Retrospective benzene exposure assessment for a multi-center case-cohort study of benzene-exposed workers in China

Abstract

Quality of exposure assessment has been shown to be related to the ability to detect risk of lymphohematopoietic disorders in epidemiological investigations of benzene, especially at low levels of exposure. We set out to build a statistical model for reconstructing exposure levels for 2898 subjects from 501 factories that were part of a nested case-cohort study within the NCI-CAPM cohort of more than 110,000 workers. We used a hierarchical model to allow for clustering of measurements by factory, workshop, job, and date. To calibrate the model we used historical routine monitoring data. Measurements below the limit of detection were accommodated by constructing a censored data likelihood. Potential non-linear and industry-specific time-trends and predictor effects were incorporated using regression splines and random effects. A partial validation of predicted exposures in 2004/2005 was performed through comparison with full-shift measurements from an exposure survey in facilities that were still open. Median cumulative exposure to benzene at age 50 for subjects that ever held an exposed job (n=1175) was 509 mg/m(3) years. Direct comparison of model estimates with measured full-shift personal exposure in the 2004/2005 survey showed moderate correlation and a potential downward bias at low (<1 mg/m(3)) exposure estimates. The modeling framework enabled us to deal with the data complexities generally found in studies using historical exposure data in a comprehensive way and we therefore expect to be able to investigate effects at relatively low exposure levels.