Transforming the management of radiotherapy-induced hypothyroidism in nasopharyngeal carcinoma through an Innovative individualized radiation dosage model: A multicenter retrospective analysis

Purpose

Current guidelines for thyroid radiation dose prescription lack uniformity and fail to consider the unique characteristics of individual patients. This study aimed to develop an individualized thyroid dosing regimen to enhance thyroid protection during radiotherapy.

Methods and Materials

In this study, we enrolled 621 patients with nasopharyngeal carcinoma (NPC) across four distinct cancer centers, stratifying the data into a training cohort and two external validation cohorts. The specific clinical characteristic-matched tolerated dose values were fitted using binary logistic regression and time-to-event Cox methods in the training cohort. The TSH-volume index (TVI), calculated as thyroid-stimulating hormone (TSH) level divided by thyroid volume (TV), was introduced as a novel parameter. A radiation-induced hypothyroidism (RIHT) parameter was developed using the volume of thyroid spared at the tolerated dose (VStd) and compared with classical normal tissue complication probability (NTCP) and machine learning models using the area under the curve (AUC) and concordance index (C-index).

Results

The follow-up periods spanned 28 (range, 1–66), 33.5 (range, 3–82), and 17 months (range, 2–56), respectively, across these cohorts. RIHT was observed in 27.7 % and 35.3 % of patients at 2 and 3 years in the training cohort, respectively; 30.0 % and 41.5 % in the external validation cohort 1; and 27.2 % and 38.0 % in the external validation cohort 2. Univariable analysis identifies sex, equivalent uniform dose (EUD), TV, TSH, and the TVI as predictors of RIHT, while multivariable analysis confirms EUD and TVI as independent prognostic factors. The TVI-based VStd parameter outperformed the VS40, VS45, and VS50 indices (representing volumes spared at 40 Gy, 45 Gy, and 50 Gy, respectively), classical NTCP models, and even machine learning models in predictive performance. To enhance clinical applicability, we have developed a thyroid dose prescription table based on TVI.

Conclusions

We developed a high-accuracy model for individualized thyroid dosing in NPC radiotherapy. The model, supported by a clinically relevant table, offers a customized approach to thyroid protection, enhancing both predictive accuracy and clinical utility.