ESTRO 2025 - Abstract Book

S3771

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Conclusion: This study highlights the preliminary potential of radiomic models to assess recurrence risk in HNC using single institution data. However, further research is required to explore post-surgical imaging patterns and validation is being undertaken to confirm the identified radiomic features in external datasets.

Keywords: radiomics, head-and-neck, locoregional recurrence

References: [1] Burnet, N. G., et al. (2017). Applying physical science techniques and CERN technology to an unsolved problem in radiation treatment for cancer: the multidisciplinary VoxTox research programme. CERN IdeaSquare Journal of Experimental Innovation, Vol. 1 No. 1 (2017): Special Issue: Experimenting in Innovation; https://doi.org/10.23726/CIJ.2017.457 [2] van Griethuysen, et al. (2017). Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Research, 77(21), e104–e107. https://doi.org/10.1158/0008-5472.CAN-17-0339

2118

Poster Discussion Application of causal inference techniques for accurate estimation of the causal effect of dose to the heart base on overall-survival in lung cancer Miren Summers 1 , Matthew Sperrin 2 , Eliana Vasquez Osorio 1 , Kathryn Banfill 3 , Alan McWilliam 1 1 Division of cancer science, The University of Manchester, Manchester, United Kingdom. 2 School of health sciences, The University of Manchester, Manchester, United Kingdom. 3 Clinical oncology, The Christie NHS Foundation trust, Manchester, United Kingdom Purpose/Objective: Voxel-based analysis aims to analyse spatial dose distributions of a given patient cohort to identify anatomical regions associated with treatment outcomes. Multivariable models used in this approach are not geared to infer causality. This work reanalyses McWilliam et al. 2017 [1] using directed acyclic graphs (DAGs) to identify the correct causal adjustment set and make an accurate causal estimate of excess dose to the heart base on overall-survival (OS). Material/Methods: A DAG was developed to inform modelling of the causal effect of dose to the region of significance identified in the base of the heart where excess dose impacts OS. Measured and unmeasured covariates were included and their effects modelled. Covariates controlled for in the original McWilliam et al. analysis were adjusted on the DAG, showing biasing pathways remained open. Therefore, a minimal adjustment set was devised which addressed all bias. Data of the 1100 patients from the original study treated in a single academic centre between 2010 and 2013, with 55Gy in 20 fractions using 3D conformal radiotherapy (3DCRT) or IMRT, were reanalysed. Data was missing for a large proportion of the units, prompting use of multiple imputation. Cox regression analysis was carried out using covariates in the original analysis and the minimal adjustment set devised. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated. Results: Using the DAG, an adjustment set which addressed bias on the causal effect of interest was derived: gross tumour volume, tumour location, staging, radiotherapy modality, prescribed dose. Some overlap existed with the original multivariable analysis (GTV, staging, prescribed dose), other variables had no effect on the causal estimate and adjustment for others caused inaccuracy (lung dose). The results of multivariable analysis on the effect of dose to