Background: While several clinical scoring systems exist to aid prognostication and patient (pt) selection for clinical trials in oncology, none are standardly used. We compared the ability of four prognostic scores to predict overall survival (OS) in pts with advanced gastric and esophageal (GE) cancer. Methods: Pts with advanced (unresectable or metastatic) GE cancer receiving first-line palliative-intent systemic therapy at the Princess Margaret Cancer Centre from 2007 to 2020 were included. High prognostic risk pts were identified using four scoring systems: Royal Marsden Hospital (RMH), MD Anderson Cancer Centre (MDACC), Gustave Roussy Immune Score (GRIm-S) and MD Anderson Immune Checkpoint Inhibitor (MDA-ICI) score. OS was estimated using the Kaplan-Meier method and compared between risk groups (high vs. not-high) for each scoring system using the log-rank test. Cox proportional hazards models were used to analyze the association between each prognostic score and OS, adjusting for baseline clinical factors. Harrell’s c-index was used to evaluate predictive discrimination of the models. Time-dependent AUCs were used to measure predictive ability for early death (within 90 days). Results: In total, 451 pts with advanced GE cancer were included. The median age was 59 years, 68% were male, 51% had ECOG status 0-1, 63% presented with de novo metastatic disease. The proportion of pts categorized as high risk was: RMH 25% (N=113), MDACC 13% (N=95), GRIm-S 24% (N=109), MDA-ICI 26% (N=117). In all scoring systems, high risk pts had significantly shorter OS (median OS 7.9 versus 12.2 months for RMH high vs. low risk, p<0.001; 6.8 vs. 11.9 months p<0.001 for MDACC; 5.3 vs. 13 months p<0.001 for GRIm-S; 8.2 vs. 12.2 months p<0.001 for MDA-ICI). On multivariable analysis, each prognostic score was significantly associated with OS (Table). The GRIm-S had the highest predictive discrimination (c-index 0.645 [0.612-0.678]) and highest predictive ability for early death (AUC 0.754 [0.675-0.832]). Conclusions: All four prognostic scoring systems compared had reasonable accuracy in predicting OS for patients with advanced GE cancer. The higher accuracy for predicting early death may render the GRIm-S as preferable. These tools can aid oncologists in discussions about prognosis, therapeutic decision-making and patient selection for clinical trials.

¹Each multivariable model adjusted for baseline characteristics: age, sex, histology, ECOG, and number of metastatic sites.