ESTRO 2020 Abstract book

S349 ESTRO 2020

and neck cancer (HNC) patients that will be used for patient selection in the randomized controlled DAHANCA35 trial, investigating photons vs. protons in HNC patients. Material and Methods Patients treated with IMRT in DAHANCA19 were contoured according to recent international recommendations. Dose metrics for all 588 patients were extracted using MatLab and all clinical parameters were retrieved from the DAHANCA database. The new Dutch dysphagia NTCP model contains the mean dose to the upper, mid and lower pharyngeal constrictor muscle, the extended oral cavity, baseline dysphagia complaints (grade 0-1, 2, 3+) and tumor site (oral cavity, pharynx, larynx). Four dysphagia models, ≥ grade 2 and ≥ grade 3, and at 6 months (± 1 mo) and 12 mo (± 1.5 mo) after treatment, were externally validated using the DAHANCA cohort in a closed testing procedure. In this procedure, a change of the model intercept, slope, or individual regression coefficients is tested for significant prediction improvements. If the model fits the data well the procedure terminates at the first step, thus avoiding multiple testing. The model performance was evaluated using calibration plot, AUC and Integrated Calibration Index (ICI). Results At 6 and 12 mo, 286 and 267 complete cases were identified, respectively. Dysphagia grade 2+ was seen in 31% at 6 mo and 22% at 12 mo, and grade 3+ in 10% and 6% at 6 and 12 mo, respectively. The closed testing procedure for physician-rated dysphagia showed that the original 6 mo grade 2+ model matched the Danish data with no model modification needed resulting in an AUC of 0.68 and low ICI of 0.03 (upper fig 1). The grade 3+ models at 6 and 12 mo needed a small but significant intercept change, resulting in a high AUC of 0.79 in both models (lower fig 1 and fig 2). The 12 mo grade 2+ model overestimated the risk of dysphagia in the corresponding validation data (red point in upper fig 2 shown original calibration), which mandated a refit of the regression coefficients for best matching to the Danish patients. Here the model AUC significantly improved from 0.71 to 0.78 and reduced the ICI from 0.09 to 0.02.

Figure 2. The decision tree and number of patients met the minimum criteria for eligibility for proton therapy on the basis of plan comparison. Conclusion Model-based selection of patients with HNC for proton therapy is clinically feasible. Around one third of HNC patients qualify for protons and these patients have the highest probability to benefit from protons in terms of prevention of late radiation-induced side effects. OC-0575 Type 4 validation of dysphagia NTCP model for selection of HNC patients to the RCT, DAHANCA35 C.R. Hansen 1,2,3 , L. Van den Bosch 4 , H.P. Van der Laan 4 , J. Friborg 2,5 , K. Jensen 2 , E. Samsøe 2,6 , L. Johnsen 1 , R. Zukauskaite 3,7 , C. Grau 2,8,9 , C. Maare 6 , J. Johansen 7 , H. Primdahl 9 , Å. Bratland 10 , C.A. Kristensen 5 , M. Andersen 11 , J.G. Eriksen 8,9 , J.A. Langendijk 4 , J. Overgaard 8 , A. Van der Schaaf 4 1 Odense University Hospital, Laboratory of Radiation Physics, Odense, Denmark ; 2 Aarhus University Hospital, Danish Center of Particle Therapy, Aarhus, Denmark ; 3 University of Southern Denmark, Institute of Clinical Research, Odense, Denmark ; 4 University Medical Center Groningen, Department of Radiation Oncology, Groningen, The Netherlands ; 5 Rigshospitalet, Department of Oncology, Copenhagen, Denmark ; 6 Copenhagen University Hospital Herlev, Department of Oncology, Copenhagen, Denmark ; 7 Odense University Hospital, Department of Oncology, Odense, Denmark ; 8 Aarhus University Hospital, Department of Experimental Clinical Oncology, Aarhus, Denmark ; 9 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark ; 10 Oslo University Hospital, The Norwegian Radium Hospital, Oslo, Norway ; 11 Aalborg University Hospital, Department of Oncology, Aalborg, Denmark Purpose or Objective The use of published NTCP models for patient selection for proton therapy is at the early stages of clinical implementation. NTCP models can be used to select patients with an anticipated benefit from proton therapy in lowering side effects. However, prediction models are developed in patient populations that may differ from the target population and may therefore perform poorly. The aim of this project was to externally validate a new model for physician-rated dysphagia after radiotherapy in head