Photoresist materials are indispensable in photolithography, a process used in semiconductor fabrication. The work process and potential hazards in semiconductor production have raised concerns as to adverse health effects. We therefore performed a health risk assessment of occupational exposure to positive photoresists in a single optoelectronic semiconductor factory in Taiwan. Positive photoresists are widely used in the optoelectronic semiconductor industry for photolithography. Occupational exposure was estimated using the Stoffenmanager R model. Bayesian modeling incorporated available personal air sampling data. We examined the composition and by-products of the photoresists according to descriptions published in the literature and patents; the main compositions assessed were propylene glycol methyl ether acetate (PGMEA), novolac resin, photoactive compound, phenol, cresol, benzene, toluene, and xylene. Reference concentrations for each compound were reassessed and updated if necessary. Calculated hazard quotients were greater than 1 for benzene, phenol, xylene, and PGMEA, indicating that they have the potential for exposures that exceed reference levels. The information from our health risk assessment suggests that benzene and phenol have a higher level of risk than is currently acknowledged. Undertaking our form of risk assessment in the workplace design phase could identify compounds of major concern, allow for the early implementation of control measures and monitoring strategies, and thereby reduce the level of exposure to health risks that workers face throughout their career.

No abstract

Addressing occupational health and safety concerns early in the design stage anticipates hazards and enables health professionals to recommend control measures that can best protect workers’ health. This method is a well-established tool in public health. Importantly, its success depends on a comprehensive exposure assessment that incorporates previous exposure data and outcomes. Traditional methods for characterizing similar occupational exposure scenarios rely on expert judgment or qualitative descriptions of relevant exposure data, which often include undisclosed underlying assumptions about specific exposure conditions. Thus, improved methods for predicting exposure modeling estimates based on available data are needed. This study proposes that cluster analysis can be used to quantify the relevance of existing exposure scenarios that are similar to a new scenario. We demonstrate how this method improves exposure predictions. Exposure data and contextual information of the scenarios were collected from past exposure assessment reports. Prior distributions for the exposure distribution parameters were specified using Stoffenmanager® 8 predictions. Gower distance and k-Medoids clustering algorithm analyses grouped existing scenarios into clusters based on similarity. The information was used in a Bayesian model to specify the degree of correlation between similar scenarios and the scenarios to be assessed. Using the distance metric to characterize the degree of similarity, the performance of the Bayesian model was improved in terms of the average bias of model estimates and measured data, reducing from 0.77 (SD: 2.0) to 0.49 (SD: 1.8). Nevertheless, underestimation of exposures still occurred for some rare scenarios, which tended to be those with highly variable exposure data. In conclusion, the cluster analysis approach may enable transparent selection of similar exposure scenarios for factoring into design-phase assessments and thereby improve exposure modeling estimates.