S1524
Interdisciplinary - Quality assurance and risk management
ESTRO 2026
1 Physics Department, Clatterbridge Cancer Centre, Liverpool, United Kingdom. 2 Department of Radiation Oncology (Maastro), Maastricht University Medical Centre+, Maastricht, Netherlands. 3 Medical, European Organisation for Research and Treatment of Cancer (EORTC) Headquarters, Brussels, Belgium. 4 Department of Information Systems and Operations Management, Tilburg University, Tilburg, Netherlands. 5 Physics, University of Liverpool, Liverpool, United Kingdom Purpose/Objective: The use of Artificial Intelligence (AI) in healthcare is expanding rapidly. Its novelty, complexity, and speed of development may lead to a lack of familiarity with, and understanding of, the risks posed, and ultimately to inadequate risk management1-3. Emerging general frameworks for prospective risk management of AI consider broader categories of potential risk than the process-driven Healthcare Failure Modes and Effective Analysis (HFMEA)4. Using commercial AI auto- contouring software as a pilot study, the purpose of this study was to explore whether the current HFMEA risk management approach remains applicable in the AI era, or whether a new framework for risk management may be required. Material/Methods: Two different radiotherapy centers in separate countries, with substantial experience in HFMEA risk assessments, independently completed assessments for their locally implemented AI-based auto- contouring software solution. Each center used their own processes, blinded to the other center’s risk assessment. Comparisons were made of the layout and structure of the risk assessment, the sub-process steps (and associated risks) identified, risk prioritisation and scoring. Risks not directly attributable to the AI-contouring software were censored. Common themes and risks within the risk assessments were identified. Results: Significant variability was seen between risk assessments in both the quantity and type of identified risks, and their associated risk scoring. ‘Center 1’ identified 11 risks falling into three broad categories, ‘Center 2’ identified 18 risks within the same categories. Only five risks were common to both risk assessments (Figure 1).
predominantly equipment failure, contributed to the generation of 14% of events. Organisational causes, indicative of a systems approach in understanding RTE were cited for 9% of events.
Conclusion: This study is the first to characterise the generation and detection of RTE within SABR/SRS radiotherapy pathways within a national ELS. It has identified that, whilst RTE are generated along the whole pathway, they arise most frequently during pretreatment activities and verification imaging. Areas of most effective detection are pretreatment planning and three treatment related activities. This analysis suggests that most providers often attribute contributory factors of RTE to individual factors, whilst factors linked to systems thinking were rarely identified. References: 1. UKHSA Safer radiotherapy: national patient safety radiotherapy event taxonomy (2025). Available at https://www.gov.uk/government/publications/safer- radiotherapy-national-patient-safety-radiotherapy- event-taxonomy (accessed 6.11.25)2. McGurk R, Woch Naheedy K, Kosak T, et al. Multi-Institutional Stereotactic Body Radiation Therapy Incident Learning: Evaluation of Safety Barriers Using a Human Factors Analysis and Classification System. J Patient Saf. 2023;19(1):e18-e24 Keywords: SABR, safety, risk
Digital Poster 2733
Is HFMEA risk management applicable in the artificial intelligence (AI) era? An inter-center study of risk for AI auto-contouring software. Jonathan Lucas 1 , Martyn Gilmore 1 , Petra Rejinders- Thijssen 2 , Daniel Portik 2,3 , Wouter van Elmpt 2,4 , Carl Rowbottom 1,5
Made with FlippingBook - Share PDF online