A Model to Systematically Employ Professional Judgment in the Bayesian Decision Analysis for a Semiconductor Industry Exposure Assessment

Abstract

Bayesian Decision Analysis (BDA) uses Bayesian statistics to integrate multiple types of exposure information and classify exposures within the exposure rating categorization scheme promoted in American Industrial Hygiene Association (AIHA) publications. Prior distributions for BDA may be developed from existing monitoring data, mathematical models, or professional judgment. Professional judgments may misclassify exposures. We suggest that a structured qualitative risk assessment (QLRA) method can provide consistency and transparency in professional judgments. In this analysis, we use a structured QLRA method to define prior distributions (priors) for BDA. We applied this approach at three semiconductor facilities in South Korea, and present an evaluation of the performance of structured QLRA for determination of priors, and an evaluation of occupational exposures using BDA. Specifically, the structured QLRA was applied to chemical agents in similar exposure groups to identify provisional risk ratings. Standard priors were developed for each risk rating before review of historical monitoring data. Newly collected monitoring data were used to update priors informed by QLRA or historical monitoring data, and determine the posterior distribution. Exposure ratings were defined by the rating category with the highest probability—i.e., the most likely. We found the most likely exposure rating in the QLRA-informed priors to be consistent with historical and newly collected monitoring data, and the posterior exposure ratings developed with QLRA-informed priors to be equal to or greater than those developed with data-informed priors in 94% of comparisons. Overall, exposures at these facilities are consistent with well-controlled work environments. That is, the 95th percentile of exposure distributions are ≤50% of the occupational exposure limit (OEL) for all chemical-SEG combinations evaluated; and are ≤10% of the limit for 94% of chemical-SEG combinations evaluated.

Keywords:

• exposure assessment
• Bayesian decision analysis
• semiconductor

ACKNOWLEDGMENTS

We would like to thank the members of our Scientific Advisory Panel for this study—Drs. Robert Herrick, Peter Lees, and John Meeker—for their thoughtful critique of the study methodology and initial results. Further, we thank Frank Bonetti, Jacob Persky, and Catherine Simmons for their data management, BDA calculation, and overall quality assurance contributions to the study. We also thank all Samsung employees who are members of the Samsung Health Research Institute who contributed by providing historical monitoring data and information on the facilities and the manufacturing processes used in developing SEGs and QLRAs for the study.

Financial support for this study was received from Samsung Electronics Co., Ltd., who did not participate in the design or implementation of the study beyond providing historical monitoring data and information on facilities and processes. The authors’ opinions are solely the responsibility of the authors, and not those of the Scientific Advisory Panel or Samsung.