ESTRO 2023 - Abstract Book

S29

Saturday 13 May

ESTRO 2023

Results 150 pts were eligible for the analyses; pts’ and tumor’s characteristics are reported in Table 1. Of these, those available for NTCP analysis and incidence of corresponding toxicities were distributed as follows: 1) 97 pts eligible for dysphagia at 6 months (5% events); 2) 88 pts eligible for tube feeding dependence (3% events); 3) 102 pts eligible for G ≥ 2 laryngeal edema (21% events); 4) 113 pts eligible for mucositis G ≥ 3 (42% events); 5) 114 pts for mucositis G ≥ 1.5 (63% events). All AUC of NTCP models are reported in Fig.1. Considering all validation tests, all the models, except for Christianen model, show adequate fit. The best discrimination is reached by Wopken model (AUC = 0.73), but the limited number of cases influences the precision of the estimates, as can be seen by the wide confidence interval (CI 95%: 0.67-0.98). No model has evidence of good calibration. Non parametric tests concerning different scores show a significant difference between pts with events and other pts, only for Rancati model (p-value = 0.02 for laryngeal edema). The low number of events for NTCP models of dysphagia at 6 months and tube feeding dependence prevented statistically robust results Conclusion All the model (except for Orlandi OM mean G ≥ 1.5) suggest a good discrimination and a sufficient fit (except for Christianen). Calibration, considered as distance as observed and predicted probabilities, is low probably because of the low number of events. Further analyses are currently ongoing to confirm the assessment of the performance of both dysphagia and tube- feeding NTCP models. MO-0062 External validation of radiomics and deep learning models for recurrence-free survival prediction Y. Li 1 , B. Ma 1 , H. Chu 1 , J. Albertus Langendijk 1 , L. Vania van Dijk 1 , N. Maria Sijtsema 1 1 University Medical Center Groningen, Department of Radiation Oncology, Groningen, The Netherlands Purpose or Objective Prediction models can be used to identify patients with a high- and low-risk for recurrence-free survival (RFS) including local, regional recurrence and distant metastases. Studies to investigate the options for treatment intensification or de- intensification for high- and low risk patients could be performed to determine the optimal treatment for individual patients. Previously, a radiomics, a deep-learning (DL) and a hybrid model were developed for RFS prediction in oropharyngeal cancer (OPC) patients within HECKTOR challenge 2022. Good model performance was obtained in the train