Quantification of a human reliability analysis method for radiotherapy applications based on expert judgment aggregation. (February 2020)

• Record Type: Journal Article
• Authors: Pandya, Dhruv; Podofillini, Luca; Emert, Frank; Lomax, Antony J.; Dang, Vinh N.; Sansavini, Giovanni
• Is Part Of: Reliability engineering & system safety. Volume 194(2020)
• Publication Date: 2020-02
• Subjects: Human reliability analysis -- Radiotherapy -- Safety analysis -- Risk analysis -- Expert judgment -- Human error; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2019.05.001
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: This paper develops the decision trees of a new HRA method, for radiotherapy. Expert judgment is elicited on a qualitative scale, aggregated via a Bayesian model. Partial validation is attempted versus literature HRA data. Differences data vs judgments are linked to differences in the performance context. Abstract: This paper develops the quantification framework of a new Human Reliability Analysis (HRA) method, for application to the radiotherapy domain. The method's reference data is obtained via expert judgment, due to the lack of domain-specific data. To avoid shortcomings of directly eliciting probabilities, experts are asked to assess the importance of specific factors for the failure probability, elicited on a qualitative scale. Each assessment is converted into statements about the order of magnitude of the probability value. The values are combined via an expert aggregation method, developed specifically for HRA. The paper includes an attempt to validate the elicitation against applicable Human Error Probability (HEP) values from existing HRA methods (thus, "convergent validation"). Indeed, some tasks are generic in nature and data can be assumed to be sector-independent (e.g. checking activities, interacting with interfaces, simple tasks such as identifying objects or characters/numbers). Differences in the values are identified and, when possible, linked to differences in the performance context characteristic of the field of application of the different methods.