Abstract
BACKGROUND AND PURPOSE: This multicentre, retrospective study aimed to develop a predictive model for radiation-induced dysphagia in head and neck cancer patients, focusing on the role of gross tumour volume (GTV) to high dose CTV (CTV1) margin size and dose-related factors. Unlike previous studies focused on peak or single time-point dysphagia, this study modelled symptom trajectories using repeated follow-up data for a more complete picture. Patient/material and methods: Between 2010 and 2015, 1,948 patients with pharyngeal or laryngeal squamous cell carcinoma received definitive intensity-modulated radiotherapy (IMRT) at three Danish centres. Data included physician-rated dysphagia (grade 0-4), tumour and treatment characteristics, and AI-based segmentations of organs at risk (OARs). Predictors included GTV-CTV1 margin size, mean doses to the oral cavity and pharyngeal constrictor muscles (PCM), GTV volume, chemotherapy, tumour site, fractionation, nimorazole, sex, smoking status, baseline dysphagia, and age. A logistic ordinal mixed-effects model was fitted with patient ID as random effect. Data were split into training (70%) and test (30%) sets. Model performance was assessed using calibration plots and area under the curve (AUC). RESULTS: After excluding incomplete cases, 1,685 patients (7,829 visits) were analysed. GTV-CTV1 margin size was not significantly associated with dysphagia, although larger margins correlated with higher OAR doses. Higher doses to the lower PCM (odds ratio [OR] = 1.30 per 5 Gy) and oral cavity (OR = 1.32 per 5 Gy) increased risk. The model demonstrated good calibration and robust discrimination (AUC = 0.77-0.84). INTERPRETATION: Radiation dose to the oral cavity and lower PCM were the strongest modifiable predictors of dysphagia risk. Margin size was not independently associated, possibly due to confounding by clinical judgement.