Methods: According to a previously published taxonomy and incident learning system, we defined important aspects of the preparation system according to responsibilities and staff involvement. We analyzed existing incident databases from the TBCC and other facilities, and used pre-existing researcher knowledge to define RT processes, incident classes, and contributing factors in fault tree structures. We then conducted structured elicitations of staff (oncologists, dosimetrists, therapists) to add to and refine the fault trees for specific processes. We elicited infrastructure (e.g. equipment, hardware, software) fault trees from medical physicists, who are involved throughout the RT process. Once the trees were finalized, we conducted training sessions in probability elicitation in order to populate the branches. Spreadsheets were provided to staff to elicit minimum, most likely, and maximum estimates of incident probabilities, which in turn are used to define "PERT" (modified Beta) distributions. These estimates include failure of existing quality control measures. Additional literature review and facility specific incident tracking are currently being conducted in order to refine probability estimates and to define aggregated PERT distributions for use in modeling risks of radiation misadministration.
Results: Results to date include a complete set of fault trees for incidents in RT patient preparation, and a preliminary set of probabilities for use in iterative risk analysis.
Conclusions: This study coordinates with an advanced incident learning system that is currently being implemented at the TBCC, and has proven valuable to staff in terms of increased knowledge of the system and identification of likely areas of concern with regard to patient safety. The probabilistic fault tree approach is consistent with advanced methods used in engineering and environmental science, and represents one of the few examples, to our knowledge, of application of this powerful methodology in health care.