Exposure assessment methods for a study of mortality and cancer morbidity in relation to specific petroleum industry exposures

In 1987 a Canadian company implemented an exposure tracking and health information system. The exposure tracking method aligned closely with published concepts for describing workplace exposure, with over 1800 similar exposure groups being used to describe occupational exposures. The database has been actively maintained and is subject to a number of quality checks. Recently, the company initiated a cancer morbidity study, with one objective being to examine whether the exposure tracking data could be used to reconstruct exposure estimates for the cohort. Five agents--hydrogen sulfide, petroleum coke/spent catalyst, hydrocarbon solvents and fuels, hydrocarbon lubricants, and an index for exposure to operations derived from noise exposure--were selected for development of occupational exposure estimates for each cohort member. The cohort consisted of workers first employed between January 1964 and December 1994 and who were employed for at least 1 year. Work history records were associated with a similar exposure group, using human resources data and knowledge of local industrial hygienists. Only employees with90% duration of their work history assigned were kept in the cohort (25,292 people out of a possible 25,617). For each similar exposure group inventory, the substances were identified that contributed to each of the five agents being studied. Exposure estimates before 1987 were modified using historic occupational exposure limits. Rules were created to sum the exposure from multiple substances found in any one similar exposure group. The validity of exposure estimates was tested via comparison with results documented in industrial hygiene survey reports. Industrial hygienists who were unaware of the derived exposure estimates evaluated several hundred industrial hygiene surveys and prepared benchmark information. The two lists were then evaluated for concordance, which was found to be significantly different from that occurring by chance. We conclude that the process described can create valid exposure estimates for use in epidemiology studies.