Purpose: We performed an integrated analysis of molecular classification systems proposed by The Cancer Genome Atlas (TCGA), the Asian Cancer Research Group (ACRG), and the tumor microenvironment (TME) score to identify which classification scheme(s) are most promising to pursue in subsequent translational investigations.
Experimental design: Supervised machine learning classifiers were created using 10-fold nested cross-validation for TCGA, ACRG, and TME subtypes and applied to 2,202 patients with gastric cancer from 11 separate publicly available datasets. Overall survival was assessed with a multivariable Cox proportional hazards model. A propensity score-matched analysis was performed to evaluate the subgroup effect of adjuvant chemotherapy on molecular subtypes. A public external cohort comprised of metastatic gastric cancer treated with immunotherapy was used to externally validate the molecular subtypes.
Results: Classification models for TCGA, ACRG, and TME achieved an accuracy ± SD of 89.5% ± 0.04, 84.7% ± 0.04, and 89.3% ± 0.02, respectively. We identified the TME score as the only significantly prognostic classification system [HR, 0.54 (95% confidence interval [CI], 0.39-0.74); global Wald test P < 0.001]. In our subgroup analysis, patients who received adjuvant chemotherapy achieved greater survival with increasing TME score (HR, 0.47; 95% CI, 0.29-0.74; interaction P < 0.05). The combination of TME-high and microsatellite instability scores significantly outperformed microsatellite instability as a univariable predictor of immunotherapy response.
Conclusions: We conclude that the TME score is a predominate driver of prognosis as well as chemotherapy- and immunotherapy-related outcomes in gastric cancer. This article provides a foundation for additional analyses and translational work.
©2024 American Association for Cancer Research.