Bayesian aggregation of expert judgment data for quantification of human failure probabilities for radiotherapy

Abstract

The paper deals with the quantification of probabilities for human failures in the radiotherapy domain. The probabilities are used as input for the development of a Human Reliability Analysis (HRA) method specific for radiotherapy. Quantification is based on expert judgment, in view of the lack of relevant data. A Bayesian aggregation model is used to aggregate the judgments collected during elicitation sessions with domain experts. A qualitative scale is first used; then the judgments are interpreted as information on the order of magnitude of the error likelihood and aggregated under the Bayesian scheme. Besides for the specific domain of interest, this work is relevant for novel HRA applications outside typical domains, for which the need to incorporate expert judgment in traceable and defendable ways is key. magnitude of the error likelihood and aggregated under the Bayesian scheme. The paper presents the results of the aggregation. The application shows the ability of the aggregation approach to formally represent the variability of the experts’ estimates. Besides for the radiotherapy domain, the work presented in this paper is relevant for the various efforts recently done to extend HRA methods for application beyond their most typical applications, i.e. nuclear power plant operation. Lack of relevant data is a major issue for these novel applications (Bye et al. 2017, NASA 2012, Gibson 2012, Mkrtchyan et al. 2015, NUREG 2016) and methods to elicit expert judgment in a formal and defendable way are needed along with specific data collection initiatives. The paper is organized as follows. The next Section provides the background on the HRA method under development, for which probability values are sought for in this paper. Section 3 presents the design of the elicitation sessions and the concepts underlying the Bayesian approach for processing and aggregation of the judgments. Section 4 presents the application to two Decision Trees part of the framework of the HRA method under development. Concluding remarks close the paper. 2 BACKGROUND INFORMATION The framework for the HRA method consists of eighteen decision trees, one for each failure mode