A related commentary discusses this research. In the early stages of treatment, paclitaxel appears to be the clear winner for patients with a PD-L1 combined positive score (CPS) of 1 or higher; median progression-free survival was 4·1 months with paclitaxel versus 1·4 months with pembrolizumab (HR 1·27, 95% CI 1·03–1·57). These results highlight that for patients with second-line gastro-oesophageal cancer, disease control is important because rapid progression could rule out further therapy. Notably, the survival curves of pembrolizumab and paclitaxel for patients with a PD-L1 CPS of 10 or higher do not cross; therefore, the detrimental effect of immune checkpoint blockade on early progression appears mitigated. Future trials of single-drug pembrolizumab should focus on this patient group, populations enriched for tumours with high mutation burden, microsatellite instability, or biomarkers that emerge through translational research on the trial dataset.