ESTRO 2025 - Abstract Book

S3777

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Material/Methods: Two DAGs were elicited based on discussions in two independent expert panels: one comprising UK-based experts and the other US-based experts. The panels included clinical oncologists, speech and language therapists, physicists, surgeons (US-based panel), a causal inference expert and a computer scientist. Only two members sat on both panels: the causal inference expert who provided methodological guidance and the computer scientist who led the discussion and operated DAGitty [2] . Minimal adjustment sets were identified using the backdoor criterion. The average causal effect of mean dose to the medulla oblongata and dysphagia was estimated for three validated measures: i ) M.D. Anderson Dysphagia Inventory (MDADI) score, ii ) performance status scale for normalcy of diet (PSS-HN), iii ) water swallow test (WST), using multivariable generalised linear model for WST and multivariable linear model for MDADI and PSS-HN. Dysphagia was assessed pre-treatment and 1 year after HN-RT for 104 sequential patients treated with definitive (chemo)RT in the UK. All dosimetric estimates were recorded as equivalent dose in 2Gy/fraction (EQD2Gy) with α/β=3Gy. Results: The elicited DAGs present varying degrees of granularity and complexity, Figure 1, with 28/53 variables and 62/126 arcs for the UK-based [3] /US-based [4] expert panels, respectively.