ESTRO 38 Abstract book

S466 ESTRO 38

there are a limited number of mortality prediction tools such as the TEACHH and Chow models. Here we build a model incorporating multiple factors to predict 30-day mortality for patients treated with palliative radiotherapy (RT) for advanced cancer. Material and Methods A cohort (n=518) of patients treated with external beam RT to a site of metastatic disease between 2012-2016 was included. Factors that may be associated with death within 30 days of RT treatment, including demographics and clinical and laboratory data were retrospectively collected. Generalized linear models (GLMNET package) were built with no regularization (logistic regression), and with regularization (Lasso and Elastic Net). Random forest (RANGER) and gradient boosting machine (GBM) models were also built to assess non-linearity in the response. Missing data was imputed by replacing values with either the mean or the mode, and adding a variable which indicated which patients had missing data for each variable. All models were built in R using cross validation withholding 25% of the data (test set), and stratified subsampling to account for unbalanced outcome. Each model was seeded in order to enable a pairwise t-test to compare the distribution of test balanced accuracies of each model. The area under the ROC curve (AUC) was also reported. The variable importance for linear models was calculated by multiplying the coefficient with the standard deviation of that variable for normalization. Results Lung, breast, and prostate cancer accounted for 55% of primary malignancies, with bone and brain accounting for 85% of treatment sites. 125 patients (24%) died within 30 days of RT. A logistic regression model resulted in a mean balanced accuracy (MBA) of 0.76 (AUC = 0.83). Adding Lasso or Elastic Net regularization offered a small improvement in MBA of 0.78 for both models (p=0.07, p=0.09) and an AUC of 0.83 and 0.84 (Figure 1) respectively. The non-linear models proved worse than logistic regression with random forest having an MBA of 0.74 (p=0.01, AUC = 0.82) and GBM having an MBA of 0.75 (p=0.08, AUC = 0.83). The most important variables in the logistic regression model are those indicative of missing data in other variables. Both Lasso and Elastic Net (Figure 2) regularizations, however, result in more interpretable variables such as TEACHH score, performance score, and hospice status. These features also have high importance in both non-linear models. For comparison, the AUC when using TEACHH score alone was 0.69.

benefit the most from SABR for spinal metastases. Probably, these metachronous lesions are the main reflection of a true oligometastatic state. However, these findings need to be investigated further in larger and prospective trials. PO-0883 Phase II trial of stereotactic body radiation therapy for abdomino-pelvic lymph node oligometastases C. Franzese 1 , T. Comito 1 , A. Tozzi 1 , F. De Rose 1 , P. Navarria 1 , P. Mancosu 1 , S. Tomatis 1 , M. Scorsetti 1 1 Humanitas Research Hospital, Radiotherapy and Radiosurgery, Rozzano Milan, Italy Purpose or Objective Stereotactic body radiotherapy (SBRT) is nowadays considered an effective approach for the management of oligometastatic patients. Few data exist on radiotherapy in the context of isolated or limited lymph node metastases. We analyzed clinical results of oligometastatic patients treated with high dose SBRT for lymph node metastases in abdomen and pelvis. Material and Methods This is a prospective, phase 2 trial. Primary end-point was the assessment of acute and late toxicity; secondary end- points were local control (LC), progression free survival (PFS) and overall survival (OS) The schedule of SBRT was 48 Gy delivered in 4 fractions of 12 Gy each. Inclusion criteria were: Histologically-proven carcinoma of gastro- intestinal, genito-urinary of gynecological primary site, WHO performance status ≤ 2, maximum 3 lymph node sites of disease, maximum diameter ≤ 5 cm, abdomen-pelvic site. Physical examinations and toxicity assessments were performed during and after SBRT according to CTCAE v4.0. Tumour response was evaluated on CT-MRI-PET scans using the RECIST modified criteria. Results From 2015 to 2018, 41 patients with 52 lymph nodes were enrolled. Median age was 69.2 years and 87.8% of patients were male. Most common primary tumour was located in genito-urinary tract (70.7%), and in particular prostate adenocarcinoma (58.%%), followed by gastro-intestinal (26.8%) and gynecological (2.5%) disease. One single lymph node was treated 32 patients, while 2 and 3 lymph node in 7 and 2 patients, respectively. Systemic therapy was administered before SBRT in 43.9% of patients and during SBRT in 14.6%. Median clinical target volume diameter was 15 mm (6 – 49). With a median follow-up of 16.7 months, only 3 patients reported grade 1 acute toxicity, in the form of pain, dysuria and fatigue. In the late setting, chronic pain was observed in 1 patient. In-field progression was observed in 5 (12.2%) patients with a 1- and 2- years rate of 96.4% and 77.1%. Systemic therapy during SBRT was associated with worse LC (HR 8.6, 95%CI 1.41 – 52.7, p=0.020). Out-field lymphnodal progression was observed in 17 (41.4%) of cases and distant progressions in 8 (19.5%) cases. Median PFS was 12.8 months. At time of analysis all patients were alive except one with 1- and 2-years OS of 100% and 94.7%. Conclusion Treatment of lymph node metastases with high dose SBRT can be considered a safe option with high rates of local control in the context of multidisciplinary management of oligometastatic patients. PO-0884 Predicting 30-day mortality for palliative radiotherapy A. Witztum 1 , S. Wu 1 , E. Gennatas 1 , G. Valdes 1 , T. Solberg 1 , S. Braunstein 1 1 Department of Radiation Oncology, University of California- San Francisco, San Francisco, USA Purpose or Objective Decision support for cancer patients at end of life is needed to optimize outcome for this population. Currently