Quantitative risk analysis in radiotherapy using Bayesian networks

Radiotherapy is a complex process, relying on different human skills and highly technical devices, which consist in exposing tumors to ionizing rays. Radiotherapy is composed of sessions which are repeated 3 to 5 times a week during several weeks. A recent study of WHO concerning risks in radiotherapy points out that only half of all incidents declared since 30 years are due to technical causes, the other half being due to human and organizational causes. To improve patient safety, our proposal considers these three dimensions (technical, human and organizational) in a global assessment of risks (over-irradiation and under-irradiation) incurred by the patients during radiotherapy and the impact of the existing safety barriers. Our previous works focused on qualitative analysis through functional, dysfunctional, and organizational analysis. These studies proposed systematic guidelines using formalisms such as SADT or FMEA/HAZOP to identify dysfunctional relations inside and between each dimension, and their impact of incurred risks. This paper aims to unify all these qualitative models in a probabilistic relational model that enables to provide not only qualitative recommendations but also a quantitative evaluation of risk. Radiotherapy can be seen as a product lifecycle, by considering treatment parameters as a product designed by a multi-disciplinary team, realized and validated during the first treatment session, and finally used for each radiotherapy session. Each of these main phases are modeled using a Bayesian network pattern (derived from cognitive engineering) that captures the different causal influences of activity inputs (technical but also organizational) with regards to the activity failure modes and outputs. These Bayesian elementary networks are aggregated according to the causal flows identified in the previous qualitative analysis. Expert knowledge and adverse event databases should be used to parameter the network and provide the expected risk evaluation for each medical center.