Predicting late radiation-induced xerostomia with parotid gland PET biomarkers and dose metrics.

PURPOSE: To assess associations between parotid gland PET biomarkers and late radiation-induced xerostomia, and to validate improvement of xerostomia predictive models by adding pre-treatment PET features to models based on dose and pre-treatment xerostomia. MATERIALS AND METHODS: Intensity PET features from 47 patients treated on institutional prospective clinical trials for HPV-associated oropharyngeal squamous cell carcinoma with uniform chemoRT were analyzed. Associations between 90th percentile of the parotid gland standardized uptake values (P90) from pre-treatment and post-treatment PET scans, mean parotid gland doses, and late xerostomia defined by the Xerostomia Questionnaire (XQ) and salivary flow rates were quantified. Multivariable analysis was applied for dose and PET features using penalized logistic regression for feature selection and generation of predictive models using the LASSO technique, and optimism bias was estimated by bootstrap resampling. RESULTS: Significant associations between late xerostomia and both mean parotid gland dose and P90 were demonstrated, and were generally stronger for post-treatment PET scans. The addition of P90 from pre-treatment PET scans improved the prediction model for late moderate or severe xerostomia compared to the base model, from AUC = 0.74 to 0.78 (p-value <0.001) for XQ summary score and from 0.77 to 0.84 (p-value <0.001) for the single eating-related XQ item with the largest inter-patient variability; however, only the latter remained significant on cross validation (AUC = 0.69 to 0.70 and 0.73 to 0.80, respectively). CONCLUSIONS: The addition of pre-treatment parotid gland PET biomarkers improved a predictive model for late patient-reported xerostomia over dose and pre-treatment xerostomia.