2384

Background/Aim Exposure to organic solvents is one of the most common health hazard in workplaces. We aimed to find an association of multiple exposure to low-level organic solvents with hepatotoxicity by Bayesian kernel machine regression(BKMR). Methods Korean special health screening data of 2013 was used. We used hippuric acid(g/g crea.), methyl hippuric acid(g/g crea.), and N-dimethylformamide(mg/L) concentrations for exposure variable and aspartate aminotransferase(AST, IU/L), alanine transaminase(ALT, IU/L), and gamma-glutamyltransferase(GGT, IU/L) for outcome variable. For statistical analysis, participants with the exposure concentration above the exposure limits of the Korea special health screening were excluded, and those with missing data were excluded. All exposure variables were log transformed and standardized prior to the analysis. We examined the association between single exposure and each outcome using multivariate linear regression. In addition, we estimated combined hepatotoxicity of three exposure variables using BKMR. All analyses were adjusted for age, sex, body mass index, smoking status, alcohol drinking habits, and physical activity. Results 615 workers’ data were included in the analysis. In the single exposure model, N-dimethylformamide concentrations were positively associated with AST (β=0.92, SE=0.35). In the BKMR analysis, increase of AST level was significantly associated with increase of combined three exposures. Posterior inclusion probabilities of the hippuric acid, methyl hippuric acid, and N-dimethylformamide were 0.148, 0.382, and 0.950, respectively. In addition, the results of BKMR represented the interactions among the three exposures and nonlinearity of each exposure. Conclusions The study found that simultaneous exposure to low-level organic solvents resulted in hepatotoxicity, and among the three exposures, N-dimethylformamide had the highest hepatotoxicity. BKMR can show combined effect, contribution of each exposure on the combined effect, nonlinear relationship, and interactions between exposures. Therefore, we should try to use this approache on occupational health study to obtain additional information not available from classical statistics.