ESTRO 38 Abstract book

S105 ESTRO 38

thereby indicating good generalizability of the models.

Material and Methods This was a prospective cohort study including 750 HNC patients treated with definitive radiotherapy. All endpoints were scored prospectively and the compliance rate was > 80% for all endpoints. OARs were re-delineated according to international consensus guidelines. Multiple imputation was used to deal with missing data to avoid the potential bias of a complete case analysis. During model development we assessed non-linear dose-effect relations and dealt with multicollinearity. Models were developed for moderate-to-severe PR and grade ³ 2 PhR (CTCAEv4.0) xerostomia, sticky saliva and changes in taste at 6 months after treatment. A closed testing procedure was used to validate and, if necessary, update the models at subsequent time points (up to 5 years) and higher toxicity grades (fig 1). The models at 6 months were externally validated on 260 HNC patients from 2 other RT-centers using the same closed testing procedure. The proposed model update was performed on the combined data to obtain a more generalized model.

Conclusion We developed and externally validated a comprehensive NTCP-profile for both PR and PhR salivary gland toxicities at different time points after head and neck radiotherapy, based on the newest OARs contouring guidelines. Such a profile can be used for treatment planning optimization and selection of patients for different RT techniques. PV-0202 3-D reconstruction of radiotherapy dose associated with advanced osteoradionecrosis after IMRT A.S.R. Mohamed 1 , K. Al Feghali 1 , S.P. Ng 1 , H. Elhalawani 1 , K. Hutcheson 1 , M. Chambers 1 , J. Phan 1 , J. Kraeima 2 , H. Glas 2 , M. Witjes 2 , G.B. Gunn 1 , A. Garden 1 , D. Rosenthal 1 , S. Frank 1 , W. Morrison 1 , C. Fuller 1 , S. Lai 1 1 The University of Texas- MD Anderson Cancer Center, Radiation Oncology, Houston, USA; 2 University of Groningen- The Netherlands, Oral Surgery, Groningen, The Netherlands Purpose or Objective IMRT carries the ability to limit dose to organs-at-risk. However, recent findings by our group and others, reported no reduction in osteoradionecrosis (ORN) rates in oropharyngeal cancer (OPC) patients after IMRT in comparison to conventional radiotherapy techniques. We recently demonstrated that a wide range of two- dimensional dose-volume parameters in the intermediate and high dose beam-path are associated with the development of advanced ORN in patients with OPC treated with IMRT. To this end, we aim to further characterize the dosimetric correlated of advanced ORN by determining the three-dimensional (3-D) spatial dose distribution of the mandibular area of ORN. Material and Methods After institutional review board (IRB) approval, we identified patients with grade IV ORN requiring major surgery among patients with OPC treated with IMRT between 2002 and 2013. The initial CT scans documenting the diagnosis of ORN were identified. The mandibular areas affected with ORN were manually segmented for all patients to create 3-D ORN volume of interest (ORN-VOI). Planning CTs and dose grids were subsequently retrieved. ORN-depicting CT scans were then co-registered to planning CT scans using a validated commercial image registration software (Velocity AI 3.0.1, Atlanta, GA). Finally, ORN-VOIs were mapped to planning CT scans, and dose grid then dosimetric parameters were extracted for each VOI (Figure 1).

Results Salivary dysfunction was best predicted by baseline complaints and mean dose to the parotid and submandibular glands and the oral cavity (table 1). For PR changes in taste, age was also a predictor. In most models, a square root or log transformation of the parotid glands or oral cavity dose improved model performance. All models performed well with an ROC-AUC ranging from 0.65 for PhR changes in taste to 0.77 for PhR xerostomia. When validated over time, none or minor model adjustments were necessary. The external validation showed good performance of the models and only minor updates sufficed to allow proper model performance,