BACKGROUND: Immune checkpoint inhibitors are a new standard of care for patients with advanced non-small-cell lung cancer (NSCLC) without EGFR tyrosine kinase or anaplastic lymphoma kinase (ALK) genetic aberrations (EGFR-/ALK-), but clinical benefit in patients with EGFR mutations or ALK rearrangements (EGFR+/ALK+) has not been shown. We assessed the effect of durvalumab (anti-PD-L1) treatment in three cohorts of patients with NSCLC defined by EGFR/ALK status and tumour expression of PD-L1. METHODS: ATLANTIC is a phase 2, open-label, single-arm trial at 139 study centres in Asia, Europe, and North America. Eligible patients had advanced NSCLC with disease progression following at least two previous systemic regimens, including platinum-based chemotherapy (and tyrosine kinase inhibitor therapy if indicated); were aged 18 years or older; had a WHO performance status score of 0 or 1; and measurable disease per Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. Key exclusion criteria included mixed small-cell lung cancer and NSCLC histology; previous exposure to any anti-PD-1 or anti-PD-L1 antibody; and any previous grade 3 or worse immune-related adverse event while receiving any immunotherapy agent. Patients in cohort 1 had EGFR+/ALK+ NSCLC with at least 25%, or less than 25%, of tumour cells with PD-L1 expression. Patients in cohorts 2 and 3 had EGFR-/ALK- NSCLC; cohort 2 included patients with at least 25%, or less than 25%, of tumour cells with PD-L1 expression, and cohort 3 included patients with at least 90% of tumour cells with PD-L1 expression. Patients received durvalumab (10 mg/kg) every 2 weeks, via intravenous infusion, for up to 12 months. Retreatment was allowed for patients who benefited but then progressed after completing 12 months. The primary endpoint was the proportion of patients with increased tumour expression of PD-L1 (defined as >/=25% of tumour cells in cohorts 1 and 2, and >/=90% of tumour cells in cohort 3) who achieved an objective response, assessed in patients who were evaluable for response per independent central review according to RECIST version 1.1. Safety was assessed in all patients who received at least one dose of durvalumab and for whom any post-dose data were available. The trial is ongoing, but is no longer open to accrual, and is registered with ClinicalTrials.gov, number NCT02087423. FINDINGS: Between Feb 25, 2014, and Dec 28, 2015, 444 patients were enrolled and received durvalumab: 111 in cohort 1, 265 in cohort 2, and 68 in cohort 3. Among patients with at least 25% of tumour cells expressing PD-L1 who were evaluable for objective response per independent central review, an objective response was achieved in 9 (12.2%, 95% CI 5.7-21.8) of 74 patients in cohort 1 and 24 (16.4%, 10.8-23.5) of 146 patients in cohort 2. In cohort 3, 21 (30.9%, 20.2-43.3) of 68 patients achieved an objective response. Grade 3 or 4 treatment-related adverse events occurred in 40 (9%) of 444 patients overall: six (5%) of 111 patients in cohort 1, 22 (8%) of 265 in cohort 2, and 12 (18%) of 68 in cohort 3. The most common treatment-related grade 3 or 4 adverse events were pneumonitis (four patients [1%]), elevated gamma-glutamyltransferase (four [1%]), diarrhoea (three [1%]), infusion-related reaction (three [1%]), elevated aspartate aminotransferase (two [<1%]), elevated transaminases (two [<1%]), vomiting (two [<1%]), and fatigue (two [<1%]). Treatment-related serious adverse events occurred in 27 (6%) of 444 patients overall: five (5%) of 111 patients in cohort 1, 14 (5%) of 265 in cohort 2, and eight (12%) of 68 in cohort 3. The most common serious adverse events overall were pneumonitis (five patients [1%]), fatigue (three [1%]), and infusion-related reaction (three [1%]). Immune-mediated events were manageable with standard treatment guidelines. INTERPRETATION: In patients with advanced and heavily pretreated NSCLC, the clinical activity and safety profile of durvalumab was consistent with that of other anti-PD-1 and anti-PD-L1 agents. Responses were recorded in all cohorts; the proportion of patients with EGFR-/ALK- NSCLC (cohorts 2 and 3) achieving a response was higher than the proportion with EGFR+/ALK+ NSCLC (cohort 1) achieving a response. The clinical activity of durvalumab in patients with EGFR+ NSCLC with >/=25% of tumour cells expressing PD-L1 was encouraging, and further investigation of durvalumab in patients with EGFR+/ALK+ NSCLC is warranted. FUNDING: AstraZeneca.