Objective To investigate whether a refined assessment of exposure to bitumen fume among workers in the European asphalt industry within a nested case-control study resulted in a different interpretation pertaining to risk of lung cancer mortality compared with the cohort study.

Methods Pearson correlation coefficients between refined and original estimates were calculated. Logistic regression and generalised additive models (penalised splines) were fitted to estimate ORs for exposure to bitumen fume using the refined and original exposure estimates, respectively, while adjusting for potential confounding.

Results 1555 subjects included in the nested case-control study had both refined and original estimates for exposure to bitumen fume. Exposure assessment in the nested case-control study (compared with the cohort phase) increased the number of subjects never-exposed to bitumen fume from 18% to 32%. From the 1282 subjects originally considered exposed in the cohort phase, 309 (24%) became unexposed after the nested case-control exposure assessment. From the 273 subjects originally considered non-exposed in the cohort phase, 87 (32%) became exposed in the nested case-control study. The majority (75%) of subjects however did not change exposure status and changes were similar among cases and controls. Correlation coefficients between refined and original exposure estimates were moderate overall (range 0.42–0.46), but varied considerably among countries. The ORs and exposure–response curves for exposure to bitumen fume were not meaningfully different between analyses that used refined and original exposure estimates. Adjustment for tobacco smoking and exposure to coal tar did not change these patterns.

Conclusions Our results showed that more detailed data collection and exposure assessment in the nested case-control study compared with the cohort study did change exposure status of many subjects, but did not alter results of the exposure–response analysis. Adjustment for tobacco smoking did not have a noticeable effect on risk estimates either.