e21625 Background: CPIs including antibodies to programmed death-1 (PD-1) and its ligand anti PD-L1 are associated with superior overall survival (OS) in NSCLC. PDL1 is an imperfect predictive biomarker. We have examined real world data to identify additional factors associated with clinical benefit. Methods: 96 consecutive NSCLC patients (pts) were treated with CPIs at our institution. Multivariate cox proportional hazard regression models were used to examine the associations between OS and baseline characteristics including age, sex, histology, PDL1 status, performance status (PS), steroid use and brain metastases. Results: 93% (89/96) of pts received pembroluzimab, 6 (6%) nivolumab, and 1 (1%) atezoluzimab. All pts had metastatic disease (68% adenocarcinoma). 15% had PS 0, and 70% PS1. 11 of the 96 pts had brain metastases prior to CPI, 4 (37%) had whole brain radiotherapy, 2 (18%) targeted radiotherapy, 2 (18%) resection with radiotherapy and 3 (27%) had no definitive treatment for brain metastases. After multivariate analysis those with PS 1-2 had an increased risk of death (PS = 1 HR 2.74, 95% CI 1.11-6.72; PS = 2 3.61, 95% CI 1.28-10.15) compared with PS 0. Outcomes for pts receiving corticosteroids at the time of initiation of CPI were also inferior (HR 2.29, 95% CI 1.10-4.74) compared to those who were not. Conversely those who had brain metastasis diagnosed prior to commencing CPI had superior OS (HR 0.29 95% CI: 0.11-0.76). No statistically significant association was seen with age, gender, histology, stage at diagnosis, smoking or mutational status. PDL < 50% compared to > 50% was associated with an increased risk of death (HR 1.53 95% CI: 0.90-2.60) though it was not statistically significant. Conclusions: Here we show, using real world data, that PS and use at baseline of oral corticosteroids are associated with inferior OS with CPI treatment for NSCLC, in line with reports from large clinical trials. We show that baseline brain metastases are associated with improved outcomes. This may be explained by survivor bias, in that those who are successfully treated for brain disease, remain PS ≤2 and are able to commence CPI represent a highly selected group.