Background <p>Pembrolizumab is an important immune checkpoint inhibitor for advanced urothelial carcinoma (aUC). However, there are no established biomarkers to predict its efficacy. Extracellular vesicles (EVs) are liquid bilayer structures released from various cells that contain genetic information such as DNAs, RNAs, and proteins. EVs-derived long non-coding RNAs (lncRNAs) have been reported to be involved play in tumor progression and drug resistance. We investigated whether EVs-derived lncRNA MAFG-AS1 is effective in predicting pembrolizumab efficacy.</p> Methods <p>The expression of lncRNA MAFG-AS1 was examined in UC tissues from 22 patients who underwent total cystectomy at our institution, and normal urothelial tissues. Functional analysis of lncRNA MAFG-AS1 were performed using bladder cancer cell lines. The relationship between EVs-derived lncRNA MAFG-AS1 expression and clinical response and prognosis was examined in 52 patients treated with pembrolizumab.</p> Results <p>The expression of lncRNA MAFG-AS1 was notably higher in UC tissues than in normal urothelial tissues. LncRNA MAFG-AS1 knockdown in bladder cancer cells by siRNA, significantly decreased cell viability, invasion, and migration. Patients with a clinical response showed significantly a lower EVs-derived lncRNA MAFG-AS1 expression than those with no clinical response. The high EVs-derived lncRNA MAFG-AS1 expression group had significantly worse progression-free and overall survival than the low expression group (Log-rank <i>P</i> = 0.0096 and log-rank <i>P</i> = 0.0349, respectively). In multivariate analysis, high EVs-derived lncRNA MAFG-AS1 expression was a significant risk factor for poor clinical response to pembrolizumab (HR = 3.2, <i>P</i> = 0.0010).</p> Conclusions <p>The EVs-derived lncRNA MAFG-AS1 may be an effective less-invasive biomarker for predicting the efficacy of pembrolizumab in aUC.</p>