Abstract Book

S198

ESTRO 37

associated with GVHD or radiation related. This series has confirmed FP IMRT- based TBI to be both safe and effective conditioning for full intensity HSCT. OC-0387 Treatment information improve machine learning survival prediction in soft tissue sarcoma patients J.C. Peeken 1 , T. Goldberg 2 , C. Knie 1 , B. Komboz 2 , M. Bernhofer 3 , F. Pasa 4 , K.A. Kessel 1 , P. Tafti 2 , B. Rost 3 , F. Nüsslin 1 , A. Braun 2 , S.E. Combs 1 1 Klinikum rechts der Isar- TU München, Department of Radiationoncology, München, Germany 2 Allianz SE, Global Data and Analytics, Munich, Germany 3 Informatik 12- TU München, Department for Bioinformatics and Computational Biology, Munich, Germany 4 Department of Physics- TU München, Chair of Biomedical Physics-, Munich, Germany Purpose or Objective Current prognostic models for soft tissue sarcomas (STS) are solely based on staging information. Treatment- related data or response-associated information are currently not regarded. Inclusion of such information could help to improve post-therapeutic risk-assessment. To this end, we created pre- and post-therapeutic machine learning-based prognostic models and compared their performance. Material and Methods A single center retrospective cohort of 136 STS patients treated with radiation was analyzed for patients characteristics, staging information, treatment modalities and radiation therapy-specific information. For neoadjuvant patients, postoperative MRI and pathology reports were analyzed for signs of response. Random forest machine learning-based models were used to predict patient’s death and disease progression at two years (2y). For pre-therapeutic models staging information, including histology and anatomic site, were used as input features. For post-treatment models therapy-related data was added. To evaluate the importance of single features for model performance, the reduction in AUC after exclusion of the respective feature was determined. Results The prognostic models achieved high performances up to an area under the curve (AUC) of 0.88 (see Figure 1). Pre- treatment models predicting 2y survival, 2y local and systemic disease progression showed high performance achieving AUCs of 0.73, 076 and 0.63, respectively. Adding post-treatment features improved the overall performance for all three classification types: prediction of death (AUC of 0.87), and of local and systemic progress (AUC of 0.88 and 0.84, respectively). In the feature importance analysis, age appeared to be a strong discriminative feature for pre- and post-therapeutic models. For pre-treatment models, tumor side appeared to be among the most discriminative features with an impact of nearly 0.20 AUC. For post-treatment models, chemotherapy and radiotherapy-related features, such as the planning target volume and total dose, had preeminent importance for prognostic performance with AUC reductions up to 0.12. The volume of the primary tumor after radiotherapy and its relative decrease during radiation were used for prediction of systemic progression showing potential for response-assessment.

Figure 1:

Conclusion Machine learning-based prognostic models combining known prognostic factors with treatment- and response- related information showed high accuracy for individualized risk-assessment of death and disease progression outperforming pre-therapeutic models. These models could be used for adjustments of follow-up procedures in patients that received radiation therapy. OC-0388 Dexamethasone for prevention of pain flare; results from a phase 3 trial in painful bone metastases Y. Van der Linden 1 , P. Westhoff 2 , R. Stellato 3 , N. Kasperts 4 , A. Van Baardwijk 5 , K. De Vries 6 , A. Reyners 7 , A. De Graeff 8 1 Leiden University Medical Center LUMC, Radiotherapy / Centre of Expertise Palliative Care, Leiden, The Netherlands 2 Radboud University Medical Center, Radiotherapy, Nijmegen, The Netherlands 3 University Medical Center Utrecht, Julius Center BioStatistics, Utrecht, The Netherlands 4 University Medical Center Utrecht, Radiotherapy,Utrecht, The Netherlands 5 MAASTRO, Radiotherapy, Maastricht, The Netherlands 6 Netherlands Cancer Institute NKI / AvL, Radiotherapy, Amsterdam, The Netherlands 7 University Medical Center Groningen, Medical Oncology / Centre of Expertise Palliative Care, Groningen, The Netherlands 8 University Medical Center Utrecht, Medical Oncology / Centre of Expertise Palliative Care, Utrecht, The Netherlands Purpose or Objective About 30% of advanced cancer patients treated with radiotherapy for painful bone metastases experience a transient aggravation of pain after radiotherapy, the so- called pain flare (PF). Previously, a Canadian placebo- controlled randomized study showed that five daily 8 mg doses of dexamethasone reduced the incidence of PF from 35% to 26% (p= 0.05, Chow, Lancet Oncol 2016). The aim of our study was to compare two different schedules of dexamethasone versus placebo to prevent the occurrence of PF after radiotherapy for painful bone metastases. Material and Methods We performed a double-blind, randomised, placebo- controlled trial, including patients from 12 out of 21 Dutch radiotherapy centres. Patients were treated with a single fraction of 8 Gy (80%) or 20-24 Gy in 5-6 fractions (20%). They were randomly allocated to receive: A- placebo at least 1 hr before start of treatment (day 1) and then every day for 3 days afterwards, B- 8 mg dexamethasone (day 1) followed by 3 days of placebo, or C- four daily 8 mg doses of dexamethasone (days 1-4). We included patients with uncomplicated painful bone metastases from solid tumors. Patients reported worst pain scores and opioid analgesic intake before treatment and daily for 14 days after (start of) radiation treatment, and on day 28 using the brief pain inventory (BPI). They completed the European Organisation for Research and Treatment of Cancer (EORTC) quality of life QLQ-C15- PAL, and the bone metastases module (BM22) at baseline, and at day 8, 15 and 28. PF was defined as at least a two- point increase on a scale of 0 to 10 in the worst pain score with no decrease in analgesic intake, or, a 25% or greater increase in analgesic intake with no decrease in