ESTRO 2021 Abstract Book

S921

ESTRO 2021

erythema, desquamation, skin tenderness, dryness, edema, pigmentation, breast pain and fatigue. Late toxicities were significantly higher in the control arm. Cosmetic effect was more favorable to the SBRT group. Conclusion This single-institution prospective randomized study compared the feasibility, toxicity, the cosmetic and economic influence of two radiotherapeutic regiments. SBRT technique turned out to be a less toxic and easier feasible approach for adjuvant radiotherapy of early-stage breast cancer patients. Consequently, the presented study increases the level of evidence for RT-indicated patients to the establishment of SBRT into daily clinical practice of APBI. Supported by grant NV19-03-00354 of the Czech Ministry of Health. PO-1107 Inflammatory markers in breast cancer patients treated with radiotherapy: Machine Learning approach E. Rodríguez-Tomàs 1 , G. Baiges-Gaya 2 , J. Acosta 3 , L. Torres 3 , H. Castañé 1 , J. Gómez 3 , M. Árquez 3 , F. Castaño 3 , J. Camps 4 , J. Joven 4 , M. Arenas 3 1 Rovira i Virgili University, Medicine and Surgery, Reus, Spain; 2 Institut d’Investigacions Sanitàries Pere Virgili (IISPV), Medicine and Surgery, Reus, Spain; 3 Sant Joan de Reus University Hospital, Radiation Oncology, Reus, Spain; 4 Sant Joan de Reus University Hospital, Medicine and Surgery, Reus, Spain Purpose or Objective Oxidative stress and inflammation are key points in the progression of breast cancer (BC). Radiotherapy (RT) has become an essential oncological treatment in this population. Computational science tools like machine learning can help us to discover predictive models for easier medical decisions making. The present study aims to identify predictive variables based on circulatory antioxidant and inflammatory markers, using Machine Learning, in patients with BC treated with RT and correlated these alterations with clinical variables. Materials and Methods The study included 243 BC patients treated with adjuvant RT and 100 control subjects. The follow-up of BC patients was an average of 6 years post-RT. Biochemical analysis and ELISA of interleukin-4 (IL4), interferon- gamma (IFN-γ), C-C motif ligand 2 (CCL2), paraoxonase 1 (PON1) concentration and activity were performed in plasma and serum samples obtained prior to treatment and 1 month after RT administration. The predictive model was designed and developed in Scikit-learn Python library. We collected clinical characteristics of patients related to age, menopausal stage, TNM system, molecular subtype, affected sentinel node, oncological treatments received (surgery, neoadjuvant chemotherapy, hormonal treatment, targeted therapy), among others. All statistical analyses and graph representations were carried out by SPSS, R software and GraphPad Prism. Results PON1 concentration and activity were lower in BC patients in compare to control (p<0.001) but after RT there was a significant increase of its concentrations (p<0.001). While INFγ and IL-4 concentrations were lower and higher in BC pre-RT in compare to control, respectively, CCL2 concentration was similar between groups. After RT only IL-4 concentration decrease significatively (p<0.001). Random forest was the best machine learning predictive model to use in these populations with less error (A). IL-4 was the best predictor of BC and the effects of RT (B and C). When compare BC pre-RT patients with control subjects, negative associations were detected between lymphocyte concentration and IL-4 (B). By contrast, when compare RT effects, this association was positive (C). Pre-RT, those patients with higher histological grades (II and III) and T staging (T3, T4) present lower PON1 concentrations but higher PON1 activity. Also, CCL2 and IL-4 concentrations were higher in higher T staging prior to treatment. After RT, we observed a significant increase in PON1 concentration and a decrease of PON1 activity in large T staging. Molecular subtypes with worse prognoses (HER2 and triple-negative) were associated with higher PON1 activity and CCL2 and INFγ concentrations. Conclusion IL-4 concentration is the best predictor to discriminate between health and BC patients and the effect of RT, which reverses IL-4 alterations. A validation study is necessary to establish IL-4 concentration as a predictive marker in BC patients treated with RT.