Article Text

Download PDFPDF
Original article
A comparison between standard methods and structural nested modelling when bias from a healthy worker survivor effect is suspected: an iron-ore mining cohort study
  1. Ove Björ1,
  2. Lena Damber1,
  3. Håkan Jonsson1,
  4. Tohr Nilsson2
  1. 1Department of Radiation Science, Oncology, Umeå University, Umeå, Sweden
  2. 2Department of Occupational and Environmental Medicine, Sundsvall Hospital, Sundsvall, Sweden
  1. Correspondence to Ove Björ, Department of Radiation Science, Oncology, Umeå University, SE-901 85 Umeå, Sweden; ove.bjor{at}


Objectives Iron-ore miners are exposed to extremely dusty and physically arduous work environments. The demanding activities of mining select healthier workers with longer work histories (ie, the Healthy Worker Survivor Effect (HWSE)), and could have a reversing effect on the exposure–response association. The objective of this study was to evaluate an iron-ore mining cohort to determine whether the effect of respirable dust was confounded by the presence of an HWSE.

Methods When an HWSE exists, standard modelling methods, such as Cox regression analysis, produce biased results. We compared results from g-estimation of accelerated failure-time modelling adjusted for HWSE with corresponding unadjusted Cox regression modelling results.

Results For all-cause mortality when adjusting for the HWSE, cumulative exposure from respirable dust was associated with a 6% decrease of life expectancy if exposed ≥15 years, compared with never being exposed. Respirable dust continued to be associated with mortality after censoring outcomes known to be associated with dust when adjusting for the HWSE. In contrast, results based on Cox regression analysis did not support that an association was present.

Conclusions The adjustment for the HWSE made a difference when estimating the risk of mortality from respirable dust. The results of this study, therefore, support the recommendation that standard methods of analysis should be complemented with structural modelling analysis techniques, such as g-estimation of accelerated failure-time modelling, to adjust for the HWSE.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.