ObjectivesTo compare clinical outcomes with programmed‐death ligand‐1 immune checkpoint inhibitors (ICIs) in patients with advanced urothelial carcinoma (aUC) who have vs have not undergone radical surgery (RS) or radiation therapy (RT) prior to developing metastatic disease.Patients and MethodsWe performed a retrospective cohort study collecting clinicopathological, treatment and outcomes data for patients with aUC receiving ICIs across 25 institutions. We compared outcomes (observed response rate [ORR], progression‐free survival [PFS], overall survival [OS]) between patients with vs without prior RS, and by type of prior locoregional treatment (RS vs RT vs no locoregional treatment). Patients with de novo advanced disease were excluded. Analysis was stratified by treatment line (first‐line and second‐line or greater [second‐plus line]). Logistic regression was used to compare ORR, while Kaplan–Meier analysis and Cox regression were used for PFS and OS. Multivariable models were adjusted for known prognostic factors.ResultsWe included 562 patients (first‐line: 342 and second‐plus line: 220). There was no difference in outcomes based on prior locoregional treatment among those treated with first‐line ICIs. In the second‐plus‐line setting, prior RS was associated with higher ORR (adjusted odds ratio 2.61, 95% confidence interval [CI]1.19–5.74]), longer OS (adjusted hazard ratio [aHR] 0.61, 95% CI 0.42–0.88) and PFS (aHR 0.63, 95% CI 0.45–0.89) vs no prior RS. This association remained significant when type of prior locoregional treatment (RS and RT) was modelled separately.ConclusionPrior RS before developing advanced disease was associated with better outcomes in patients with aUC treated with ICIs in the second‐plus‐line but not in the first‐line setting. While further validation is needed, our findings could have implications for prognostic estimates in clinical discussions and benchmarking for clinical trials. Limitations include the study’s retrospective nature, lack of randomization, and possible selection and confounding biases.