Subregional Parotid Dose Predictors of Xerostomia In Head and Neck Cancer
Abstract
Purpose
Radiation induced xerostomia is a common adverse sequela among head and neck cancer (HNC) patients with ~45,000 new cases treated with radiation in the US annually. Current guidelines focus on sparing the mean dose to the parotid glands to reduce xerostomia. This study aimed to determine if other dosimetric predictors provide improved xerostomia prediction.
Methods
Xerostomia outcomes and dose distributions were analyzed for 102 HNC patients enrolled in a double-blind parotid stem cell rich region (SCR) sparing study at University Medical Center Groningen. Xerostomia was assessed using CTCAE and Groningen Radiation Induced Xerostomia (GRIX) questionnaires at acute (≤3 months), late (3–12 months), and recovery (24 months) timepoints. For volumetric analysis, ipsilateral and contralateral parotids were subdivided into 10 anatomical subregions, and DVH features (D5%, D25%, D50%, D75%, D90%) were extracted. For voxel-based analysis, parotids were deformably registered to a common reference patient and uniformly sampled. Voxel doses were used as predictors. Logistic regression with ridge regularization was performed using 5-fold cross-validation for each endpoint and timepoint. Model performance was evaluated by AUC, and predictor significance was assessed using permutation testing (p≤0.05).
Results
For CTCAE-defined xerostomia, volumetric models achieved AUCs of 0.63 (acute), 0.66 (late), and 0.54 (recovery), while voxel-based models achieved AUCs of 0.65, 0.65, and 0.57, respectively. GRIX-based recovery models demonstrated the strongest performance (AUC=0.69-0.75). Across volumetric and voxel-based analyses, the superior region of the contralateral parotid consistently showed the highest predictive importance and statistically significant dose differences, corresponding to low-dose bath areas and overlapping the SCR.
Conclusion
Dose to the superior contralateral parotid, overlapping the SCR, was the strongest predictor of xerostomia incidence and recovery. Because this cohort originated from a parotid SCR-sparing trial, whole-parotid mean dose was decoupled from SCR dose demonstrating the importance of subregional and low-dose effects, supporting these strategies in treatment planning.