Abstract

The purpose of this study is to determine how inter-coder differences in coding jobs impact the exposures subsequently assigned by a JEM.

1,000 jobs were selected from among the job histories reported by subjects in a case-control study of lung cancer conducted in Montreal. Two coders independently each job to four different occupation classifications (OC) and three different industry classifications (IC) which can be linked to the CANJEM job exposure matrix. For jobs coded differently by the two experts, CANJEM was used to obtain various metrics of exposure to gasoline engine emissions (GEE): exposed or unexposed (based on three cut-points of probability of exposure: 5%, 25% or 50%), exposure intensity (categorical) and frequency-weighted intensity (FWI – continuous). Interrater agreement between the exposure metrics to GEE was measured using Kappa statistics for categorical metrics and Intra-class Correlation Coefficients (ICCs) for FWI.

Depending on the classification used, at the highest level of resolution, the proportion of jobs coded differently by the two experts varied from 46.8% to 64.3% for OC and from 21.5% to 36.8% for IC. Based on jobs coded differently, Cohen’s kappa statistic for exposure status ranged from 0.32 to 0.49 when using OC and from −0.02 to 0.50 when using IC depending on the cut-point used. The corresponding numbers for intensity of exposure ranged from 0.34 to 0.48 for OC and from 0 to 0.44 for IC. When restricting analysis to jobs considered as exposed using both codes, ICC for FWI varied from −0.24 to 0.35 (OC) and from −0.13 to 0.68 (IC).

Similar to other studies, a quite high proportion of jobs were coded differently by the two experts, especially for OC. Jobs with different codes had null to moderate agreement in exposure estimates. This exercise highlights the importance of improving and standardising coding of occupations and industries.