ESTRO 2025 - Abstract Book

S4427

Late-breaking abstracts

ESTRO 2025

References [1] Jintao Ren, Jesper G Eriksen, Jasper A Nijkamp and Stine Korreman, ”Comparing different CT, PET and MRI multi-modality image combinations for deep learning-based head and neck tumor segmentation”, Acta Oncologica 60(11):1399-1406, doi: 10.1080/0284186X.2021.1949034, 2021. [2] Mathis Ersted Rasmussen, Casper Dueholm Vestergaard, Jesper Folsted Kallehauge, Jintao Ren, Maiken Haislund Guldberg, Ole Nørrevang, Ulrik Vindelev Elvstrøm, and Stine Sofia Korreman, ”RadDeploy: A framework for integrating in-house developed models seamlessly into radiotherapy workflows”, Physics in Imaging and Radiation Oncology 31:100607, doi: 10.1016/j.phro.2024.100607, 2024.

5022

Proffered Paper First RAPID-RT analysis: Using rapid-learning to assess the survival impact of a new cardiac avoidance area during lung cancer radiotherapy Gareth Price 1,2 , Catharine Morgan 3 , Evan Kontopantelis 3 , Tjeerd van Staa 3 , Tom Marchant 2 , Kathryn Banfill 4 , Abigail Walker 4 , Rebecca Holley 1 , Harry Crawford 4 , Alan McWilliam 1 , Marcel van Herk 1 , Neil Bayman 4 , David Woolf 4 , Claire Barker 4 , Jennifer King 4 , Clara Chan 4 , Laura Pemberton 4 , Joanna Coote 4 , Hamid Sheikh 4 , Danya Abdulwahid 4 , Catherine Harris 2 , Joe Wood 2 , Corinne Faivre-Finn 1,4 , On behalf of the RAPID-RT team 1 Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom. 2 Christie Medical Physics and Engineering, The Christie NHS Foundation Trust, Manchester, United Kingdom. 3 Division of Informatics, Imaging and Data Sciences, University of Manchester, Manchester, United Kingdom. 4 Department of Clinical Oncology, The Christie NHS Foundation Trust, Manchester, United Kingdom Purpose/Objective Evaluating changes in routine radiotherapy care with conventional trials can be challenging due to equipoise concerns, selection biases, learning effects, and resource limitations. Emerging evidence has suggested that limiting radiation dose to a new Cardiac Avoidance Area (CAA), composed of the right atrium, proximal coronary arteries, and aortic valve 1 , may improve survival in lung cancer patients receiving curative-intent radiotherapy 1–3 . In addition to the above challenges, there is also uncertainty about the optimal CAA definition and dose limit 4 . A rapid-learning approach, iteratively assessing and optimising new techniques using real-world data within routine clinical care 5,6 , offers a potential solution. The RAPID-RT study (ISRCTN17129364) uses rapid-learning to evaluate the impact of limiting dose to the CAA on survival 7 . We report the first learning cycle results. Material/Methods In April 2023 our institution became the first to implement CAA-based planning for all patients receiving non SABR, curative radiotherapy for stage I-III lung cancer 8,9 . D max to 1cc of the CAA was limited to 19.5Gy for 20 fractions (or equivalent) unless conflicting with target coverage. Patients treated with the CAA dose limit (cohort 2) were prospectively enrolled via an informed opt-out consent process and compared to those treated prior to its introduction (cohort 1). All data was collected from the electronic patient record, including demographic, tumour, treatment, and survival data. Missing data was addressed via Multiple Imputation 10 . Survival was modelled using a Bayesian parametric Weibull distribution adjusting for prognostic factors and using uninformative priors. Results Between 1/2021 and 2/2025, 922 and 786 patients were included in cohorts 1 and 2 (average missingness 6.7%). Cohorts were well balanced, except for radiotherapy technique, with VMAT use increasing sharply (93.2% of cohort 2 vs 31.6% of cohort 1, Figure-1). The proportion of patients not achieving the CAA constraint reduced from 56.8% to 29.7% with little impact on other OAR DVHs 9 . Model diagnostics confirmed good convergence and performance. Analyses suggest a survival benefit in cohort 2 (Figure-2) with a population average 12-month survival probability of 0.738 (0.673, 0.802) in cohort 2 compared to 0.682 (0.642, 0.744) in cohort 1.