Background: To determine the frequency of discordance in programmed death-ligand 1(PD-L1) expression between primary tumors and paired distant metastases in advanced cancers.Methods: We searched MEDLINE and EMBASE for eligible studies and assessed their methodologic quality using QUADAS-2 tool. We estimated the discordant rates (positive to negative or vice versa) of PD-L1 expression in primary tumors and paired distant metastases using logistic-normal random effects model. We performed subgroup analyses based on the PD-L1 status of primary tumors (positive or negative), location of primary tumors (lung or others) and distant metastases (central nervous system or others), timing of distant metastases (synchronous or metachronous), positivity thresholds of PD-L1 expression (1% or 5%) and types of antibody clones used (E1L3N or SP142).Results: Thirteen eligible studies including 451 cases were identified. The included studies were judged to have low to unclear risk of bias. The pooled estimate of discordant rates in PD-L1 expression was 31% (95% CI= 19-47%), with high heterogeneity across the studies (I2 = 75%). There was no significant effect modification in the discordant rates according to the predefined subgroups.Conclusion: Approximately one-third of advanced cancer cases have discordance in PD-L1 expression between primary tumors and paired distant metastases. A more liberal testing of PD-L1 expression in both primary and metastatic tumors is recommended in order to identify patients who may benefit from immune checkpoint blockade treatment. Further research exploring the mechanisms and its impact are warranted.