Dose-Volume Histogram (DVH) Patterns within the Salivary Glands and Clinical Parameters Predict Xerostomia in Head and Neck Cancer (HNC) Patients, from Injury to Recovery

Abstract

NCI-CTCAE-graded xerostomia is routinely captured in our analytic database demonstrating that its maximum severity and prevalence occurs within 18 mos after RT (injury), with the prevalence decreasing (recovery) where a lower rate of decreasing prevalence can be observed thereafter. Studies have indicated that the spatial distribution of radiation to the salivary glands differentially affects the risk of xerostomia. We sought to characterize the spatial dosimetry to the salivary glands and the oral cavity as it related to the risk of maximum xerostomia injury and its recovery as defined above while controlling for clinical parameters. HNC patients treated from 2007 to 2017 in our institution with CTCAE graded xerostomia available for at least 18 mos of follow-up were evaluated. Patients with moderate to severe xerostomia (CTCAE grade ≥ 2) before RT were excluded. 22 spatial zones were geometrically created from the parotid glands (PG), submandibular glands (SMG) and oral cavity (OC) for each patient. DVH features (D10-D90) for each zone as well as clinical parameters were used to predict xerostomia injury (CTCAE grade ≥ 2) within 18 mos post-RT, and recovery (grade decrease to < 2) after 18 mos of follow-up. A total of 217 HNC patients were identified with 146 developing moderate to severe xerostomia (CTCAE grade ≥ 2) within 18 mos of follow-up. The dataset was randomly split into a training and test dataset at the ratio of 7:3 for each outcome. Ridge logistic regression with 10-fold cross-validation was applied to predict xerostomia injury and recovery respectively in the training datasets. The critical zones associated xerostomia injury was the medial inferior portion of the ipsilateral PG (relative to the primary tumor) and the superior portion of the ipsilateral PG for recovery. The dose – recovery pattern was significantly influenced by the low - dose bath (D80-D90) across the superior portion of ipsilateral PG. The area under the receiver operating characteristic curve (AUC) was 0.78 and 0.74 for xerostomia injury and recovery respectively in validation datasets. The AUC for injury and recovery are 0.72 and 0.70 respectively in test datasets. Our data science methodology demonstrated that different spatial-dose patterns for xerostomia injury vs. recovery and highlighted the strength of this analytic technique along with the importance of assessing xerostomia during follow-up. These observations if validated provide insights into new strategies for RT deintensification. This work was supported by the Radiation Oncology Institute.