OBJECTIVES: Recent clinical trials have established non-vitamin K antagonist oral anticoagulants (NOACs) as viable alternatives to warfarin in stroke prevention for atrial fibrillation. A major challenge in routine clinical practice is the high rate of medication discontinuation and switching. There are currently few large studies offering estimates of drug persistence with NOACs. Moreover, little is known about what drives patients’ choice and behavior regarding patterns of long-term medication use. Using a large heterogeneous cohort of patients, we aimed to investigate the patterns and predictors of switch and discontinuation among patients initiating NOACs. METHODS: We performed a retrospective cohort analysis using a large U.S. commercial insurance database, OptumLabs Data Warehouse. We identified 10,147 privately insured and Medicare Advantage patients with nonvalvular atrial fibrillation who initiated apixaban, dabigatran and rivaroxaban in 2013 and 2014, the time period when all three NOACs had become available in the U.S. Discontinuation of anticoagulation was defined as having at least 3 months of gap in days of supply of any oral anticoagulants. Two multivariable logistic regression models were performed to assess the risk factors related switch and discontinuation, respectively. RESULTS: Within one year of NOACs initiation, 13% of the patients switched to another oral anticoagulant, among whom the majority (64%) switched to warfarin. Nearly 40% of patients discontinued oral anticoagulants. Patients initiating apixaban were less likely to switch or discontinue compared to those initiating dabigatran or rivaroxaban. Other risk factors for switch include high out-of-pocket costs, prior warfarin use, and a stroke during follow up. Risk factors for discontinuation include high out-of-pocket costs, no prior warfarin exposure, and a major bleeding during follow up. CONCLUSIONS: Medication discontinuation and switching are common among patients treated with NOACs in routine clinical practice. Medication use may be affected by the choice of initial medication, out-of-pocket costs, and clinical events occurring during follow up.