S1525
Interdisciplinary - Quality assurance and risk management
ESTRO 2026
Proffered Paper 2753
Improving the quality of collegial decision-making in radiotherapy: identification and reduction of cognitive biases in multidisciplinary meetings Charlotte LE ROY, Ronan TANGUY Radiotherapy, Centre Léon Bérard, Lyon, France Purpose/Objective: Collegial decision-making in multidisciplinary team (MDT) meetings is essential to the quality, consistency, and safety of radiotherapy care. Evidence from high- reliability sectors such as aviation and industry shows that cognitive biases can impair decision accuracy and promote errors (1). The healthcare environment shares similar vulnerabilities to human error (2). More than 30 types of cognitive biases have been described (3), some arising from individual predispositions that may lead to factually incorrect choices, and others influencing decisions that are suboptimal without being objectively wrong (e.g., risk aversion, ambiguity tolerance). In addition, certain biases emerge from group dynamics and can compromise the objectivity and quality of therapeutic decisions during MDT discussions (4, 5).Few studies have examined cognitive biases in radiotherapy (6), particularly within group decision-making. This study aimed to identify the main biases affecting MDT discussions in our radiotherapy department and to propose corrective measures to improve the quality of collegial decisions. Material/Methods: A six-month prospective qualitative study was conducted during weekly MDT meetings. Using an analytical grid based on classical cognitive biases (Kahneman and Tversky), two independent observers analyzed each meeting, categorizing biases by type, frequency, and context. Findings were shared in a team feedback session to co-design quality- improvement actions. Results: Across 24 MDT meetings and 141 therapeutic decisions, cognitive biases appeared in 65% of discussions. The most frequent observed bias were framing effect (80%), anchoring bias (72%), posterior probability error (65%), conformity bias (59%), and authority bias (37%), along with recurrent patterns such as sunk cost and status quo biases.Several corrective strategies were proposed by the team, including:* Meeting Scheduling: Hold meetings during dedicated time slots, preferably after lunch, avoiding end-of-day sessions and task interruptions.* Case Presentation Standardization: Require standardized case presentation templates, submitted to the department one day in advance.* Devil’s Advocate Step: Include a structured “devil’s advocate” phase to challenge group consensus.* Cognitive Bias Training: Conduct an initial collective training on
Center 1 considered more risks involving staff training and documentation, whereas Center 2 identified more risks in the use and fidelity of the software. Comparing risks common to both centers, Table 1 shows the RPN scoring, and highlights greatest disagreement in the evaluation of occurrence and detectability, rather than the severity of a given risk. Occurrence and detectability scores differed by a factor of 2 to 3.
Conclusion: The study demonstrates variations between centers in
risk management applied to AI. Two separate centers, with long-term experience of risk
assessment, identified different areas of risk focus and differed in their scoring of risk, particularly in expected failure rates and detectability. Additionally, comparing to broad categories within the AI risk atlas taxonomy2 for example, both risk assessments focused mostly on the output category. Development and harmonization of prospective risk management tailored to AI applications, with the inclusion of more centers and greater consideration of broader categories including training data, model behavior and non-technical risks are necessary. References: 1. National Institute of Standards (2020). Artificial Intelligence Risk Management Framework (AI RMF 1.0). https://doi.org/10.6028/NIST.AI.100-12. Bagehorn et al (2025). AI Risk Atlas: Taxonomy and Tooling for Navigating AI Risks and Resources. https://github.com/IBM/risk-atlas-nexus3. Wells et al (2025). A practical framework for appropriate implementation and review of artificial intelligence (FAIR-AI) in healthcare. npj Digit. Med. 8, 514. https://doi.org/10.1038/s41746-025-01900-y4. Giardina et al (2016). A Review of Healthcare Failure Mode and Effects Analysis (HFMEA) in Radiotherapy. Health Phys. 111(4):317-26. https://doi.org/10.1097/HP.0000000000000536 Keywords: AI, risk management
Made with FlippingBook - Share PDF online