Occup Environ Med 71:A30 doi:10.1136/oemed-2014-102362.94
  • Oral presentation

0223 Marginal Structural Models in Occupational Epidemiology: An Application in the US Aluminium industry

  1. Ellen Eisen1
  1. 1Environmental Health Sciences, UC Berkeley School of Public Health, Berkeley, CA, USA
  2. 2Department of Internal Medicine, Stanford University, Stanford, CA, USA


Objectives To apply Marginal Structural Models (MSM) to address healthy workers survivor effect in a cohort study of active workers when time varying variables on health status and exposure are measured.

Method We used Cox MSMs and inverse probability weighting to assess the effect of PM­2.5 exposure on incident ischaemic heart disease (IHD) in an active cohort of 11 966 US aluminium workers. The outcome was assessed using medical claims data from 1998 to 2012. Quantitative exposure metrics of current exposure to PM2.5 were dichotomized using different cutoffs and effects were assessed separately for smelters and fabrication. Risk score based on insurance claims was available as a time varying health status variable.

Results Defining binary PM2.5 exposure by the 10th percentile cut-off, health status was affected by past exposure and predicted subsequent exposure in smelters, but not in fabrication. A Traditional cox model was appropriate for fabricators; the hazard ratio was 1.51(95% CI: 1.12 – 2.06) and was attenuated when considering higher exposure cutoffs. In smelters, Cox MSM Hazard Ratios for IHD comparing the effect of exposure in a population had everyone always been exposed to everyone always unexposed, using the 10th percentile exposure cutoff was 1.83 (95% CI: 1.14 – 2.94). Higher exposure cutoffs also resulted in attenuated effects.

Conclusions Marginal Structural Models can be used in active employment occupational cohorts to address time varying confounding. Results from the current study suggest that occupational exposure to PM2.5 in the aluminium industry increases the risk of IHD in both smelters and fabrication.