A comparison of standard methods with g-estimation of accelerated failure-time models to address the healthy-worker survivor effect: application in a cohort of autoworkers exposed to metalworking fluids

Epidemiology. 2012 Mar;23(2):212-9. doi: 10.1097/EDE.0b013e318245fc06.


Background: Studies of autoworkers exposed to straight metalworking fluids report excess risks of several cancers. These studies, however, have not addressed the healthy-worker survivor effect. Most methods proposed to address this bias do not consider that it may be caused by time-varying confounders affected by prior exposure. G-estimation of accelerated failure-time models was developed to handle this issue but has never been applied to account for the healthy-worker survivor effect.

Methods: We compare results from Cox models and g-estimation in 38,747 autoworkers exposed to straight metalworking fluids. Exposure was defined based on job records and air samples. We examine relationships between duration of exposure and mortality from all causes, cancers, ischemic heart disease, and chronic obstructive pulmonary disease (COPD).

Results: In standard models, hazard ratios were elevated for cancers of the larynx, prostate, and rectum, but below or approximately equal to 1.0 for all other outcomes considered. Adjustment for the healthy-worker survivor effect using time off work, employment status, time since hire, and restriction to inactive workers after 15 years of follow-up did not substantially change the hazard ratios. However, g-estimation yielded higher hazard ratios than standard Cox models for most outcomes. Exposure was related to increased risks of mortality from all causes combined, heart disease, COPD, and all cancers, as well as lung and prostate cancers.

Conclusions: G-estimation may provide a better control for the healthy-worker survivor effect than standard methods.

Publication types

  • Comparative Study
  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Adult
  • Aged
  • Automobiles*
  • Cohort Studies
  • Data Interpretation, Statistical
  • Female
  • Healthy Worker Effect
  • Humans
  • Industrial Oils / adverse effects*
  • Male
  • Metallurgy / statistics & numerical data*
  • Middle Aged
  • Models, Statistical*
  • Occupational Diseases / epidemiology*
  • Occupational Exposure / adverse effects
  • Occupational Exposure / statistics & numerical data
  • Proportional Hazards Models
  • Time Factors