ESTRO 2025 - Abstract Book

S1031

Clinical – Head & neck

ESTRO 2025

Inventory (MDADI), Performance Status Scale for Normalcy of Diet (PSS-HN), and Water Swallowing Test (WST). Four Bayesian models were fit using Markov Chain Monte Carlo with two chains and 2000 iterations for each outcome. The first model used non-informative priors with risk factors selected via best subset selection for linear regression and stepwise selection for logistic regression models. The subsequent three models used the posterior results as informative priors for risk factors and included mean doses to MO, PCM, and both, respectively, each with an uninformative prior. Model performance was assessed using R-squared (R²), Widely Applicable Information Criterion (WAIC), and leave-one-out cross-validation (loo-CV). Estimated marginal means (EMMs) compared effects of mean doses to both regions on predicted dysphagia. Results: Baseline dysphagia measures, comorbidity scores, smoking history, and tube feeding status were selected in all models, with other risk factors varying according to outcome, Table 1. Including MO mean dose in the model yielded the lowest WAIC value, suggesting superior fit compared to the clinical model alone. R² demonstrated a 5% improvement in predictive accuracy for models that included either the MO mean dose alone, or both MO and PCM mean doses, compared to models without dose variables for these regions, Table 1. EMMs analysis revealed that the model including MO mean dose offered the strongest predictive performance, Figure 1.