ESTRO 37 Abstract book

S26

ESTRO 37

OC-0061 Big-RT: Big Data analysis to identify predictors of radiotherapy toxicity for personalised treatment N. Somaiah 1 , S. Poelsterl 2 , C. Rodriguez-Gonzalvez 2 , A. Wilkins 3 , S. Gulliford 4 , J. Campbell 2 , S. Yu 2 , V. Garcia- Perez 2 , C. Griffin 5 , J. Bliss 5 , J. Yarnold 4 , U. Oelfke 4 , D. Dearnaley 6 , E. Hall 5 , B. Al-Lazikani 2 1 The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust, Division of Cancer Biology and Division of Radiotherapy and Imaging, Sutton, United Kingdom 2 The Institute of Cancer Research, Division of Cancer Therapeutics, Sutton- London, United Kingdom 3 The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust, Division of Radiotherapy and Imaging, Sutton- London, United Kingdom 4 The Institute of Cancer Research, Division of Radiotherapy and Imaging, Sutton- London, United Kingdom 5 The Institute of Cancer Research, Clinical Trials & Statistics Unit- Division of Clinical Studies, Sutton- London, United Kingdom 6 The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust, Division of Radiotherapy and Imaging, Sutton, United Kingdom Purpose or Objective Nearly two thirds of cancer patients will receive radiotherapy (RT), with about 20% experiencing long- term RT-induced toxicity. Currently patients cannot be stratified by risk of side effects, which restricts treatment intensity for all patients. The pathogenesis of RT-induced toxicity is complex and multi-factorial, yet most predictive analyses to date are restricted to isolated data types and standard statistical techniques, with limited success. We developed and applied bespoke, state-of-the-art machine learning techniques to large- scale, high-dimensional multidisciplinary data from the CHHiP prostate RT-fractionation trial (CRUK/06/016) to identify multi-parametric predictors of RT toxicities. Material and Methods We performed fully integrative analyses of clinician- reported outcomes, co-morbidities, dosimetry and genetic data (via RAPPER/PRACTICAL consortium) collected as part of the trial. We explored a range of methodologies for dimensionality reduction and feature extraction, including Bayesian univariate tests, elastic- net penalisation, Multi-locus SNP analyses using genetic annotations and matrix factorisation. This was particularly important for the germline data, which contained over 9 million variant calls. Selected features were modelled using a suite of linear and non-linear machine learning methods, as well as an in-house- developed hybrid-functional model. We evaluated over 1300 models using K-fold cross-validation and Bayesian optimisation. Results We applied our methodology to 964 patients (out of 3,212 recruited) with complete data profiles, successfully integrating these diverse data sets to predict the risk of rectal bleeding. Naïve Bayes - our best performing model - was trained on a feature matrix consisting of 12 clinical, 28 germline and 6 dosimetric variables and yielded a median ROC AUC of 0.69. Following integration and preliminary analysis of dosimetric, clinical and genetic variables, a higher risk of rectal bleeding was associated with increase in volume for a number of dose levels close to the prescription dose, whilst lower risk was linked to novel genetic variants in a region of chromosome 8. Conclusion We have demonstrated the predictive power of our data- rich, integrative machine learning analysis driven by a multidisciplinary team. Further analysis using deep learning methodology and network analysis is underway. The resulting novel combinatorial markers predicting RT-

induced toxicity will need to be validated in an independent data set. The successful techniques developed in this project will allow similar approaches to be applied to other tumour types treated with RT. OC-0062 Phase II trial of SBRT and Hormone therapy for Oligometastases in Prostate Cancer (SBRT-SG 05). A. Conde Moreno 1 , F. Lopez Campos 2 , A. Hervás Morón 2 , C. Vallejo Ocaña 2 , J. Valero Albarrán 3 , A. Mendez Villamón 4 , M. Puertas Valiño 4 , A. Gómez-Iturriaga 5 , P. Samper Ots 6 , A. Sola Galarza 7 , M. Rico Oses 7 , C. Ibañez Villaoslada 8 , L.A. Pérez-Romasanta 9 , J. Pastor Peidró 10 , J. López-Torrecilla 11 , F. Ferrer González 12 , N. Ortiz Rodil 1 , F. García Piñon 13 , A. Rodríguez Pérez 14 , C. Ferrer Albiach 1 1 Consorcio Hosptialario Provincial de Castellon, Radiation Oncology, Castellon, Spain 2 Hospital Universitario Ramón y Cajal, Radiation Oncology, Madrid, Spain 3 Hospital Universitario Sanchinarro Grupo HM, Radiation Oncology, Madrid, Spain 4 Hospital Universitario Miguel Servet, Radiation Oncology, Zaragoza, Spain 5 Hospital Universitario Cruces, Radiation Oncology, Bilbao, Spain 6 Hospital Rey Juan Carlos, Radiation Oncology, Móstoles, Spain 7 Complejo Hospitalario de Navarra, Radiation Oncology, Pamplona, Spain 8 Hospital General de la Defensa Gómez Ulla, Radiation Oncology, Madrid, Spain 9 Hospital Universitario de Salamanca, Radiation Oncology, Salamanca, Spain 10 Hospital General de Valencia. ERESA., Radiation Oncology, Valencia, Spain 11 Hospital General de Valencia. ERESA, Radiation Oncology, Valencia, Spain 12 Instituto Catalán de Oncología, Radiation Oncology, Barcelona, Spain 13 Consorcio Hosptialario Provincial de Castellon, Fundación Hospital Provincial- BIOSTATISTICS, Castellon, Spain 14 Hospital Ruber Internacional, Radiation Oncology, Madrid, Spain Purpose or Objective SBRT-SG 05 (ClinicalTrials.gov Identifier: NCT02192788) is a collaborative (SBRT-SG, GICOR and SEOR) prospective multicenter phase II trial testing SBRT and hormone therapy in oligometastatic prostate cancer patients. Material and Methods Prostate cancer patients (hormone-sensitive or castration-resistant) in an oligorrecurrent stage defined as less than 5 bone or lymph node metastases (including spinal bones) by Choline PET-CT or/and WB-DWI-MRI after primary treatment for their disease, were assigned to receive SBRT (Vertebral metastases: 1x16-18Gy or 3x8-- 9Gy. Lymph node metastases: 3x10-11 Gy or 6x7,5Gy. Non-spinal bone metastases: 1x16Gy or 3x10Gy). Inclusion criteria included: time from primary treatment to biochemical recurrence of more than 1 year and PSA doubling time> 2 months. A minimum of 24 months of LhRh analogues from the time of the enrollment was required. Local control rates, biochemical control rates, progression free survival, chemotherapy free survival and SBRT impact on patient’s quality of life were assessed. Toxicity was prospectively evaluated according to CTCAE criteria. Concomitant treatment with chemotherapy, abiraterone or enzalutamide was not allowed. Results From 07/2014 to 06/2017, 67 patients from 11 Spanish centers were recruited with a total of 100 oligometastases treated, 54 in lymph nodes, 42 in non-- -spinal bones and 4 in spinal bones. Twelve patients had castration-resistance disease. With a median follow-up after SBRT of 9 months, (range 1–30 months) local and