Dosimetry Predictors of Late Skin Reactions after Whole Breast Radiotherapy on a Large Mono-Institutional Cohort of Patients

Abstract

To develop an NTCP model predicting late skin toxicity using dosimetric parameters from the breast dermis to identify possible RT constraints on such a structure. The skin structure was defined as the 5 mm inner isotropic expansion from the outer CT body contour. It was retrospectively segmented on a large mono-Institutional cohort of early-stage breast cancer patients enrolled between 2009 and 2017 (n = 1066). Patients were treated with tangential-field RT, delivering 40 Gy in 15 fractions without a RT boost. Toxicity was reported during FU using SOMA/LENT scoring. The study endpoint was moderate-severe late toxicity consisting of Fibrosis-Atrophy-Telangiectasia-Pain (FATP G2+) developed within 42 months after RT completion. Automatic delineation of skin and DVH extraction were accomplished by scripting using the MIM_assistant software. Also, the impact of changes in the dose calculation algorithms during enrolment time was quantified. A logistic model was created by combining multifactorial variables, considering both clinical factors and the absolute skin DVH (cc). Variance Inflation Factor (VIF) was performed to reduce the multicollinearity. Repeated 5-fold cross-validation with SMOTE approach to overcome the class unbalance was applied for model feature selection. The predictive model was then developed on the entire population due to the limited G2+ events. The FATP G2+ rate was 3.8% with 40/1066 experiencing late toxicity. Among them, a 40% had already developed acute symptoms after RT completion showing a consequential effect. The multicollinearity analysis selected 27 clinical-treatment-dosimetric factors. After repeated (20 times) 5-fold cross-validation, the best-performing model included Post-Surgery Cosmetic alterations, Aromatase Inhibitors (as a protective factor), V20 Gy (50% of the prescribed dose - DVH plateau region) and V42 Gy (105% of the prescribed dose - DVH high-dose tail). Accuracy and f1-score were 0.76 and 0.58 in both training and test sets, providing good reliability for selected variables. AUC for the final model on the entire population was 0.76+/-0.04. We quantified the association between fibrosis and skin DVH when delivering 40 Gy in 15fr. The model suggested an independent role of V20 and V42 Gy and a heavy risk modulation by surgical effects and aromatase inhibitors. This last factor could interfere with adipose tissue and water-content distribution within the breast. For this purpose, a CT-based densitometry characterization of toxicity patients is ongoing.