Evaluating exposure–response associations for non-hodgkin lymphoma with varying methods of assigning cumulative benzene exposure in the Shanghai Women's Health study

Objectives: To provide insight into the contributions of exposure measurements to job exposure matrices (JEMs), we examined the robustness of an association between occupational benzene exposure and non-Hodgkin lymphoma (NHL) to varying exposure assessment methods. Methods: NHL risk was examined in a prospective population-based cohort of 73087 women in Shanghai. A mixed-effects model that combined a benzene JEM with >60000 short-term, area benzene inspection measurements was used to derive two sets of measurement-based benzene estimates: ‘job/industry-specific' estimates (our presumed best approach) were derived from the model's fixed effects (year, JEM intensity rating) and random effects (occupation, industry); ‘calibrated JEM'estimates were derived using only the fixed effects. ‘Uncalibrated JEM' (using the ordinal JEM ratings) and exposure duration estimates were also calculated. Cumulative exposure for each subject was calculated for each approach based on varying exposure definitions defined using the JEM's probability ratings. We examined the agreement between the cumulative metrics and evaluated changes in the benzene–NHL associations. Results: For our primary exposure definition, the job/industry-specific estimates were moderately to highly correlated with all other approaches (Pearson correlation 0.61–0.89; Spearman correlation > 0.99). All these metrics resulted in statistically significant exposure–response associations for NHL, with negligible gain in model fit from using measurement-based estimates. Using more sensitive or specific exposure definitions resulted in elevated but non-significant associations. Conclusions: The robust associations observed here with varying benzene assessment methods provide support for a benzene–NHL association. While incorporating exposure measurements did not improve model fit, the measurements allowed us to derive quantitative exposure–response curves.