Background: ICI with anti-PD-1 therapy was approved by the US FDA as 1st/2nd-line treatment of advanced GEC regardless of PD-L1 status. It is argued that patients (pts) with PD-L1-low GEC do not benefit from ICI. To avoid spurious exclusion of pts, we systematically reviewed phase 3 trials to assess the magnitude and consistency of PD-L1 to predict benefit from ICI. We also examined other potential predictive factors to understand their magnitude and consistency in comparison with PD-L1 status. Methods: MEDLINE, Embase, Scopus, Web of Science, and Cochrane Central Register were searched for all randomized clinical trials (RCT; 2000-2021) in advanced GEC comparing ICI, alone or with chemotherapy, vs standard of care (SOC). The primary endpoint was overall survival (OS). Study screening, data abstraction, and bias assessment were done independently by two reviewers per PRISMA guidelines. The mean HR for ICI vs SOC was calculated by random effects model. The predictive value of a factor was quantified using relative magnitude and consistency (table). Results: 14 RCTs at low risk of bias were included (N = 9,210 pts; KN62, CM649, CM648, ESCORT1st, KN590, ATT4, JAV100, KN181, KN61, ATT3, ESCORT, RATIONALE302, JAV300, ATT2). 6 trials were 1^st-line, 8 were after 1^st-line. Treatment comparisons were ICI + chemo vs chemo (6 trials), ICI vs chemo (9 trials), and ICI vs placebo (1 trial). ICI was always anti-PD-1/-L1 (1 trial also had anti-CTLA-4). Table shows primary results. To enable head to head comparisons, subgroup analyses of trials containing both levels of a factor or having similar design were done and generally yielded consistent results. Conclusions: PD-L1 CPS is the strongest predictive biomarker, after MSI, for OS benefit from ICI-containing therapy vs SOC. Additional research on gender is warranted.

^a Comparison means ICI-containing vs SOC. Some trials had >2 arms, thus >1 comparison.^b 1 – (ratio of mean HRs of one factor level vs another). Factors with <15% predictive value not shown (ie, tumor site, geography, liver met, # mets, prior surgery, age, ECOG PS, Lauren, chemo backbone).^c Highest consistent cutpoint reported across studies (usually CPS >10).Abbrev: CPS, combined positive score; TPS, tumor proportion score