Approach Combining Dosimetry And Biology To Predict Severe Toxicity Of Radiotherapy For Head And Neck Squamous Cell Carcinomas

Radiotherapy (RT) used alone or added to surgery in head and neck squamous cell carcinomas (HNSCC) leads to severe acute toxicity in up 30-40% patients (pts). Recent studies have stressed the role of the nucleoshuttling of the phosphorylated form of the ATM protein (pATM) in the response to radiation, and the discrimination power of this molecular endpoint in predicting individual radiosensitivity. Furthermore, a predictive assay based on the quantification of the pATM by ELISA method was developed on fibroblasts. In order to simplify the existing assay, and make it closer to a possible use in the clinics, we assessed the reliability of pATM quantification on lymphocytes to predict the pts’ risk of acute toxicity. We here propose and evaluate a model combining dosimetry and the lymphocyte assay. A retroprospective study was performed on HNSCC pts treated with IMRT/VMAT/3DCRT and IMRT system. Acute toxicity was assessed with CTCAE and two endpoints were selected: grade≥3 mucositis (OM3) and grade≥3 dysphagia (DY3). Global quantity of pATM molecules was assessed by ELISA on lymphocytes. Two previously developed dosimetric models were considered: including dose to oral cavity (Equivalent Uniform Dose, EUD, with n=0.05, OR=1.02 for 1 Gy increase) and mean dose to parotid glands (OR=1.06 for 1 Gy increase) for OM3 and including EUD to the oral cavity (n=0.15, OR=1.04 for 1 Gy increase), EUD to larynx (n=0.35, OR=1.02 for 1 Gy increase) and the volume of pharyngeal constrictor muscles receiving more than 50Gy (V50Gy, OR=1.02, for 1% V50Gy increase) for DY3. The combined model was defined as a logistic model including prediction from the previously published dosimetric model (without any adjustment, continuous variable with possible values in the 0-100 range) and classification by the pATM assay (radiosensitive vs not-radiosensitive, without any adjustment, dichotomous yes/no variable). Discrimination of models was evaluated through area under the ROC curve (AUC). Thirty-five pts were available for evaluation of OM3 (8/35 events, 23.8%) and 33 for DY3 (12/33 events, 36.4%). Detailed results are presented in Figures 1 and 2. When OM3 was considered, the dosimetric model resulted in AUC=0.61, while the pATM assay had AUC=0.68. The combined model resulted with OR=1.06 for the dosimetric model and OR=7.1 for the pATM assay, AUC=0.79. For DY3, the dosimetric model and the pATM assay resulted in AUC=0.56 and AUC=0.77, respectively. The combined model had OR=1.01 for the dosimetric model and OR=27 for pATM assay, with AUC=0.8. A better prediction of occurrence of radio-induced toxicities can help treatment optimization and personalization, minimize treatment interruptions and improve management of side effects. Results from this small validation study suggest that molecular assays may improve the performance of dosimetric predictive models, leading to combined dosimetric-biological tools which might improve management of HNSCC pts during RT.