ESTRO 36 Abstract Book

S90 ESTRO 36 _______________________________________________________________________________________________

Purpose or Objective The aim of this study was to evaluate the acute and late toxicities and biochemical disease-free survival and overall survival after high-dose-rate brachytherapy as a salvage modality for locally recurrent prostate Between 2007 and 2014, we retrospectively analyzed 20 consecutively patients. Median age of first treatment was 62 years (range 51-73). The majority of the patients in this study (65%) were low risk. 5p received hormonal blockade. 11p received treatment with low-dose-rate brachytherapy (LDR-BT) and 9p received treatment with external beam radiotherapy with median dose of 75Gy (70- 78Gy). Time to biochemical recurrence was 62 months (range 14-119). Median presalvage PSA was 3.72 (range 1,83-12,29). After biochemical relapse, we confirm local recurrence with biopsy. Patients received high-dose-rate brachytherapy (HDR-BT). The schedule was three implantations, every two weeks, with 10,5Gy per implant. By the time of salvage BT, only 1p received ADT. Acute and late genitourinary and gastrointestinal toxicities were graded using Common Terminology Criteria for Adverse Events (CTCv4.0). Overall survival (OS) and biochemical (bDFS) control were calculated using Kaplan-Meier method. Results After first treatment, acute toxicities consisted of genitourinary toxicities grade 1 (3p) and grade 3 (1p). Not late gastrointestinal toxicities. After HDR-BT, acute toxicities consisted of genitourinary grade 1 (4p), grade 2 (5p) and grade 3 (3p), gastrointestinal toxicities grade 1 (3p) and grade 2 (4p) and impotence in 4p. Not acute toxicities grade 4 were reported. Late toxicities consisted of genitourinary grade 3 were observed in 2p. Not grade 4 complications. With a median follow-up after salvage HDR-BT of 47 months (range 11-112 months), local control was achieved on PSA levels in all patients. Among 20 patients studied, 1 lost follow-up and he was excluded from the survival analysis. Using Kaplan-Meier analysis the 2-year and 5-year OS were 100% and 84,2%, respectively. The 2-year and 5-year biochemical disease-free survival (bDFS) were 85% and 81%, respectively. Conclusion Prostate BT is an effective salvage modality in some selected prostate local recurrence patients after radiation therapy. HDR-BT is a good choice to deliver high-dose radiation in prostate recurrence tumors after external beam radiotherapy or LDR-BT. This treatment offers adequate locoregional control with acceptable range of complications. OC-0176 Identifying Patients Who Benefit the Most from Salvage HDR Brachytherapy G. Valdes 1 , A.J. Chang 1 , O. Kenton 1 , A. Cunha 1 , T.D. Solberg 1 , H. I-Chow 1 1 University of Californ ia UCSF, Radiation Oncology, San Francisco CA, U SA Purpose or Objective To use mac hine learning to better identify patients that could benefit from prostate salvage HDRB (HDR brachytherapy). Material and Methods Data was analyzed for 52 consecutively accrued patients that underwent salvage HDRB between 1998 and 2009 for locally recurrent prostate cancer following previous definitive radiation therapy at the University of California, San Francisco (UCSF). All patients were treated with 36 Gy in 6 fractions after pathologic confirmation of locally recurrent disease without evidence of metastatic radiotherapy failure. Material and Methods

disease. Determination of biochemical failure after salvage HDRB was based on the Phoenix definition. All non-failure patients were followed for a minimum of 5 years. Eighteen different clinical risk features were collected from each patient. Machine Learning was used to identify subpopulations that would most likely to remain biochemically disease free after the treatment. Decision tree algorithms were constructed using Matlab R 2011a. The complexity of the decision tree was fine-tuned by selecting the optimum number of observations per terminal node that minimized the “Leave One Out Cross-Validation” estimation of the deviance. Results were compared to those obtained using Ensemble Methods. Random permutation experiments were also performed to estimate the probability that the tree found was the result of random variations. Results A subpopulation of patients with a high risk of biochemical failure after salvage HDRB was identified. Those patients with a fraction of positive nodes from those sample that was greater than 0.354 and disease free interval less than 4.12 years had a failure rate after salvage HDRB of 0.75 vs 0.38 for the remainder of the population, Figure 1 . The probability that the conclusions reached in this paper are not due to random fluctuations is 0.7, Figure 2 .

Figure1. The Optimal Decision Tree obtained for predicting failure after Salvage HDRB.

Figure 2. Random permutation of output labels. 1000 iterations were created in each case were the outcome of was randomly permuted. Not correlation between features and outcome should be present in this case. The probability of obtaining a cross-validated leave one out error smaller than the one obtained by the tree in Figure 1 was calculated. Conclusion Patients with a fraction of positive nodes higher than 0.35 and a disease free interval bigger than 4.12 years are at higher risk of biochemical failure after salvage HDBR. Machine Learning is effective in identify subtle variables that can affect the treatment outcome.