ESTRO 2020 Abstract book

S248 ESTRO 2020

trials. Both use markedly different delineation protocols to generate similiar target volumes. This study aims to see if one protocol introduces more variation in delineated volumes, identify common sources of error or variation and provide insights to inform future protocols. Material and Methods Ten clinical oncologists with experience in upper GI cancers from across the UK were given two identical CT datasets (a mid and lower-third oesophagus case) with pre- delineated GTV. The clinicians were asked to generate target volumes (CTVA, CTVB, PTV) for each dataset using protocols from the NeoSCOPE and Neo-AEGIS studies. Submitted volumes were reviewed by an experienced RTQA clinician with protocol deviations identified. To evaluate the inter-observer concordance, the generalized conformity index (CI gen ), defined as the ratio of the sum of all overlapping volumes between pairs of observers and the sum of all overlapping and all non-overlapping volumes between the same pairs, was used. A CI gen of 1 indicates 100% concordance and a CI gen of 0 indicates no concordance in delineation. Results

For future trial delineation protocols, we recommend utilising clear anatomical landmarks to avoid ambiguity when defining target volumes. Where possible, we recommend utilising geometric expansion of GTV with editing and to limit the use of a ‘free-hand’ approach. We also recommend including images and multiple worked examples to reflect different cases and situations. PD-0423 Risk group classification for locoregional failure following upfront surgery for pancreatic cancer A. Elamir 1 , S. Gallinger 2 , C. Moulton 2 , I. McGilvary 2 , L. Lu 3 , W. Xu 3 , J. Knox 4 , R.M. Prince 4 , G. O'Kane 4 , J. Kim 1 , J. Ringash 1 , L.A. Dawson 1 , R. Wong 1 , A. Barry 1 , J. Brierley 1 , A. Hosni 1 1 Princess Margaret cancer Center, Radiation oncology department, Toronto, Canada ; 2 Princess Margaret cancer Center, Department of Surgery, Toronto, Canada ; 3 Princess Margaret cancer Center, Department of Biostatistics, Toronto, Canada ; 4 Princess Margaret cancer Center, Medical oncology department, Toronto, Canada Purpose or Objective The aims of the study were: (1) to identify predictors of locoregional failure (LRF) following upfront Whipple, distal or total pancreatectomy for resectable pancreatic adenocarcinoma (PA) (2) Development of a prediction risk score model of LRF, and (3) evaluation of the impact of post-operative radiation therapy (PORT) in the identified subgroups. Material and Methods Following research ethics board approval, a retrospective review was conducted on patients (pts) with stage I-III PA who underwent upfront surgery with adjuvant chemotherapy (± PORT) at our institution between 2005- 2016. PORT was offered based on multidisciplinary tumor board discussion, patient preference, and was considered in fit patients, with positive margins or nodes. Analysis was performed in three steps: 1) Identification of LRF predictors in pts not receiving PORT: univariable analysis (UVA) using Cox proportional-hazard model was used to evaluate clinical and pathological factors associated with increased risk of LRF. All variables with p-value <0.1 in UVA were entered into multivariable analysis (MVA) and <0.05 to stay in the risk score model. 2) Development of prediction risk score model for LRF: The risk score was calculated based on the sum of coefficients of the predictors of LRF. The high (HR) and low (LR) risk group cut point was determined by a minimal p value approach. 3) Evaluation of the impact of PORT on the identified risk groups: The developed model was applied to the entire cohort (including pts who received PORT), then the LRF was calculated in HR and LR groups among those who received PORT vs those who did not receive PORT. Results A total 480 pts were identified, of whom: 395 (82.3%) had pT3, 4 had pT4, 347 (72%) had pN+, 391 (82%) had Grade 2-3, 171 (36%) had involved or close (<1 mm) margins, and 24 (5%) received PORT. With a median follow up of 13 months, the 1 yr-LRF for the entire cohort was 23% (95% CI 19-27 %). Results from our analysis showed that: Predictors of LRF were: pT3-4, pN+, Grade 2-3, and involved or close margins (See table 1). The HR group (n=266, 55%) was defined by the sum of risk score coefficients ≥ 1.2. The 1- year LRF for the HR group vs the LR group was 29% vs 17% (p<0.001); Figure 1A. After including pts who received PORT: (1) among the HR group, the 1 yr-LRF was 31% for pts who did not receive PORT (n=246) and 5% among pts who received PORT (n=20, p=0.03), (2) Among the LR

Middle Third Lower Third Volume NeoSCOPE Neo-AEGIS NeoSCOPE Neo-AEGIS CTVA 0.93 0.95 0.92 - CTVB 0.84 0.70 0.76 0.60 PTV 0.84 0.77 0.81 0.67 Table 2: Generalised Conformity Index (CIgen) for CTVA, CTVB and PTV Middle Third Lower Third

Image 1: CTVB contours for lower third case below the gastro-oesophageal junction.

Conclusion There was more TVD variation using the Neo-AEGIS protocol. For lower third cases, 6/10 cases had unacceptable variation from protocol. The most significant non-concordance was seen in the CTVB volume using the Neo-AEGIS protocol for lower third cases. Analysis of the protocols show that the NeoSCOPE uses systematic geometric expansion of GTV which is then edited back to generate CTVB. Neo-AEGIS uses anatomical landmarks to define the ’fat pad’ which forms the CTVB. Below the GOJ, both protocols use a ‘free-hand’ approach with anatomical landmarks to define CTVB. Both protocols included sample images, but in addition, NeoSCOPE included multiple worked examples.