ESTRO 36 Abstract Book

S859 ESTRO 36 _______________________________________________________________________________________________

assessed according to CTCAE v4.0 at baseline and weekly during RT. Four endpoints were considered: mean grade (G) ≥1.5 and G≥3 oral mucositis (OM), G3 dysphagia and G≥2 salivary dysfunction (SD). The selected OARs were: oral cavity (OC) and parotid glands (PG) considered as a single organ for OM and SD; OC, pharyngeal constrictor muscles (PCM), supraglottic and glottic larynx (GL) for dysphagia. DVHs were reduced to Equivalent Uniform Dose (EUD): for each OAR the best volume parameter n was determined through numerical optimization. When this procedure did not converge, we chose to evaluate DVH cutpoint through t-test. EUD was inserted into a multivariable logistic (ML) model together with clinical/treatment features; variables selection was guided by LASSO. Goodness of fit was evaluated with Hosmer-Lemeshow test and calibration plot. Results Data were collected for 132 pts. MeanG≥1.5 and G≥3 OM were reported in 40 pts (30%), G3 dysphagia in 50 (38%) and G≥2 SD in 90 (68%). ML models (figures 1-2) consisted in: a single variable for meanG≥1.5 OM, i.e. OC EUD with n=1 (mean dose) (OR=1.07); 3 variables for G≥3 OM including OC EUD with n=0.05 (OR=1.02), PG EUD with n=1 (OR=1.06), BMI≥30 (OR=3.8, obese pts); 3 variables for dysphagia including PCM V50Gy (OR=1.02), GL and OC EUD with n=0.35 and 0.15 respectively (OR=1.02 and 1.04); 4 variables for SD including PG D98% (OR=1.04), OC EUD with n=0.05 (OR=1.11), age (OR=1.08, 5-year intervals), smoke (OR=1.37, yes vs no). Calibration was good in all cases.

Conclusion OC resulted to be a parallel organ for mean G≥1.5 OM, while severe OM was associated to a synergic effect between PG mean dose and high doses received by small OC volumes, with BMI acting as a dose-modifying factor. On the other hand, OC resulted as a fairly serial organ for G≥2 SD; this could be due to the inclusion of minor salivary glands in OC contour, that the target often overlaps. Older age and smoking history were also associated to a SD increased risk. We did not find a strictly significant association between G3 dysphagia and dosimetric features, but the step trend resulting from our model calibration suggests that the ML model may not describe well the dose-response relationship in this case, with a step function possibly being more suitable. Validation of these models in a larger pts cohort is ongoing.