3638 - Individualized Thyroid Prescriptions Based on a Novel Tolerance Dose Model for Radiation-Induced Hypothyroidism in Nasopharyngeal Carcinoma Patients

J. Ding¹, Y. Lin¹, Z. R. Li², J. Hong¹, C. Huang¹, Q. Zhou³, Z. Fei¹, and C. Chen¹; ¹Department of Radiation Oncology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, Fujian, China, ²Department of Research Algorithms, Manteia Technologies Co., Ltd, Xiamen, Fujian, China, ³Manteia Technologies Co.,Ltd, Xiamen, Fujian, China

Purpose/Objective(s): Researches into normal tissue complication probability (NTCP) models for Radiation-Induced Hypothyroidism (RIHT) have almost harnessed intricate algorithms that, while precise, are challenging to apply in clinical practice. We developed a new model that proposes a unique thyroid tolerance dose prescription to accurately forecast RIHT and simplifies clinical protocols for safeguarding thyroid tissue in nasopharyngeal carcinoma. Materials/

Methods: Our study collected clinical data and dose distribution imaging from 336 nasopharyngeal carcinoma patients. The radiation doses were converted into the Equivalent Dose in 2 Gray fractions (EQD2) and the Dose-Volume Histogram (DVH), Dose-Remaining Volume Histogram (DRVH), and Dose-Remaining Absolute Volume Histogram (DRAVH) were calculated. Generated the DRVH and DRAVH for dose d ranging from 1 to 80 Gy, in 1 Gy increments, resulting in 79 sets of DRVH(d) and DRAVH(d). Patients were stratified into two groups based on individualized clinical features using a threshold value X, and univariate analyses were performed using DRVH and DRAVH as variables within each group. The doses d were ranked according to univariate analysis indicators, with the top 15 doses designated as the tolerance dose distribution. The average of these tolerance dose distributions was selected for each groups tolerance dose. These individualized clinical features and their corresponding tolerance doses were fitted using least squares regression, leading to a tolerance dose model. This model was then compared with classical NTCP and machine learning models based on clinical and dosimetric factors. A recommending customized prescription for thyroid irradiation was proposed.
Results: Among the 336 patients enrolled in our study, 36.9% developed RIHT. We noted significant differences in the distribution of variables such as gender, Thyroid Volume(TV), Equivalent Uniform Dose (EUD), Thyroid Stimulating Hormone (TSH), and TVI (TVI=TSH/TV) between RIHT and non-RIHT patients. Due to solid multicollinearity, gender was excluded from further study after correlation analysis. TV and TVI had stronger linear trends with tolerance doses than TSH, thus we included them as distinct variables in our tolerance dosage model. This model demonstrated superior predictive performance (TVI: AUC 0.814, C-index 0.738; TV: AUC 0.712, C-index 0.639) compared to classical NTCP models (Lyman: AUC 0.667, C-index 0.401; Ronjom: AUC 0.718, C-index 0.347; Boomsma: AUC 0.708, C-index 0.347; Cella: AUC 0.668, C-index 0.399) and machine learning algorithms (highest AUC 0.812, C-index 0.738). Our model yielded equal or superior predictive accuracy with more straightforward calculations. We also developed recommending thyroid dose prescription tables based on TV and TVI, respectively.
Conclusion: We established a high-accuracy thyroid tolerance dose model and provided clinically applicable tables offering customized dosing prescriptions for thyroid irradiation.