ESTRO 2020 Abstract book

S327 ESTRO 2020

Conclusion We observed only modest decrease of discriminatory performance from training to validation for DM and Death NED, indicating a robust set of predictors and modeling. LRF risk prediction performance dropped, with primarily loss in intermediate/high risk group separation. This should be investigated if the ambition is to use the model to select patients for trials of treatment intensification. An apparent increased risk of death NED compared to predicted is concerning if associated with increasingly intense therapy, but the causes of this observation need to be further investigated. 1 Håkansson et al . “A failure-type specific risk prediction tool for selection of head-and-neck cancer patients for experimental treatments”. Oral Oncology 74 (2017) 77-82. PD-0545 Validation of a predictive model for salivary dysfunction during chemo-IMRT for head-neck cancer A. Cavallo 1 , T. Rancati 2 , N.A. Iacovelli 3 , N. Facchinetti 3 , S. Alfieri 4 , S. Cavalieri 4 , T. Giandini 1 , A. Cicchetti 2 , R. Ingargiola 3 , D.A. Romanello 3 , S. Di Biaso 1 , M. Sabetti 1 , C. Fallai 3 , L. Licitra 4,5 , L. Locati 4 , E. Pignoli 1 , R. Valdagni 2,5,6 , E. Orlandi 3,6 1 Fondazione IRCCS Istituto Nazionale dei Tumori, Medical Physics, Milan, Italy ; 2 Fondazione IRCCS Istituto Nazionale dei Tumori, Prostate Cancer Program, Milan, Italy ; 3 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiation Oncology 2, Milan, Italy ; 4 Fondazione IRCCS Istituto Nazionale dei Tumori, Head and Neck Medical Oncology Unit, Milan, Italy ; 5 University of Milan, Department of Oncology and Hemato-oncology, Milan, Italy ; 6 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiation Oncology 1, Milan, Italy Purpose or Objective Radiation-induced xerostomia is one of the most prevalent adverse effects of head&neck cancer (HNC) treatment and could seriously affect and impair several domains of patients’ quality of life. It results primarily from a damage to the salivary glands, but its onset and severity may also be influenced by other patient-, tumor-, and treatment- related factors. We aimed to validate a previously developed predictive model for acute salivary dysfunction (aSD) for loco- regionally advanced nasopharyngeal carcinoma (NPC) patients (pts). Material and Methods For model development, a cohort of consecutive NPC pts treated curatively with IMRT at 70 Gy (2-2.12Gy/fr) and chemotherapy was retrospectively considered: 132 pts, grade≥2 (G≥2) aSD was reported in 90 pts (68.2%). Parotid glands (cPG, considered as a single organ) and oral cavity (OC) were selected as organs-at-risk. aSD was assessed according to CTCAE v4.0 at baseline and weekly during RT. DVHs were reduced to Equivalent Uniform Dose (EUD). Dosimetric and clinical/treatment features selected via LASSO were inserted into a multivariable logistic model, resulting in a 4-variables model, including dose to 98% of cPG (cPG_D98%) with OR=1.04, OC EUD (calculated with n=0.05) with OR=1.11, age (OR=1.08, 5-year interval) and smoking history (OR=1.37, yes vs. no). Validation was performed on two cohorts of pts treated at the same department and with prospective acute toxicity scoring using the same schedule and scale which was used for the population used to train the model. A first cohort (NPC_V) consisted of NPC pts (as in the model training phase), the second cohort was a mixed population of non- NPC HNC pts (HNC_V). Results

NPC_V cohort included 38 pts with G≥2 aSD scored in 34 pts (89.5%). HNC_V cohort included 93 pts, 77 with G≥2 aSD (92.8%). As a general observation, the incidence of G≥2 aSD was significantly different in the training and validation populations (p=0.01), thus impairing calibration-in-the- large, while effect size for the two dosimetric factors was confirmed: cPG_D98% OR=1.07 (95%CI: 1.001 to 1.15) and OC EUD OR=1.06 (95%CI: 1.001 to 1.13) (Figure 2). The model confirmed good discrimination in NPC_V cohort, with AUC=0.73 (p=0.02) and with negative predictive value (NPV) and positive predictive value (PPV) of 0.65 and 0.75 (cutoff at 65% predicted toxicity probability), respectively. Discrimination was also satisfactory in HNC_V cohort, AUC=0.68 (p=0.035) and with NPV and PPV of 0.52 and 0.69 (cutoff at 65% predicted toxicity probability), respectively (Figure 1).