OBJECTIVES: Recent trials have shown an overall survival (OS) benefit in 10-40% advanced cancer patients treated with programmed cell death 1 (PD-1) or programmed death-ligand 1 (PD-L1) inhibitors. Here, we aimed to evaluate the relationship between PD-L1 expression and the therapeutic efficacy of PD-1 or PD-L1 inhibitors in patients with cancer with recurrent or metastatic disease, compared with control treatments. METHODS: We systematically searched Medline (PubMed), Embase, and Cochrane Library databases up to Jan 2019 and pooled the treatment effects (hazard ratio or relative ratio) of PD-1/PD-L1 inhibitors in patients with different PD-L1 expression. RESULTS: Overall, twenty-four qualifying trials with over 14,860 subjects were eligible in this study. Compared with conventional agents, anti-PD/PD-L1 drugs significantly reduced the risk of death (hazard ratio 0.72, 95% CI 0.66 to 0.78), irrespective of the tumor type. Additionally, when PD-L1 expression ≥1% was defined as positive, anti-PD-1/PD-L1 monotherapy correlated with prolonged overall survival in patients with nonsmall cell lung cancer (NSCLC) (0.72, 0.61 to 0.86) and other cancer types (0.66, 0.57 to 0.76) patients with PD-L1 positive, rather than those with PD-L1 negative (hazard ratio for NSCLC and other cancer types: 0.84 and 0.87, respectively; all > 0.05). The subgroup analyses to experimental agents, PD-1/PD-L1 inhibitors, PD-L1 antibody clone, and type of IHC scoring method validated the robustness of these findings. However, anti-PD-1/PD-L1 combination therapies can reduce the risk of death for patients with different cancer types, regardless of PD-L1 expression ( < 0.05 for all PD-L1 expression status). CONCLUSIONS: We recommend PD-L1 expression as a predictive biomarker in patient selection for anti-PD-1/PD-L1 monotherapy, but not for combination therapies.