S1323
Interdisciplinary - Education in radiation oncology
ESTRO 206
Workshops/seminars and Certificate/Degrees (p = 0.21). Self-reported knowledge and comfort strongly correlated with objective performance (p < 0.001, all combinations), though a subset who self-rated “None/Basic” achieved high scores, indicating under- recognised proficiency that targeted teaching could unlock. Career stage showed no association with scores. Perceived importance of AI was 36.3% important/critical, 48.4% somewhat important, and 15.3% not important. Six questions generated the lowest scores (<50% correct) and highlight priority topics for training and education.
Discipline of Radiation Therapy, Trinity College Dublin, Dublin, Ireland. 7 Department of Radiation Oncology, University of Alabama at Birmingham, Birmingham, AL, USA. 8 Inpictura Ltd, Inpictura, Abingdon, United Kingdom. 9 Division of Cancer Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester, United Kingdom. 10 Discipline of Medical Imaging and Radiation Therapy, School of Medicine, University College Cork, Cork, Ireland. 11 Department of Medical Physics, University Hospitals Birmingham NHS Foundation Trust, Birmingham, United Kingdom. 12 Centre for Physics in Health and Medicine, School of Physics, University College Dublin, Dublin, Ireland Purpose/Objective: To quantify current AI knowledge across radiation oncology (RO) professionals, benchmark performance by role, education and experience, and identify priority-learning needs. Secondarily, to report reliability and item performance of the assessment instrument. Material/Methods: A multidisciplinary, international group with representation from ESTRO AI working groups as well as clinical practice, academia, and industry generated a 55-item question bank mapped to core AI-in- radiotherapy concepts. Content was refined via three rounds (Round 1 blinded scoring; Rounds 2-3 transparent consensus). The final assessment included 21 knowledge based questions plus demographics and perception of AI items. It was deployed as an online app globally in 7 different languages and shared to ESTRO National Societies. We report overall percentage performance and performance per role, associations with career stage, training background, self-reported knowledge/comfort and individual question and category scores. Group comparisons used non-parametric tests and multivariable modelling; pairwise comparisons used Holm correction for multiple testing. Internal consistency (Cronbach’s α ) and item discrimination were also calculated. Results: Completion was 502 respondents (70% completion rate across knowledge items). The instrument showed excellent internal consistency ( α = 0.90) and good/very good item discrimination (Mean: 0.63±0.13, range=0.32-0.82). Across the cohort, the median total score was 63.6% [IQR 45.5-81.8]. By profession, Medical Physicists were highest (72.7% [59.1-86.4], n=293), followed by Radiation Oncologists (50.0% [40.9-68.2], n=81) and Radiation Therapists (38.6% [13.6-54.5], n=78), with all differences significant (p<0.001). Level of training showed a significant difference (No training<Self-study<Workshops, p<0.001). However, no difference was found between
Conclusion: AI knowledge varies systematically by profession and by training background, with workshops performing on par with formal certificates/degrees (others lower), while career stage shows no effect. Self-ratings track measured performance and, combined with psychometrically robust items, offer a scalable proxy for workforce monitoring. Self-ratings also highlighted under-recognised proficiency among some low- confidence groups. This consensus-built 21-item instrument shows excellent reliability and discrimination and may benchmark curricula, target and prioritise training needs, and evaluate interventions across the Radiation Oncology community. Keywords: AI, Artificial Intelligence, Assessment
Poster Discussion
Made with FlippingBook - Share PDF online