We included 51,690 new users of OAC in this population-based cohort study, 41,146 of them had dispensing data available in SIDIAP (79.6%). Approximately 20% of them patients initiated DOAC, pointing out that VKA are still the first therapeutic option for anticoagulation in NVAF in our setting, as recommended by AEMPS. Most patients (83.9%) had CHA2DS2VASc score ≥ 2, which is the criterion to anticoagulate in NVAF patients according to guidelines [32, 33]. Patients with highest risks of stroke were those in the groups of acenocoumarol and apixaban, as shown in previous studies [14, 22, 34].
Therapeutic adherences at implementation and persistence to OAC were assessed in those patients who were adherent at initiation and started anticoagulation treatment before 2015. Only one third of naive patients received DOAC treatment during at least one year of follow-up, for VKA the proportion was higher. Between them, rivaroxaban group showed the highest percentage of patients with good adherence at implementation (MPR ≥ 80) and dabigatran the lowest. Similar results were also found by Forslund et al.  or by Beyer-Westendorf et al.  although this last study only analysed rivaroxaban and dabigatran. On the opposite, other studies found higher MPR in apixaban-treated patients [14, 35].
Regarding persistence to DOAC in naive patients, apixaban showed higher discontinuation rates during the first month of treatment but at one year, all DOAC showed similar rates. Several studies analysed discontinuation rates at different times during follow-up. After one year, apixaban users were more persistent than other DOAC and VKA users in the studies conducted by Forslund et al.  and Johnson et al. 
Other studies which analysed persistence in naive patients only included dabigatran and rivaroxaban. Rivaroxaban presented better persistence than dabigatran and VKA [23, 36].
Manzoor et al.  or Martínez et al.  studied all DOAC together as a group. The first one showed higher levels of persistence to DOAC in anticoagulant-experienced patients over time. Martínez et al. found higher levels of persistence in DOAC versus VKA users, which slightly decreased during the first year of follow-up (from 94.7% of persistents to DOAC at 3 months to 72.9% at 12 months).
For anticoagulant-experienced patients with more than one year of follow-up in our study, again rivaroxaban showed the largest proportion of patients with good adherence during implementation. Only Manzoor et al.  analysed MPR in non-naive patients and apixaban showed higher MPR in apixaban at 6 months and dabigatran at 9 months. Discontinuation rates in our non-naive patients were much lower than for the naive ones; during the first month since treatment initiation 1.6% apixaban, 4.9% rivaroxaban and 12.6% dabigatran patients discontinued the treatment, and after one year apixaban users showed higher persistence rates than dabigatran and rivaroxaban (66.1% > 64.6% > 48.5%). Manzoor et al.  compared persistence between naive and non-naive patients receiving DOAC and the last ones showed higher levels of persistence. Johnson et al.  described similar discontinuation rates than for naive patients and at the end-of follow-up patients prescribed apixaban showed improved persistence over dabigatran, rivaroxaban and VKA.
The differences in treatment persistence between naive and anticoagulant-experienced patients in our study could be motivated by a better knowledge of the anticoagulation importance of these patients who previously received mainly VKA, and they were used to attend monthly to PHC centres for INR determination and had optimal levels of drug adherence.
Suboptimal adherence to anticoagulant therapy places patients with AF at risk for stroke or bleeding complications. Our study concludes as most observational studies, that the guidelines recommendations regarding anticoagulant therapy are not routinely followed in clinical practice, and adherence is substantially lower than in clinical trials [3, 38].
Some specific limitations in our database are the lack of association between GP’s prescriptions and dispensing associated with these prescriptions. This study has missing data from pharmacy claims and for some variables as it is common in observational studies using electronic databases (information bias). The strengths of our study are representativeness for the general population, with a database that covers almost the 80% of the Catalonian population, with complete sociodemographic and health records, long follow-up, and real clinical practice data.