulated datasets when the outcome is sufficiently rare.

In addition to dynamic (realistic) interventions that assign exposure according to workers' employment status, causal contrasts can be defined under static (etiologic) interventions that additionally prevent leaving work. Causal effect estimates of the two classes of interventions can differ substantially. While ideally the choice of intervention would be driven by the research question, in practice it may be dictated by the available data. Furthermore, when estimates of the long-term etiologic effects of occupational exposures are not available, guidelines for exposure limits may be based on studies that estimated effects of realistic interventions. In a simulation study we investigate the conditions under which the two effect measures are comparable, and identify factors that drive the differences between the two.

Main Content
Current View