ESTRO 2025 - Abstract Book

S3743

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

model and 919 patients for the 12-month model. Dysphagia prevalence was 19% at 6 months and 14% at 12 months, compared to a 28% prevalence in the Van den Bosch et al. cohort (2) at both time points. The closed testing procedure indicated that complete model revisions achieved the best fit in the validation cohort. For the 6-month dysphagia prediction (Figure 1), the original model's AUC of 0.72 (95%CI: 0.67-0.77) improved to 0.74 (0.70-0.79) after revision. Similarly, the 12-month model (Figure 2) showed an AUC improvement from 0.69 (0.63-0.74) to 0.74 (0.69-0.79).

Conclusion: External validation of the NTCP models suggested that complete revisions were needed to best predict dysphagia at 6 and 12 months in the validation cohort, possibly influenced by the differing dysphagia prevalence between cohorts. These findings contribute to understanding model transportability across different clinical settings and demonstrate the value of validation for local populations.