Evaluation of radiobiological models for estimation of radiation-induced hypothyroidism after radiotherapy in patients with head-and-neck and breast cancers

Abstract

The effectiveness of four normal tissue complication probability models for prediction of radiation-induced hypothyroidism (RHT) in patients with head-and-neck cancers (HNCs) and breast cancer (BC) was evaluated. The dose-response relationship of the thyroid for RHT in these patients was determined, according to the four mathematical models, and the best-fit parameters of the models were found. Methods and materials: Clinical and dose-volume data of 62 patients treated with three-dimensional conformal radiation therapy for HNCs and BC were prospectively analyzed. Thyroid function assessment was evaluated by the level of thyroid hormones from patient serum sample. Thyroid DVHs were converted to 2 Gy/fraction equivalent doses using the linear-quadratic formula with α/β=3 Gy. The evaluated models included the following: LKB (LEUD); Logit-EUD (gEUD); mean dose; and relative seriality models. Cox semi-parametric regression models were used to predict the risk of RHT. Model performance and model ranking was evaluated through the area under the receiver operating characteristic curve (AUC) and Akaike's information criterion (AIC), respectively. The parameters of the models were obtained by fitting the patients’ data using a maximum likelihood estimation method. The goodness of fit of the models was determined by the Chi-Square test. Results: 20 of 62 patients developed RHT at a median follow-up of 11.92 months after radiotherapy. Kaplan-Meier curves showed a decrease in RHT with thyroid mean dose less than 30 Gy for the whole dataset. Simple and multiple analysis for the whole dataset revealed that RHT risk was higher for smaller thyroid volumes. For BC patients, V20, V30 and V40 was found to be influencing parameters in prediction of RHT. Based on the AUC, the generalized equivalent-uniform-dose (EUD) model performed better on the whole dataset. The ROC comparison showed no significant difference in the prediction capability of the 4 models for the whole dataset. gEUD and RS models were ranked as the best models based on the AIC values, respectively. None of the models was rejected according to the evaluation of the goodness of fit. The D50 estimated from the models was approximately 44.13 Gy.