Ensuring comparability of benzene exposure estimates across three nested case-control studies in the petroleum industry in support of a pooled epidemiological analysis

Background: Three case-control studies each nested within a cohort of petroleum workers assessed exposure to benzene in relation to risk of haematopoietic cancers. These studies have each been updated and the cases will be pooled to derive a more powerful study. The benzene exposure of new leukemia cases and controls was estimated in accordance with each respective study's original methods. An essential component of the process of pooling the data was comparison and rationalisation of the exposure estimates to ensure accuracy and consistency of approach. This paper describes this process and presents comparative estimates before and after appropriate revision took place. The original petroleum industry studies, in Canada, the UK and Australia, were conducted at different points in time by different study teams, but the industry used similar technology in similar eras in each of these countries.

Methods: A job history for each subject giving job title, dates of starting and leaving the job and location of work, was assembled. For each job or task, the average benzene exposure (Base Estimate (BE) in ppm) was derived from measurements collected at applicable worksites. Estimates of exposure intensity (workplace exposure estimates (WE)) were then calculated for each line of work history by adjusting the BEs for site- and era-specific exposure-related variables such as loading technology and percentage benzene in the product. To ensure that the exposure estimates were comparable among the studies, the WEs were allocated to generic Job Categories, e.g. Tanker Driver (by technology used e.g. bottom loading), Motor Mechanic. The WEs were stratified into eras, reflecting technological changes in the industry. The arithmetic mean (AM), geometric mean (GM) and range of the stratified WEs were calculated, by study, for each generic Job Category. These were then compared. The AMs of the WEs were regarded as substantially similar if they were within 20% in all three studies in one era or for at least two studies in two eras. If the AM of the WE group differed by more than 20%, the data were examined to see whether the difference was justified by differences in local exposure conditions, such as an enclosure versus open work area. Estimates were adjusted in the absence of justification for the difference.

Results: Reconciliation of differences resulted in changes to a small number of underlying BEs, particularly the background values, also the BEs attributed to some individuals and changes to the allocation of jobs between Job Categories. Although the studies covered some differing sectors of the industry and different time periods, for 22 Job Categories there was sufficient overlap, particularly in the downstream distribution sector, to make comparisons possible. After adjustment 12 Job Categories were judged to be similar and 10 were judged to be justifiably different. Job-based peak and skin exposure estimates were applied in a uniform way across the studies and a single approach to scoring the certainty of the exposure estimates was identified.

Conclusions: The revised exposure estimates will be used in the pooled analysis to examine the risk of haematopoietic cancers and benzene exposure. This exercise provided an important quality control check on the exposure estimates and identified similarly exposed Job Categories that could be grouped for risk assessment analyses.