A Model-Based Approach to Predict Short-Term Toxicity Benefits With Proton Therapy for Oropharyngeal Cancer.

Abstract

The aim of this study was to generate normal tissue complication probability (NTCP) models in patients treated with either proton beam therapy (PBT) or intensity-modulated radiation therapy (IMRT) for oropharynx cancer and to use a model-based approach to investigate the added value of PBT in preventing treatment complications. For patients with advanced-stage oropharynx cancer treated with curative intent (PBT, n = 30; IMRT, n = 175), NTCP models were developed using multivariable logistic regression analysis with backward selection. For PBT-treated patients, an equivalent IMRT plan was generated to serve as a reference to determine the benefit of PBT in terms of NTCP. The models were then applied to the PBT-treated patients to compare predicted and observed clinical outcomes (calibration-in-the-large). Five binary endpoints were analyzed at 6 months after treatment: dysphagia ≥ grade 2, dysphagia ≥ grade 3, xerostomia ≥ grade 2, salivary duct inflammation ≥ grade 2, and feeding tube dependence. Corresponding toxicity grading was based on National Cancer Institute Common Terminology Criteria for Adverse Events version 4. Paired t tests and Wilcoxon rank tests were used to compare mean NTCP results for endpoints between PBT and IMRT. NTCP models developed based on outcomes from all patients were applied to those receiving PBT. NTCP values were calculated for the equivalent IMRT plans for all PBT-treated patients, revealing significantly higher NTCP values with IMRT. PBT was associated with statistically significant reductions in the mean NTCP values for each endpoint at 6 months after treatment, with the largest absolute differences in rates of ≥grade 2 dysphagia and ≥grade 2 xerostomia. NTCP models predict significant improvements in the probability of short-term, treatment-related toxicity with PBT compared with IMRT for oropharyngeal cancer. This study demonstrates an NTCP model-based approach to compare predicted patient outcomes when randomized data are not available.