Purpose: Health is important for continued employment and therefore continued accrual of occupational exposure; furthermore, steady employment can benefit health. Consequently, bias can occur in estimates of cumulative exposure-mortality associations. This has been called the healthy worker survivor effect (HWSE). The processes associated with the HWSE tend to lead to variation in mortality rates with time-since-termination of employment, most notably a peak in mortality shortly after termination of employment. We use simulations and an empirical example to demonstrate that time-since-termination can be a confounding factor in analyses of occupational-exposure-mortality associations.
Methods: Simulation data were generated for 20,000 workers followed for 40 years under a model of no effect of employment duration (a proxy for cumulative exposure) on mortality. Proportional hazards regression methods were used to quantify exposure-mortality associations and evaluate methods to control for the HWSE. Results were derived after 100 iterations of the simulation. Relationships between employment duration and mortality were also investigated in a cohort of 122,247 male utility workers with adjustments for time since termination.
Results: Simulation data show a peak in mortality rates in the first year after termination of employment which declined in magnitude with continued time since termination of employment; average employment duration also declined with time since termination of employment. This led to confounding of cumulative-exposure-mortality associations, with spurious evidence of a positive association between cumulative exposure and mortality in the post-termination period. Adjustment for time-since-termination eliminated this spurious association; in contrast, adjustment for a binary indicator of employment status led to positively-biased relative rate estimates. A similar pattern was observed in analyses of utility worker data. The log relative rate of all cancer mortality is -0.12+/-0.03 per decade of employment without adjustment for time-since-termination, and -0.01+/-0.03 with adjustment for time-since-termination of employment.
Conclusions: The HWSE can lead to temporal variation in mortality rates that is correlated with cumulative exposure. Under these conditions, adjusting for time-since-termination of employment may reduce bias in estimates of cumulative-exposure-mortality trends more effectively than the commonly-used method of adjusting for a binary indicator of employment status.