Development and validation of a prognostic model for advanced gastric and oesophageal carcinoma (AGOC) using individual patient data from two clinical trials.

499 Background: Estimating survival in advanced gastric and oesophageal carcinoma (AGOC) remains a significant challenge. We developed a prognostic model incorporating readily accessible patient and clinical data and specific patient-reported outcomes (PROs), utilizing individual participant data from two randomized trials. Methods: We used data from 2 trials comparing regorafenib to placebo; AGITG INTEGRATE IIa (n=251) for model development and AGITG INTEGRATE (n=152) for validation. Both trials enrolled metastatic or AGOC following failure of 1 or more prior lines of systemic therapy. Candidate variables were chosen from systematic literature review and expert consultation. Significant prognostic factors for inclusion in the multivariable model were identified using univariable Cox proportional hazards models with p-value threshold of 0.1. Multivariable models were developed via Lasso regression. An initial model was created using clinical variables alone, followed by the integration of PROs. C statistics were used to discriminate the model’s efficiency. Results: Univariable analysis identified 9 clinical variables and 4 PRO domains as significant, including body mass index (BMI)(p=0.08), ECOG PS (0.02), extent of cancer (p<0.001), liver involvement (p=0.04), treatment with regorafenib (p=0.005), neutrophil-lymphocyte ratio (NLR) (p<0.001), LDH (p<0.001), albumin (p<0.001), and CA 19-9 (p=0.007). Significant baseline PROs included appetite loss (p<0.001), constipation (p<0.001), fatigue (p<0.001), and pain (p<0.001). Gastrectomy violated the Cox model proportional hazards assumption necessitating that the analysis be stratification by gastrectomy. The primary model (M1) incorporated region (Asia vs non-Asia, p=0.25, was included because it was significant in previous analysis (INTEGRATE), ECOG PS status, cancer extent, treatment with regorafenib, NLR, BMI, LDH, CA 19-9, and albumin. The final model (M2), which included PROs, demonstrated superior discriminative ability with the highest c-statistic values in both the gastrectomy and non-gastrectomy strata. The c-statistic increased from 0.695 for M1 to 0.723 for M2, indicating the added prognostic value of including PROs. Conclusions: The prognostic model that integrated both clinical data and PROs, proved robust for predicting survival in AGOC. The addition of PROs significantly enhanced prognostic accuracy, highlighting their importance in survival models. Further validation and refinement of the model is ongoing.