ESTRO 38 Abstract book

S1031 ESTRO 38

Material and Methods Complete data of 65 patients (pts), including overall, loco- regional relapse and distant metastasis-free survival (OS, LRFS, DMFS) information were available. Pts received 41.4Gy in 18 fr (2.3 Gy/fr) delivering ART concomitant boost on the residual GTV in the last 6 fr (3 Gy/fr, GTV D mean : 45.6Gy). Chemotherapy consisted of oxaliplatin (OXA) 100 mg/m 2 on days -14, 0 (start of RT), and +14, and 5-fluorouracil (5-FU) 200 mg/m 2 /d from day -14 to the end of RT. Uni- and multi-variable Cox regression models for OS, LRFS and DMFS were assessed considering several clinical (age, sex, OXA dose, 5-FU dose, time to surgery, stage) and histological (pCR, pCR or clinical complete response (cCR) followed by surgery refusal (pCR/cCR), Tumor regression grade, Residual vital cells<5%,<10%) variables. High resolution T2-weighted MRI taken before RT (MRI pre ) and at half RT (MRI half ) were available and GTVs were contoured by a single clinician (V pre , V half ). The parameter ERI TCP = -ln[(1 – (V half / V pre )) Vpre ], previously introduced to quantify early response, was considered. Models including/not including ERI TCP (CONV_model and REGR_model respectively) were assessed and their ability in discriminating relapsing pts compared. Results The median time between RCT and surgery was 11w (range:7-19); 63 pts were operated and two refused surgery after cCR; pCR were 20/63 (32%). The median follow-up was 30 months (range: 5.5-100). OS, LRFS and DMFS at 30 months were 96%, 97% and 80% respectively. Due to the few events the analysis was focused on DMFS: the best CONV_model included pCR/cCR (HR:0.12,p=0.038) and 5-FU dose>90% (HR: 0.35,p=0.039), with AUC=0.73 (95%CI: 0.62-0.83). The best REGR_model included ERI TCP (HR:1.019, p<0.0001) and 5-FU dose>90% (HR:0.18,p=0.005); AUC was 0.87 (95%CI: 0.76-0.94), significantly higher than CONV_model (p=0.03, Figure 1). When grouping pts according to the best cut-off value for REGR_model, DMFS at 30 months was 97 % vs 63 % (p=0.0006) for pts below (n=33, 1 event) and above (n=32, 12 events) this value (Figure 2). Higher ERI TCP values were also associated to worse OS (p=0.036). Conclusion Early regression during RCT for Rca pts as modeled by a Poisson-based TCP formula predicted DMFS better than pCR. An independent impact of the individually administered drug dose was also quantified. ERI TCP should be considered as a strong outcome predictor with large potentials in treatment individualization.

1 Institut Jen Godinot, Radiation Oncology, Reims, France ; 2 Centre Hospitalier Universitaire de Reims, Radiology, Reims, France ; 3 Centre Oscar Lambret, Biostatistics unit, Lille, France ; 4 Institut Jen Godinot, Surgery, Reims, France ; 5 Centre Oscar Lambret, Radiation Oncology, Lille, France Purpose or Objective Baseline contrast-enhanced Computed Tomography (CT)- derived texture analysis in locally advanced rectal cancer could be useful in order to offer the best personalised treatment. The purpose of this study was to determine the value of baseline-CT texture analysis in the prediction of downstaging in patients with locally advanced rectal cancer. Material and Methods We retrospectively included all consecutive patients treated with neoadjuvant chemoradiation therapy (CRT) followed by surgery for locally advanced rectal cancer. Tumour texture analysis was performed on the baseline pre-CRT contrast-enhanced CT. Based on the selected model of downstaging with a penalized logistic regression in a training set, a radiomics score (Radscore) was calculated as a linear combination of selected features. A multivariable prognostic model was built including Radscore and clinical factors. Results Among the 121 patients included in the study, 109 patients (90%) were T3-T4 and 99 (82%) N+ at diagnosis. A downstaging response was observed in 96 patients (79%). In the training set (79 patients), the best model (ELASTIC- NET method) reduced the 36 texture features to a combination of 6 features. The multivariate analysis retained, as independent factors, the Radscore (Odds Ratio, OR=13.25; 95%-Confidence Interval, 95%CI, 4.06- 71.64; p<0.001) and the age (OR=1.10/1 year; 1.03-1.20; p=0.008). The model was evaluated in the test set leading to an area under the curve of 0.70 (95%CI, 0.48-0.92). Conclusion This study presents a prognostic score for downstaging, from initial computed tomography derived texture analysis in locally advanced rectal cancer, which may lead to a more personalised treatment for each patient. EP-1898 Encouraging the use of decision support systems in routine clinical practice R. Fijten 1 , B. Reymen 2 , R. Houben 3 , P. Fick 4 , T. Hendrik 4 , S. Puts 1,4 , J. Veugen 4 , A. Dekker 1 1 Maastricht University Medical Centre, Department of Radiation Oncology Maastro, Maastricht, The Netherlands ; 2 Maastro Clinic, Department of Radiotherapy, Maastricht, The Netherlands ; 3 Maastro Clinic, Data Center Maastro, Maastricht, The Netherlands ; 4 Maastro Clinic, Informatisering and Services, Maastricht, The Netherlands Purpose or Objective In the current era of machine learning, many new predictive models are generated yearly. In routine clinical practice, predictive models can assist physicians in interpreting multidimensional clinically relevant data and therefore aiding the (shared) decision making process. However, their use is not widespread due to a number of reasons, such as time constraints during consultations and the difficulty of accessing relevant models. In order to alleviate these concerns and encourage the use of predictive models in routine clinical practice, we have developed an easily accessible web-based interface to host clinically relevant models. Material and Methods A linear regression model was created to determine the survival probability of small-cell lung cancer patients 26 weeks after the start of Prophylactic Cranial Irradiation (PCI). The model was based on data from 151 patients and was internally validated.

EP-1897 Texture analysis of the initial CT to predict the response to neoadjuvant CRT in rectal cancer B. Vandendorpe 1 , C. Durot 2 , L. Lebellec 3 , M. Le Deley 3 , D. Sylla 3 , A. Bimbai 3 , K. Amroun 4 , F. Ramiandrisoa 1 , A. Cordoba 5 , X. Mirabel 5 , C. Hoeffel 2 , D. Pasquier 5 , S. Servagi-Vernat 1