|| Checking for direct PDF access through Ovid
We pooled data from three nested case control studies from cohorts of Canadian, British and Australian petroleum industry workers and were able to examine relatively rare outcomes such as myelodysplastic syndrome (MDS) and myeloproliferative disease (MPD). The exposure assessment methodology was similar for all three studies. We examined the base estimates attributed to specific jobs before pooling the data and made some adjustments, notably to the background exposures attributed to individuals. In some cases the differences in exposure were explained by different circumstances, e.g. legal requirements. These inter-study comparisons were made blind as to case-status and before the data were pooled and the analyses were carried out. Multivariate analyses were used to identity main predictors of exposure intensity in ppm benzene. Site type (terminal, refinery, upstream) job (e.g. office terminal operator craft worker) and decade were strong predictors of exposure. Study was also a predictor of exposure which was of concern but the three studies covered different industry sectors (and hence jobs) and different decades. We therefore sought site type/jobs and decades which were represented in all three studies. There was little overlap with sufficient numbers to carry out such a comparison, with the exception of terminal workers in four job categories. We showed that job and era were significant predictors of exposure intensity but study was not a major predictor although the interaction terms study X job and study X decade were significant predictors of exposure intensity. Data pooling is more powerful than meta-analyses, allowing analyses by new metrics or new groupings perhaps derived from findings from one study which can be tested in a larger setting. Careful comparisons of data sets before they are pooled are essential to provide reassurance in the quality of the pooled data set and aid interpretation of pooled analyses.