ESTRO 2024 - Abstract Book

S1183

Clinical - Head & neck

ESTRO 2024

234

Digital Poster

External validation of a severe dysphagia prediction model using real world data

Vinod Patel 1 , Laia Humbert-Vidan 2 , Francesca De Felice 3 , Teresa Guerrero-Urbano 4

1 Guy's & St Thomas' Hospital, Oral Surgery Dept, London, United Kingdom. 2 King's College London, Oncology, London, United Kingdom. 3 Sapienza Universita De Roma, Oncology, Roma, Italy. 4 Guy's & St Thomas' Hospital, Oncology, London, United Kingdom

Purpose/Objective:

Acute severe dysphagia (ASD) is a problematic treatment complication peri- and immediately post-head and neck (HN) (chemo)radiotherapy. Its presence can impact the timely delivery of radiotherapy, nutritional intake, and quality of life. Determining which patients will suffer from the complication has been challenging to predict. To improve prediction, normal tissue complication probability (NTCP) models have been developed to help identify high risk patients.1 This study aims to test the validity of an ASD NTCP model developed on clinical trial data on a real-world population dataset.2

Material/Methods:

Using a retrospective approach (2018-2021) clinical records were reviewed to identify a study population of patients that matched the eligibility criteria for the model being tested. Patient, tumour, and clinical data was retrospectively collected from medical records including the presence or absence of ASD (> grade 3)3 in the peri-, immediate and up to 3-months post-radiotherapy phase. Furthermore, the extended pharynx was volumed to retrieve radiation dose to the structure.2 Clinical variables (gender, chemotherapy regime, chemotherapy drug, tumour site, age) and radiation dose-volume histogram data for the pharyngeal mucosa was then input into the dysphagia model for analysis via functional principal component-logistic regression (FPCLR) as per the original model.2

Results:

A total of 184 eligible patients defined the study population of which 98 (53%) patients were identified to have had ASD. Oropharyngeal cancer patients (n=134/184, 73%) and chemo-radiotherapy (n=153/184, 83%) were most populous. Tumour (T) size >3 (p=0.011) and those treated via chemo-radiotherapy (p=0.018) were found to be significant clinical factor for the development of ASD. ASD prediction versus observed outcome via the model showed the performance and discriminative ability to be sub-optimal based upon the Brier score (0.337) and the ROC AUC (0.516).

Conclusion: