ABSTRACT: BACKGROUND:Atezolizumab is a humanised antiprogrammed death-ligand 1 (PD-L1) monoclonal antibody that inhibits PD-L1 and programmed death-1 (PD-1) and PD-L1 and B7-1 interactions, reinvigorating anticancer immunity. We assessed its efficacy and safety versus docetaxel in previously treated patients with non-small-cell lung cancer. METHODS:We did a randomised, open-label, phase 3 trial (OAK) in 194 academic or community oncology centres in 31 countries. We enrolled patients who had squamous or non-squamous non-small-cell lung cancer, were 18 years or older, had measurable disease per Response Evaluation Criteria in Solid Tumors, and had an Eastern Cooperative Oncology Group performance status of 0 or 1. Patients had received one to two previous cytotoxic chemotherapy regimens (one or more platinum based combination therapies) for stage IIIB or IV non-small-cell lung cancer. Patients with a history of autoimmune disease and those who had received previous treatments with docetaxel, CD137 agonists, anti-CTLA4, or therapies targeting the PD-L1 and PD-1 pathway were excluded. Patients were randomly assigned (1:1) to intravenously receive either atezolizumab 1200 mg or docetaxel 75 mg/m2 every 3 weeks by permuted block randomisation (block size of eight) via an interactive voice or web response system. Coprimary endpoints were overall survival in the intention-to-treat (ITT) and PD-L1-expression population TC1/2/3 or IC1/2/3 (?1% PD-L1 on tumour cells or tumour-infiltrating immune cells). The primary efficacy analysis was done in the first 850 of 1225 enrolled patients. This study is registered with ClinicalTrials.gov, number NCT02008227. FINDINGS:Between March 11, 2014, and April 29, 2015, 1225 patients were recruited. In the primary population, 425 patients were randomly assigned to receive atezolizumab and 425 patients were assigned to receive docetaxel. Overall survival was significantly longer with atezolizumab in the ITT and PD-L1-expression populations. In the ITT population, overall survival was improved with atezolizumab compared with docetaxel (median overall survival was 13·8 months [95% CI 11·8-15·7] vs 9·6 months [8·6-11·2]; hazard ratio [HR] 0·73 [95% CI 0·62-0·87], p=0·0003). Overall survival in the TC1/2/3 or IC1/2/3 population was improved with atezolizumab (n=241) compared with docetaxel (n=222; median overall survival was 15·7 months [95% CI 12·6-18·0] with atezolizumab vs 10·3 months [8·8-12·0] with docetaxel; HR 0·74 [95% CI 0·58-0·93]; p=0·0102). Patients in the PD-L1 low or undetectable subgroup (TC0 and IC0) also had improved survival with atezolizumab (median overall survival 12·6 months vs 8·9 months; HR 0·75 [95% CI 0·59-0·96]). Overall survival improvement was similar in patients with squamous (HR 0·73 [95% CI 0·54-0·98]; n=112 in the atezolizumab group and n=110 in the docetaxel group) or non-squamous (0·73 [0·60-0·89]; n=313 and n=315) histology. Fewer patients had treatment-related grade 3 or 4 adverse events with atezolizumab (90 [15%] of 609 patients) versus docetaxel (247 [43%] of 578 patients). One treatment-related death from a respiratory tract infection was reported in the docetaxel group. INTERPRETATION:To our knowledge, OAK is the first randomised phase 3 study to report results of a PD-L1-targeted therapy, with atezolizumab treatment resulting in a clinically relevant improvement of overall survival versus docetaxel in previously treated non-small-cell lung cancer, regardless of PD-L1 expression or histology, with a favourable safety profile. FUNDING:F. Hoffmann-La Roche Ltd, Genentech, Inc.