ESTRO 2025 - Abstract Book

S3859

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

[5] X. Yang, et al. FastCPH: Efficient survival analysis for neural networks, Workshop on Learning from Time Series for Health, NeurIPS (2022).

4148

Proffered Paper Identifying Cardiac Substructures Influencing Survival in Stage III NSCLC Patients Undergoing Radiotherapy Alessandra Catalano 1 , Tiziana Rancati 1 , Eliana Gioscio 1 , Francesco Pisani 1 , Andrea Riccardo Filippi 2 , Anna Cavallo 3 , Silvia Meroni 3 , Claudia Sangalli 2 , Albina Allajbej 2 , Valentina Bartolomeo 2 , Francesco Dionisi 4 , Valeria Landoni 5 , Lucia Goanta 6 , Marco D'Andrea 4 , Livia Marrazzo 7 , Emanuela Olmetti 8 , Andrea Botti 9 , Patrizia Ciammella 10 , Sara Broggi 11 , Roberta Tummineri 12 , Miriam Torrisi 12 , Italo Dell'Oca 13 , Alessandro Cicchetti 1 1 Data Science Unit, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 2 Radiation Oncology Dep., Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 3 Medical Physics, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 4 Radiation Oncology Dep., IRCCS Regina Elena National Cancer Institute, Rome, Italy. 5 Medical Physics, IRCCS Regina Elena National Cancer Institute, Rome, Italy. 6 Radiation Oncology Dep., IRCCS Regina Elena National Cancer Institute, Milan, Italy. 7 Medical Physics, Azienda Ospedaliera Universitaria Careggi, Florence, Italy. 8 Radiation Oncology Dep., Azienda Ospedaliera Universitaria Careggi, Florence, Italy. 9 Medical Physics, IRCCS Azienda Unità Sanitaria Locale, Reggio Emilia, Italy. 10 Radiation Oncology Dep., IRCCS Azienda Unità Sanitaria Locale, Reggio Emilia, Italy. 11 Medical Physics, IRCCS Ospedale San Raffaele, Milan, Italy. 12 Radiation Oncology Dep., IRCCS Ospedale San Raffaele, Milan, Italy. 13 Radiation Oncology Dep., ASST Sette Laghi, Varese, Italy Purpose/Objective: To identify cardiac substructures associated with increased mortality in a multicenter study including stage III Non Small Cell Lung Cancer (NSCLC) patients treated with RT. Material/Methods: The retrospective study involved five different Italian centres. We enrolled 309 patients treated with RT alone or combined with chemotherapy and immunotherapy. Clinical, patient and DICOM data were collected. Ten cardiovascular structures were contoured through an automatic software (MVision AI™): Aorta, Left Anterior Descendent Artery (LAD), Pulmonary Artery, Cardiac Chambers, Inferior (IVC) and Superior Vena Cava, Heart, and Healthy Lungs. A propensity score analysis was performed matching age, KPS, sex, stage at the start of RT, PTV (T+N) volume (<300cc, 300cc-500cc, >500cc), chemotherapy and immunotherapy. Control and treatment groups were defined through the median threshold of the dose distribution in each cardiovascular substructure. In total, we considered 20 propensity-matched scenarios: 10 structure x 2 dose metrics (EUD with n=0.1, Mean dose computed with α/β = 5Gy). We used Kaplan-Meier (KM) curves to analyse differences in survival at 24 months after RT for the 20 above described settings. Results: Fifty-nine % of the enrolled patients were alive two years after RT. KM curves were significantly different for the following risk structures (Fig.1): the EUD for LAD (p=0.016 from the log rank test), the Mean Dose for IVC (p=0.026), and the EUD for healthy lungs (p=0.063). Of note, despite including PTV grouping in the propensity matching, it was impossible to completely disentangle the association between PTV and doses to some cardiac substructures. Specifically, we still saw significantly higher PTVs in patients with higher LAD EUDs (t-test p-value=0.035). Consequently, to explicitly assess the influence of PTV, we performed an additional multivariable Cox regression for the matched scenarios, including the doses to the cardiovascular structures and PTV values as a continuous feature. The IVC emerged as a critical structure for this cohort of patients. Cox modelling found that the PTV and the dose to this structure were significant predictors (p=0.01 and p=0.03, respectively) for survival. We depicted combined KM curves as a simplified representation of these factors (Fig.2). The association between survival and doses of LAD and healthy lungs faded with the explicit inclusion of PTV.