ESTRO 2020 Abstract book

S261 ESTRO 2020

Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE TM ). Material and Methods PRO-CTCAE questionnaires were administered at baseline, end-of-treatment, 3, 6, 12 months, then annually. Eligible patients were treated with radiation therapy with curative intent and completed selected PRO-CTCAE items based on cancer type between 2013 and 2019. A patient was considered to have patient-reported treatment-related symptomatic AEs if he/she had a score of 3 or 4 for any PRO-CTCAE item that was higher than baseline. Artificial neural network analysis with 4-fold cross-validation was utilized to predict patient-reported AEs with local interpretable model-agnostic explanations (LIME) to approximate clinical effect. All data was analyzed to determine the predictive power of the artificial neural network. Inherent differences in the proton vs. photon patient populations mandated a propensity score analyses utilizing inverse probability of treatment weighting (IPTW) to assess the effect of modality on patient-reported AEs. Artificial neural networks were utilized to predict proton vs. photon treatment assignment to create the inverse weights, while logistic regression models assessed the effect of modality for patient-reported AEs. Results 1,930 patients were eligible; 695 received proton and 1481 received photons. The median time from treatment to last survey was 33.6 mos. Site-specific patient-reported AE rates include(N): 44.2% Prostate (821), 6.7% Breast (624), 20.0% Esophagus (129), 11.2% Pancreas (106), 30.8% Colorectal (91), 37.3% Head & Neck (76), 26.4% Sarcoma (71), and 58.3% Gynecologic (12). Overcoming the subjective nature of patient-reported AEs, the artificial neural network correctly predicted 76.6% (95% CI: 74.5%, 79.6%) of AEs. Prostate cancer and higher total doses were associated with increased risk of patient-reported AEs while breast cancer, pancreas cancer, and hypo- fractionation were associated with lower risk of patient- reported AEs (Fig.1A). Artificial neural networks and the resulting LIMEs showed deterministic factors for modality (Fig.1B). Subsequently, propensity score analysis demonstrated that patients treated with proton therapy had significantly less patient-reported AEs (p=0.045) with an odds’ ratio of 0.79 (95% CI: 0.63, 0.99) when accounting for baseline differences. Conclusion This large, multi-site prospective registry demonstrates a significant benefit of proton therapy in reducing patient- reported AEs. Based on patient characteristics and treatment, artificial neural networks can predict patient- reported AEs after treatment completion. Utilizing multi- level modelling an adjusting for confounding, the direct treatment effect of proton therapy can be predicted. OC-0441 An interim analysis of outcome data from the UK Proton Overseas Programme S. Gaito 1 , N. Burnet 1,2 , M. Aznar 3 , P. Foden 4 , C. Howell 4 , S. Pan 1 , D. Saunders 1 , G. Whitfield 1,2 , A. Crellin 5 , E. Smith 1,2,4 1 The Christie, Proton Beam Therapy Centre, Manchester, United Kingdom ; 2 The University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom ; 3 The University of Manchester, Radiotherapy Related Research, Manchester, United Kingdom ; 4 The Christie, Proton Clinical Outcomes Unit, Manchester, United Kingdom ; 5 NHS England, National Clinical Lead Proton Beam Therapy, Manchester, United Kingdom

Results In total, 10 short (SCRT) and 10 long (LCRT) course radiotherapy treatment schedules were included, resulting in 300 fractions for evaluation. Coverage, expressed as D99% remained the same. For the total cohort the median volume of the normal tissue irradiated with 95% of the prescribed dose dropped from 642 cm 3 (PS) to 237 cm 3 (online ART), which was statistically significant (Figure 2). Online ART reduced dose to the bowel bag and bladder for all tested dose levels significantly for both SCRT and LCRT. The average difference for bowel bag was -127 cm 3 for the volume receiving 0.6Gy per fraction in LCRT and -57 cm 3 the volume receiving 95% of the prescribed dose in SCRT. The average difference for bladder was -24% for the volume receiving 0.6Gy per fraction in LCRT and -9% and for the volume receiving 95% of the prescribed dose in SCRT. The average difference for the total cohort in Dmean for bladder was -7%. Conclusion Radiotherapy with online ART of locally advanced rectal cancer reduces dose to the bladder and small bowel significantly for large volumes, compared to a clinically implemented plan selection strategy. OC-0440 Proton Therapy Reduces Patient-Reported Adverse Events: A Neural Network for Large-Volume Practice T. DeWees 1 , M. Golafshar 2 , M. Petersen 2 , T.J. Whitaker 3 , A. Amundson 4 , K. Klein 4 , T. Pisansky 4 , B. Davis 4 , K. Corbin 4 , B. Stish 4 , C.R. Choo 4 , C. Hallemeier 4 , R. Mutter 4 , W. Wong 5 , I. Petersen 4 , S. Park 4 , T. Daniels 5 , L. McGee 5 , N. Laack 4 , A. Dueck 2 , C. Vargas 5 1 Mayo Clinic, Health Sciences Research- Radiation Oncology, Scottsdale, USA ; 2 Mayo Clinic, Health Sciences Research, Scottsdale, USA ; 3 Baylor University, Radiation Oncology, Waco, USA ; 4 Mayo Clinic, Radiation Oncology, Rochester, USA ; 5 Mayo Clinic, Radiation Oncology, Phoenix, USA Purpose or Objective We instituted a prospective registry to capture provider- and patient-reported clinical outcomes in a large, multi- site radiation oncology practice. Herein, we evaluate the association between proton vs. photon therapy and adverse events as (AEs) assessed by the Patient-Reported Proffered Papers: Proffered papers 21: Protons

Purpose or Objective