RESUMEN

Objective:To explore the dosimetric differences of different dose accumulation method for brachytherapy combined with external beam radiation therapy (EBRT) of cervical cancer and establish clinical prediction models for radiation-induced late rectal injury (RLRI) after radiotherapy.Methods:A retrospective analysis was conducted for the clinical data of patients who received radical concurrent chemoradiotherapy (CCRT) for cervical cancer in the Department of Oncology of the Affiliated Hospital of North Sichuan Medical College from January 1, 2020 to November 30, 2021. EBRT combined with brachytherapy was employed for the patients, and dose assessment was performed in two means: the direct accumulation using equivalent dose in 2-Gy fractions (EQD2) and deformable image registration (DIR)-based dose accumulation of 3D planning images. The toxicity criteria of the Radiation Therapy Oncology Group were adopted as the RLRI grading criteria. The prediction models of RLRI using both dose assessment method were constructed. The areas under the receiver operating characteristic (ROC) curves were calculated to assess the predictive accuracy of the different dose assessment method.Results:In the case of brachytherapy, the D95% and D90% EQD2 doses to high-risk clinical target volumes (HR-CTVs) were 2.18 and 2.92 Gy higher respectively and the D2 cm 3, D1 cm 3, and D0.1 cm 3 EQD2 doses to the rectal were 1.74, 2.28, and 2.26 Gy higher, respectively compared to DIR-based dose accumulation ( t = 3.82, 5.21, 4.58, 5.17, 2.05, P < 0.05). For EBRT combined with brachytherapy, the D2 cm 3, D1 cm 3, and D0.1 cm 3 EQD2 doses to the rectal were 6.22, 7.61, 9.56 Gy higher than DIR-based doses, respectively, and the dosimetric differences were statistically significant ( t = 9.40, 10.59, 7.87, P < 0.001). The joint prediction model yielded an area under the ROC curve of 0.788. The sensitivity and specificity of the optimal cut-off value were 0.850 and 0.660, respectively. Furthermore, the Hosmer-Lemeshow goodness-of-fit tests indicated high goodness-of-fit ( P > 0.05). The prediction model for DIR-based dose accumulation of traditional predictors yielded areas under the ROC curves for D2 cm 3 and D1 cm 3 to the rectal of 0.784 and 0.763, respectively. The sensitivities of the optimal cut-off values were 0.850 and 0.750, respectively, and the specificities were 0.679 and 0.717, respectively. Conclusions:There are dosimetric differences between the direct dose accumulation using EQD2 and DIR-based dose accumulation of 3D planning images for brachytherapy combined with EBRT. Both the joint prediction model and the DIR-based dose accumulation of D2 cm 3 and D1 cm 3 to the rectal are effective in predicting RLRI. Given the complex calculation of the joint prediction model, it is recommended that RLRI should be predicted through DIR-based dose accumulation of D2 cm 3 and D1 cm 3 to the rectal clinically.