Radiation-induced Hypothyroidism in Survivors of Head-and-Neck and Breast Cancers After 3-Dimensional Radiation Therapy: Dose-Response Models and Clinical-Dosimetric Predictors

The prediction of normal tissue complications in treatment planning plays a critical role in radiation therapy of cancer. Objectives: The aim of the current study was to evaluate mathematical models and clinical-dosimetric variables for prediction of radiation-induced hypothyroidism (RHT) in patients with head-and-neck cancer (HNC) and breast cancer (BC). Methods: Clinical and dose-volume data from 62 patients treated with three-dimensional conformal radiation therapy were prospectively analyzed in terms of HNCs and BC. Thyroid function assessment was monitored by the level of thyroid hormones from patients’ serum samples. Cox semi-parametric regression models were used to predict the risk of RHT. Model performance and model ranking were evaluated in accordance with the area under the receiver operating characteristic curve (AUC) and Akaike’s information criterion (AIC), respectively. Results: Out of 62 patients, 17 persons developed RHT at a median follow-up of 11.4 months after radiation therapy. Thyroid volumes above the cut-off points of 14.2 cc and 11.4 cc showed a decrease in RHT risk for patients with HNC and BC, respectively. Moreover, the thyroid mean dose above the cut-off points of 53 and 27 Gy increased the risk of RHT for patients with HNC and BC, respectively. Simple and Multiple Cox regression analyses of the complete dataset revealed that thyroid volume and thyroid mean dose were the strongest predictors of RHT. According to AUC, Boomsma’s model, and the generalized equivalent-uniform-dose (EUD) model in the HNC dataset outperformed the BC dataset. Conclusions: The probability of RHT rises with an increase in the mean dose to the thyroid gland; however, it decreases with increasing thyroid gland volume. Regarding the AUC analysis, gEUD model showed an acceptable predictive performance; however, the logistic Boomsma’s model wassomehowmore effective in predicting RHT on theHNCdataset. Cella’s model revealed a relatively acceptable prediction of RHT on the BC dataset.