ESTRO 38 Abstract book

S205 ESTRO 38

gadolinium injection (Dotarem® 279.3 mg/ml, Guerbet Roissy, France). From the multi-echo data both the T 1 - weighted dynamic curve and the R 2 * -weighted curve were extracted for dynamic contrast enhanced (DCE) and dynamic susceptibility contrast (DSC) analysis, respectively. The extended Tofts model was applied for DCE analysis, estimating transfer constants (K trans , k ep ) and plasma- and interstitial volumes (v p , v e ). The model-free deconvolution approach was used for DSC analysis, calculating tumour blood flow (BF). For both DCE and DSC analysis, individual arterial input functions were used. A bi-exponential model was used for IVIM analysis, estimating perfusion fraction (f), pseudo-diffusion (D * ) and diffusion (D). Tumours were delineated by two radiologists on T 2 -weighted images and co-registered to parametric images before median tumour values were extracted. Pearson correlation coefficients were estimated to evaluate correlations between parameters, Student’s t- test to assess differences between patients with good and poor tumour response to the neoadjuvant radiation and Cox-regression for survival analysis. The median follow-up for the selected patients was 21 months (range 3 - 51 months). Results BF was significantly higher in tumours with good response to the treatment (ypT0-1, n = 9; BF = 120 ml/min/100 g) compared to those with poor response (ypT2-4, n = 34; 96 ml/min/100 g) (p = 0.014). High BF was also significantly associated with overall survival (hazard ratio = 0.97, p = 0.017), Figure 1. Further, BF was correlated to both D * (r = 0.46, p = 0.001) from IVIM and v p (r = 0.57, p <0.001) from the DCE analysis.

Purpose or Objective Prediction of overall survival is important to select appropriate treatment for the individual patient. In order to use prediction models clinically, they require independent validation to check their robustness between different institutions (Type 4 TRIPOD external validation). The aim of this study was to validate the larynx survival model published by Egelmeer et al 2011 in an external cohort from one institution. Material and Methods Using the Computer Assisted Theragnostics (CAT) approach all (n=615) patients treated for larynx cancer were retrieved from the Danish DAHANCA database from 2005-2015 and using the CAT system combined with data from our local Record and Verify system (Mosaiq). Patients that received surgery alone (n=58) or those with missing data (n=169 - predominantly hemoglobin at start of RT) were excluded. In total, 388 patients were used for the validation. All patients were treated according to the DAHANCA guidelines and followed up yearly until year 5. The median follow-up of the 388 patients was 2.6 years. The parameters used in the model were: age, hemoglobin and equivalent dose (EQD2T) as continuous variables, and T-classification (1-4), N- classification (-/+), sex (F/M) and tumor site (glottis/other) as categorical variables. The 2-year and 5-year model performance was validated using calibration plots with the patients grouped in 10 equal sized groups. The plots contain raw data as open circles (1 = alive) and group averages as filled circles. The error bars represent one standard deviation. The validation cohort was split into three groups (two outer quartiles and the inner quartiles as one group) according to the risk assessment of the published model in a Kaplan- Meier plot. The Harrell C-index, the equivalent of AUC for a survival model, was calculated. Bootstrap was used to calculate a 95% confidence interval. Results The external model validation shows a very good calibration for the 2 year survival, while the 5-year survival model underestimates the survival (offset), however with a good calibration slope (fig. 1). The three risk groups showed highly significant (p<0.001) different survival rates in the Kaplan-Meier plot (fig. 2), validating that the model discriminates this larynx patient cohort well. The 2- and 5-year survival rates for the three risk groups were: low risk 93.0±2.8% and 90.1±3.9%, intermediate risk 84.9±2.7% and 66.5±4.3%, and high risk 45.1±5.2% and 20.6±4.8% (std error). The model has a high Harrell C-index of 0.78 [0.74-0.82] compared to 0.73 [0.70-0.77] from the original article.

Conclusion In this cohort of rectal cancer patients, BF from DSC acquired at baseline, reflecting tumour perfusion, predicted both local tumour response to neoadjuvant radiation as well as overall survival. OC-0403 Type 4 TRIPOD external validation of a larynx survival model C. Rønn Hansen 1,2,3,4 , N. Sarup 1 , R. Zukauskaite 4,5 , J. Johansen 5 , J.G. Eriksen 5,6 , S.L. Krogh 1 , A. Bertelsen 1 , D.I. Thwaites 2 , U. Bernchou 1,4 , C. Brink 1,4 1 Odense University Hospital, Laboratory of Radiation Physics, Odense, Denmark ; 2 The University of Sydney, Institute of Medical Physics- School of Physics, Sydney, Australia ; 3 Aarhus University Hospital, Danish Centre for Particle Therapy, Aarhus, Denmark; 4 University of Southern Denmark, Department of Clinical Research, Odense, Denmark; 5 Odense University Hospital, Department of Oncology, Odense, Denmark; 6 Aarhus University Hospital, Department of Experimental Clinical Oncology, Aarhus, Denmark