ESTRO 2023 - Abstract Book

S28

Saturday 13 May

ESTRO 2023

therapy, of which 23 patients exhibit Contrast-enhancing Brain Lesions (CEBL). We have recently analyzed this cohort by means of voxel-level NTCP modeling, revealing clinical evidence for a variable proton RBE and an increased sensitivity of the periventricular 4mm-fringe [1]. Both effects cannot be accounted for on a patient level by established NTCP methods. Here, we calculated an enhanced EUD (eEUD_varRBE,VP) that incorporates the voxel-level model predictions and regressed a univariate NTCP model. To discern the different contributions on model performance, we regressed additional models based on: an EUD with variable RBE (EUD_varRBE), thus ignoring the periventricular sensitivity; the best-performing DVH parameter D5ml with variable (D_5mlvarRBE); and with fixed RBE (D_5mlRBE=1.1). We compared model performance by: the dose-response steepness parameter γ ; the Akaike information criterion AIC; and the AUC of the ROC curve. Results The results are listed in Tab. 1. eEUD_varRBE,VP outperforms all other measures and yields a very steep dose response with a steepness parameter γ =5.4. The D50 is listed for completeness.

Conclusion In our example study, incorporating the periventricular sensitivity into the eEUD increases model performance and particularly steepness of the dose response. This not only allows more accurate risk evaluation, but the virtual dose employed to calculate the eEUD can be directly translated into treatment planning in the form of modified dose constraints to organs at risk. [1] E. Bahn, et al. Int J Radiat Oncol Biol Phys 107(3), 571–578, 2020, doi: 10.1016/j.ijrobp.2020.03.013. MO-0061 Independent external validation of four NTCP models for head and neck cancer patients F. Emiro 1 , C. Bonfiglio 2,3 , M.G. Vincini 4 , S. Vignati 5 , S. Gandini 5 , M. Zaffaroni 2 , S. Volpe 2,6 , G. Marvaso 2,6 , D. Genovesi 3 , G.L. Gravina 7,8 , D. Alterio 2 , B.A. Jereczek-Fossa 2,6 1 European Institute of Oncology IRCCS, Unit of Medical Physics, Milan, Italy; 2 European Institute of Oncology IRCCS, Division of Radiation Oncology, Milan, Italy; 3 SS. Annunziata Hospital, “G. D’Annunzio” University of Chieti, Radiation Oncology Unit, Chieti, Italy; 4 European Institute of Oncology IRCCS, Division of Radiation Oncology , Milan, Italy; 5 European Institute of Oncology IRCCS, Department of Experimental Oncology, Milan, Italy; 6 University of Milan, Department of Oncology and Hemato-Oncology, Milan, Italy; 7 University of L'Aquila, Laboratory of Radiobiology, Department of Biotechnological and Applied Clinical Sciences, L'Aquila, Italy; 8 University of L'Aquila, Division of Radiation Oncology, Department of Biotechnological and Applied Clinical Sciences, L'Aquila, Italy Purpose or Objective We aimed to externally validate normal tissue complication probability (NTCP) models for radiation-related toxicity in head and neck (HN) cancer patients (pts) treated with curative radiotherapy (RT) Materials and Methods Retrospective analysis of pts treated from 2015 to 2021. Four NTCP models were evaluated: 1) physician-rated swallowing dysfunction at 6 months (m) (Christianen et al.); 2) tube feeding dependence at 6 m (Wopken et al.) 3) Incidence of G ≥ 2 laryngeal edema within 15 m from RT (Rancati et al.); 4) acute oral and oropharyngeal mucositis (OM) G ≥ 3 at any time during RT and OM mean G during RT treatment weeks ≥ 1.5 (Orlandi et al.). Pts already undergone oncological treatments on the HN district and/or with distant metastases and/or synchronous tumors or treated with altered fractionation RT schedule were excluded. Toxicities were retrieved from electronic medical charts. Validation analyses included: 1) Model performance using the Brier score 2) discrimination ability using the area under the receiver operating characteristic curve (AUC) 3) calibration using calibration intercept and slope 4) Hosmer–Lemeshow goodness-of-fit test to evaluate the calibration of the model. The model’s predictions fit the data at an acceptable level if the Hosmer– Lemeshow goodness of-fit test statistic is>0.05