Article Text
Abstract
The determination of appropriate standards for workplace exposure requires accurate estimation of the health effects. This will sometimes require adjustment for the Healthy Worker Survivor Effect (HWSE) using g-methods, which can adjust for the time-varying confounding on the causal pathway that characterises this bias. These methods allow the estimation of a target parameter corresponding to an intervention on the data generating distribution.
These target parameters generally capture the disease experience of workers under a fixed regimen of exposure at a constant level. However, any exposure regimen must acknowledge that terminated workers will no longer experience any occupational exposure. We consider two contrasting target parameters – the etiologic and the realistic – that both recognise this fact. An etiologic parameter corresponds to a hypothetical intervention that fixes exposure levels and keeps all workers employed throughout follow-up. A realistic parameters corresponds to an intervention that fixes exposure levels as long as workers are employed, but allows them to leave work.
An etiologic parameter provides a better estimate of the biological effect of the exposure of interest. However, it requires information and assumptions about the process of leaving work. In addition, it corresponds to an infeasible intervention which may be of limited interest to policy makers.
Realistic parameters, on the other hand, allow some of the processes that characterise the HWSE to affect the true parameter value. The true value of the realistic parameter could even be null when in fact exposure causes disease; interventions setting exposure at higher limits may induce earlier termination and less cumulative exposure accrual than lower limit interventions.
We use directed acyclic graphs as our theoretical framework to define both types of target parameters in two contrasting types of occupational cohort designs. We discuss the rationale for each type and the ramifications of the choice.