How to identify the healthy worker survivor effect empirically and how to interpret results from published studies: the NIOSH ethylene oxide cohort as a case study.

Modern causal methods are underutilized in occupational epidemiology, despite the development of robust methods to adequately control time-dependent confounding affected by prior exposure, the root of the healthy worker survivor effect. We demonstrate how to detect the healthy worker survivor effect empirically and explain how to interpret analyses that have not adjusted for it. For lymphohematopoietic cancer mortality and female breast cancer mortality, we performed pathway analyses assessing whether employment is a time-varying confounder affected by prior workplace exposure to ethylene oxide. These analyses ascertained whether the relevant causal relationships depicted in a directed acyclic graph were present. For both outcomes, workers employed longer were at lower risk. Workers exposed to higher levels of ethylene oxide were also more likely to leave work. Thus, employment is a time-varying confounder affected by prior exposure. The directions of these associations imply that healthy worker survivor effect is operating. Previously published estimates of health effects of workplace exposures to ethylene oxide on both lymphohematopoietic cancer mortality and female breast cancer mortality are underestimates of the true impacts. Applying these methods to other occupational cohorts can aid interpretations of analyses that have not adjusted for the healthy worker survivor effect.