ESTRO 2020 Abstract book

S882 ESTRO 2020

of the coefficients obtained. As measurements of the metric used, we provide the area under the curve (AUC) of the ROC curve and optimism. Results The variables most selected with the method selected (method 1) were (descending order): irradiation of the lymph nodes, the range of breast density, V95PTV, D98PTV, CTV boost volume, age and Her2. Figure 1 shows the ROC curve performed and the resulting coefficients of the RLM. The model obtained provides an AUC of 0.75 with an optimism of 4.1%. The most significant variable is the treatment of the areas. Patients with lymph node involvement were 8 times more likely to suffer LF than those without lymph node involvement.

PO-1539 Predictive modelling of late fibrosis in breast cancer radiotherapy M. Lizondo 1 , J. Fuentes-Raspall 2 , N. Jornet 3 , A. Latorre- Musoll 3 , P. Delgado-Tapia 3 , P. Carrasco 3 , J. Pérez-Alija 3 , P. Gallego 3 , P. Simón 3 , A. Ruiz-Martínez 3 , M. Adrià 3 , I. Valverde-Pascual 3 , M. Barceló 3 , N. Garcia 3 , M. Ribas 3 1 Institut de Recerca Hospital de la Santa Creu i Sant Pau, Servei de Radiofísica i Radioprotecció, Barcelona, Spain ; 2 Hospital de la Santa Creu i Sant Pau, Servei d'Oncologia Radioteràpica, Barcelona, Spain ; 3 Hospital de la Santa Creu i Sant Pau, Servei de Radiofísica i Radioprotecció, Barcelona, Spain Purpose or Objective Fibrosis is one of the late complications associated with radiotherapy treatment in breast cancer. In this work, different variables are analysed to develop a predictive model of radioinduced fibrosis with a view to further personalizing the treatment. Material and Methods This study includes 195 breast cancer patients treated with conservative radiotherapy since May 2013 and with a minimum follow-up time of 32 months. Prescription dose was 50 Gy to breast and lymph nodes, and 66 Gy to the boost (2Gy/fraction). All patients were treated with a Clinac2100(Varian). This study includes the following treatment techniques: 3DCRT (6MV and/or 15MV) + sequential boost with electrons or integrated boost with IMRT (6MV). Table 1 shows the 18 variables that were considered to develop the predictive model. The effect studied, late fibrosis (LF), was evaluated following the Common Terminology Criteria for Adverse Events (CTCAE v.3.0). 67 patients presented maximum G1 toxicity, and 4 G2 patients. FT was evaluated as a binary variable: fibrosis diagnosed or not during the follow-up period.

Conclusion The analysis performed indicates that the predictive factors of late fibrosis are irradiation of the lymph nodes, greater V95PTV and a wider range of breast density. The model obtained allows us to calculate probabilities of occurrence with an AUC of 0.75. Acknowledges: This work was financed by the Spanish Association Against Cancer (AECC). PO-1540 Radiomic models for validation in patients with locally advanced HNSCC treated with primary RTCx A. Rabasco 1,2,3 , A. Zwanenburg 1,3,4 , S. Leger 1,3,4 , K. Pilz 1,3,5 , F. Lohaus 1,2,5 , A. Linge 1,3,4,5 , K. Zöphel 6,7 , J. Kotzerke 6,7 , A. Schreiber 8 , I. Tinhofer 9,10 , V. Budauch 9,10 , M. Stuschke 11,12 , P. Balermpas 13,14 , C. Rödel 13,14 , U. Ganswindt 15,16 , C. Belka 15,16 , S. Pigorsch 15,17 , S. E.Combs 15,17 , D. Mönnich 18,19 , D. Zips 18,19 , C. Richter 1,2,3,5 , E. G.C. Troost 1,2,3,4,5 , M. Krause 1,2,3,4,5 , M. Baumann 5,20 , S. Löck 1,3,5 1 OncoRay - National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus- Technische Universität Dresden- Helmholtz-Zentrum Dresden – Rossendorf, Dresden, Germany ; 2 Helmholtz-Zentrum Dresden-Rossendorf, Institute of Radiooncology - OncoRay, Dresden, Germany ; 3 German Cancer Research Center DKFZ, Heidelberg and German Cancer Consortium DKTK partner site Dresden, Dresden, Germany ; 4 National Center for Tumor Diseases NCT, Partner Site Dresden, Dresden, Germany ; 5 Department of Radiotherapy and Radiation Oncology- Faculty of Medicine and University Hospital Carl Gustav Carus, Faculty of Medicine and University Hospital Carl Gustav Carus- Technische Universität Dresden, Dresden,

The prediction model was developed as follows:

- Selection of variables. Two different methodologies were tested choosing the one with lower Akaike information criterion. • Method 1: On each one of 1000 samples selected by bootstrap, a logistic regression was made between each variable and the LF, choosing those variables whose adjustment had a p- value<0.1. Then we performed a multiple logistic regression (RLM) with stepwise selection with backward elimination. We selected the most frequent variables, complying with the rule of no more than 1 variable for every 10 patients in whom toxicity was presented. Method 2: Random Forest. - RLM. We used the set of variables obtained at the end of the previous process to perform the RLM that will be our predictive model. For an internal validation, we performed a RLM on each one of another 1000 samples selected through bootstrap. The final coefficients will be the means •