Objective To estimate inter-rater reliability of expert assessment of occupational exposures.

Methods A population-based case-control study conducted in Montreal was used to obtain detailed information on lifetime occupational histories. Two trained industrial hygienists assessed the 4,362 reported jobs to assign exposures using a checklist of 258 agents. The jobs were divided between the two experts for evaluation (initial coding); then, each reviewed the others’ to reach a consensus. A job was considered ‘exposed’ to an agent if that agent was present at levels above the non-occupational environment. Experts rated exposure for each job/substance combination according to confidence that the exposure occurred (possible, probable, definite), and to concentration (low, medium, high), where, low and high represented the extremes in the range of levels encountered in a work environment. An inter-rater reliability sub-study was conducted among a random sample of 185 jobs. Each expert coded the 185 jobs (1st coding); then, 6 months later, a 2nd coding occurred, whereby each expert coded the other’s evaluation but did not have access to their own 1st evaluation. The statistical unit of observation was each job/substance decision (185 jobs×258 substances=47,730 decisions/expert). Chance-corrected weighted kappa statistic and Gwet’s AC1 estimated the concordance between the experts in the 1st and 2nd coding.

Results Over 98% agreement was found and >97% (n=36,497) of decisions were to attribute no exposure to a particular job/substance combination by both experts. Restricting to combinations rated as exposed by both experts (n=508), Kappa=0.44 (95%CI: 0.37–0.50) and Gwet=0.55 (0.48–0.61) was found for confidence; while, Kappa=0.30 (0.15–0.45) and Gwet=0.92 (0.90–0.95) was found for concentration. After the 2nd coding, agreement improved for both confidence (Kappa=0.68, 0.63–0.73; Gwet=0.70, 0.65–0.75) and concentration (Kappa=0.65, 0.50–0.80; Gwet=0.96, 0.95–0.98).

Conclusion This sub-study provides some evidence supporting the reliability of expert assessment of occupational exposures in large-scale epidemiologic studies.