BACKGROUND: Prognostic scores that can identify patients at risk for early death are needed to aid treatment decision-making and patient selection for clinical trials. We compared the accuracy of four scores to predict early death (within 90 days) and overall survival (OS) in patients with metastatic gastric and esophageal (GE) cancer. METHODS: Advanced GE cancer patients receiving first-line systemic therapy were included. Prognostic risks were calculated using: Royal Marsden Hospital (RMH), MD Anderson Cancer Centre (MDACC), Gustave Roussy Immune (GRIm-Score), and MD Anderson Immune Checkpoint Inhibitor (MDA-ICI) scores. Overall survival (OS) was estimated using the Kaplan-Meier method. Cox proportional hazards models were used to analyze associations between prognostic scores and OS. The predictive discrimination was estimated using Harrell's c-index. Predictive ability for early death was measured using time-dependent AUCs. RESULTS: In total, 451 patients with metastatic GE cancer were included. High risk patients had shorter OS for all scores (RMH high- vs. low-risk median OS 7.9 vs. 12.2 months, P < .001; MDACC 6.8 vs. 11.9 months P < .001; GRIm-Score 5.3 vs. 13 months, P < .001; MDA-ICI 8.2 vs. 12.2 months, P < .001). On multivariable analysis, each prognostic score was significantly associated with OS. The GRIm-Score had the highest predictive discrimination and predictive ability for early death. CONCLUSIONS: The GRIm-Score had the highest accuracy in predicting early death and OS. Clinicians may use this score to identify patients at higher risk of early death to guide treatment decisions including clinical trial enrolment. This score could also be used as a stratification factor in future clinical trial designs.