Immune checkpoint inhibitors (ICIs) initiate a new stage for gastric cancer (GC) therapeutics, and plenty of patients have already benefited from ICIs. Liquid biopsy promotes the development of precision medicine of GC. However, due to the lack of precision biomarkers of immune-related adverse events (irAEs), the safety of ICIs-treated GC patients cannot be guaranteed. In our study, GC patients treated with ICIs were included for investigating the correlation between irAEs of ICIs and corresponding outcomes. We also explored the potential of biomarkers of irAEs via EV-derived proteins. Dynamic plasma was taken from 102 ICIs-treated GC patients generated retrospectively or prospectively, who were divided into discovery and validating cohorts. Plasma EV-derived protein profiles were described, and two EV-proteins, inducible T-cell co-stimulator (EV-ICOS) and indoleamine 2,3-dioxygenase 1(EV-IDO1), from 42 vital proteins were screened to predict the prognosis of ICIs with irAEs. Our work is the first to propose that EV-proteins can predict ICIs-corresponding irAEs, which can be conducive to the diagnosis and treatment of GC patients, and to facilitate the screening of beneficiaries.