357 Background: Radiomic characterisation of tumour phenotypes can generate image-driven biomarkers that potentially aid in clinical decision-making. We sought to identify radiomic features in metastatic GEA that may be predictive for survival outcomes. Methods: A retrospective analysis between 2009-20 identified patients (pts) with metastatic GEA. All pts received chemotherapy (CTx), with a ‘baseline’ and 8-12 week ‘on-treatment’ contrast-enhanced CT chest/abdomen/pelvis performed. Radiomic analysis was performed with LIFEx (livexsoft.org). Population demographics and clinical outcomes were recorded. Univariable Cox proportional hazards model (UVA) assessed clinical variables (n=26) predictive of overall survival (OS) and progression-free survival (PFS) with p=0.05 indicating significance. Multivariable Cox model (MVA) was used to assess radiomic features (n=78) in the presence of clinical variables. Concordance index (C-index) was calculated to assess model performance (≥0.7 = high predictive accuracy). A ‘validation’ cohort analysis was performed to validate the model. Results: 166 pts were identified (primary cohort n=143; validation cohort n=23). 123 had de-novo metastatic disease, 43 recurrence following curative-intent therapy. In the primary cohort the median age was 58.1y, 101 (71%) were male, 120 (84%) were non-Asian and 131 (92%) were ECOG 0-1. Similar demographics were observed in the validation cohort. Both ‘baseline’ and ‘on-treatment’ scans UVA identified Her2 status, ethnicity, and the number of CTx cycles as predictive of PFS, while ECOG, brain metastases, neutrophil count (ANC), albumin and number of CTx cycles were predictive of OS. ‘Baseline’ model analysis for PFS and OS identified consistent radiomic features (HUskewness; HUpeakSphere), with an observed C-index 0.6 and 0.657 respectively. No radiomic features were identified on ‘on-treatment’ PFS analysis. ‘On-treatment’ OS analysis is shown in the table with 3 radiomic features (SHAPE Surface; SHAPE Compacity; PARAMS ZSpatial-Resampling) predictive for OS. The C-index is 0.76. Analysis of the validation cohort supported the model (C-index 0.815) for ‘on-treatment’ OS. Conclusions: Radiomic analysis identified a number of features associated with PFS and OS. The features specifically identified on ‘on-treatment’ scans were highly predictive for OS. Our analysis suggests radiomic features in addition to clinical variables can be predictive of outcome in patients with metastatic GEA receiving CTx.[Table: see text]